Skip to content

The ultimate harness for coding agents and beyond. All your agents. All your machines. One command center. Start with code, then follow your curiosity and build across disciplines: CAD, circuits, robots, games and music.

Agents start harnesses only when it pays, and Harness cleans up after them (#1234)

Dee1ff0bebac290
diff --git a/cli/AGENTS.md b/cli/AGENTS.md
index dac907d27..fae72aa46 100644
--- a/cli/AGENTS.md
+++ b/cli/AGENTS.md
@@ -28,7 +28,7 @@ they show as `harnessd`, `harnessd-core` and `harnessd-<process>` (`src/harnessd
| engine-claude, engine-codex | history, last-turn readers, live transcript parsers, runtime profiles/catalogs, screen interpretation, submission verification readings, model control, question navigation, and (Codex) the shared app-server connection | on demand: an attach or read for that engine |
| search | session search | always |
| viewers | the harness viewers, their remote streams and rendered surfaces, and the Store | always |
-| edge | the shell service, workspaces, usage, the monitor, the project readers, the change-agent handoff, the recaps | always |
+| edge | the shell service, workspaces, usage, the monitor, the project readers, the change-agent handoff, the recaps, the agent cleanup | always |
| gateway | the relay and its E2EE (`src/gateway/`) | on demand: as the core starts when signed in or anything is paired here, or when something needs it (a pairing, a window's E2EE request, a key command, the Wi-Fi device) |
| models | grid, local models, the Model Manager | on demand: once grid is in use here (a managed grid, saved grid pictures, local models), or on a models request |
| updater | checks, downloads and stages a new build (`src/services/updaterProcess.ts`) | by the master, for the installed copy only |
diff --git a/cli/e2e/agentClient.e2e.ts b/cli/e2e/agentClient.e2e.ts
index bdfadb3c1..4f5470970 100644
--- a/cli/e2e/agentClient.e2e.ts
+++ b/cli/e2e/agentClient.e2e.ts
@@ -19,7 +19,7 @@ import { FakeBackend, type FakeMachine } from './harness/fakeBackend.js'
/** The line (lib/agentAttribution.ts), as a reader of the pull request sees it. */
const AGENT_ATTRIBUTION_LINE = '🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.'
-const HARNESS_INSTRUCTIONS = 'You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.'
+const HARNESS_INSTRUCTIONS = 'You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.'
const ME: FakeMachine = { machineId: 'b8'.repeat(16), computerId: 'e2e-computer-0000-0000-0000000000b8', name: 'studio', token: 'e2e-token-agent-client' }
const signIn = (d: IsolatedDaemon) => writeFileSync(join(d.root, 'auth', 'session.json'), JSON.stringify({
version: 1, accessToken: ME.token, autonomousEnv: 'prod', computerId: ME.computerId,
@@ -177,6 +177,90 @@ describe('Harness for agents, end to end', () => {
client.close()
})
+ it('a harness another agent started may not start more, is closed with its starter, and is the person\'s once they type into it', async () => {
+ // Owner, October 12: "we need to be very careful about when to start a new agent so we don't waste user token and we
+ // need to properly clean up after too". The cleanup runs in its own process here, as it does by default (the edge
+ // host): what the core tells it and what it asks back cross the link.
+ const relay = await FakeBackend.start()
+ backend = relay
+ relay.addMachine(ME)
+ relay.agentMessages = { revision: 1, machines: [{ machineId: ME.machineId, name: 'studio' }], tabs: [] }
+ const d = await IsolatedDaemon.create({ env: {
+ BACKEND_WS_URL: relay.wsUrl, WEB_URL: relay.httpUrl, ADAPTER_COMPUTER_ID: ME.computerId,
+ HARNESSD_SERVICE_INITIAL_BACKOFF_MS: '200', HARNESSD_SERVICE_MAX_BACKOFF_MS: '1000',
+ } })
+ daemon = d
+ onTestFailed(() => { console.log(`---- daemon log\n${d.log().split('\n').slice(-200).join('\n')}`) })
+ signIn(d)
+ await d.start()
+ const client = await LocalClient.connect(d, { machineId: ME.machineId })
+ const list = async (includeStopped = false) => (await client.request<{ agents: Array<Record<string, any>> }>('agents_list', { includeStopped }, 30_000)).agents
+ const lead = async (name: string) => {
+ const dir = join(d.projectsDir, name)
+ mkdirSync(dir, { recursive: true })
+ const made = await client.request('agent_create', { engine: 'claude', cwd: dir, bypassPermission: true, name }, 90_000)
+ expect(made.error, JSON.stringify(made)).toBeUndefined()
+ const bound = await until(`${name} to bind its conversation`, async () => {
+ const row = (await list()).find((a) => a.id === made.agent.id)
+ return row?.sessionId && row.status === 'active' ? row : null
+ }, 60_000, 500)
+ // One turn of the person's first: a close saves a conversation, and one with nothing in it yet has none to save.
+ const from = client.frames.length
+ client.send('message', { agentId: bound.id, content: `hello ${name}` })
+ const opened = await client.waitFor((f) => f.type === 'turn_started' && f.agentId === bound.id, 60_000, `${name}'s first turn`, from)
+ await client.waitFor((f) => f.type === 'turn_ended' && f.agentId === bound.id, 60_000, `${name}'s first turn to end`, client.frames.indexOf(opened) + 1)
+ return bound
+ }
+ const start = async (from: Record<string, any>, name: string) => {
+ const started = await harness(d, String(from.tmuxPane), ['new', 'claude', name, '--name', name, '--', `hello ${name}`])
+ expect(started.code, started.out + started.err).toBe(0)
+ return until(`${name} to run`, async () => (await list()).find((a) => a.name === name && a.status === 'active' && a.tmuxPane) ?? null, 60_000, 500)
+ }
+ const stateOf = async (agent: Record<string, any>) => (await list(true)).find((a) => a.id === agent.id)?.status
+
+ const first = await lead('lead')
+ const kid = await start(first, 'kid')
+
+ // One level down: the harness lead started may not start one itself, and nothing is started.
+ const deeper = await harness(d, String(kid.tmuxPane), ['new', 'claude', 'deeper', '--name', 'deeper', '--', 'hello deeper'])
+ expect(deeper.code, deeper.out).toBe(1)
+ expect(deeper.err.trim()).toBe('A harness another agent started may not start harnesses itself. Do the work yourself, or say in your answer what else is needed.')
+ expect((await list(true)).some((a) => a.name === 'deeper')).toBe(false)
+
+ // Lead reads its answer, and is told to close it; `ps` counts what it has open.
+ const waited = await harness(d, String(first.tmuxPane), ['wait', 'kid', '--timeout', '2m'])
+ expect(waited.code, waited.out + waited.err).toBe(0)
+ const shown = await harness(d, String(first.tmuxPane), ['show', 'kid'])
+ expect(shown.out.trim().split('\n').at(-1)).toBe('Close it once you have what you need: harness close kid')
+ const ps = await harness(d, String(first.tmuxPane), ['ps'])
+ expect(ps.out.trim().split('\n').at(-1)).toBe('You have 1 harness of your own open, 1 idle. Close each once you have its answer: harness close <name>.')
+
+ // A second lead with two: the person types into one of them, from a window.
+ const second = await lead('lead2')
+ const joined = await start(second, 'joined')
+ const left = await start(second, 'left')
+ client.send('message', { agentId: joined.id, content: 'hello from the person' })
+
+ // The app closes lead, as a pane's close does (`agent_close`): kid closes with it.
+ const closed = await client.request('agent_close', { agentId: first.id, sessionId: first.sessionId, createdAt: first.createdAt, mode: 'now' }, 60_000)
+ expect(closed, JSON.stringify(closed)).toMatchObject({ closed: true })
+ await until('kid to close with lead', async () => (await stateOf(kid)) === 'stopped' || null, 60_000, 500)
+ // A window closing the tab both are in closes kid's pane next, a moment later: closed, not "it changed".
+ const again = await client.request('agent_close', { agentId: kid.id, sessionId: kid.sessionId, createdAt: kid.createdAt, mode: 'idle' }, 60_000)
+ expect(again, JSON.stringify(again)).toMatchObject({ closed: true })
+
+ // The person closes lead2 from a terminal: the one they typed into stays, the other closes.
+ const byPerson = await harness(d, '', ['close', 'lead2', '--now'])
+ expect(byPerson.code, byPerson.out + byPerson.err).toBe(0)
+ await until('left to close with lead2', async () => (await stateOf(left)) === 'stopped' || null, 60_000, 500)
+ expect(await stateOf(joined)).toBe('active')
+ // One line for each close.
+ expect(d.log()).toMatch(/\[agent-cleanup\] closed kid \(\w+\): the harness that started it was closed\. Its conversation is kept\./)
+ expect(d.log()).toMatch(/\[agent-cleanup\] closed left \(\w+\): the harness that started it was closed\./)
+ expect(d.log()).not.toMatch(/\[agent-cleanup\] closed joined/)
+ client.close()
+ })
+
it('signed out, the person starts, waits for, reads and closes a harness from a terminal', async () => {
// Never signed in: the daemon runs as this computer (core/main.ts), and the app made harnesses while `harness new`
// refused with "This computer is not signed in to Harness".
diff --git a/cli/src/agentClient/commands.spec.ts b/cli/src/agentClient/commands.spec.ts
index b8e0231b6..2db5d76eb 100644
--- a/cli/src/agentClient/commands.spec.ts
+++ b/cli/src/agentClient/commands.spec.ts
@@ -2,7 +2,7 @@ import { mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs'
import { tmpdir } from 'node:os'
import { join } from 'node:path'
import { afterEach, beforeEach, describe, expect, it } from 'vitest'
-import { addressOf, agentOrigin, AGENTS_START_AGENTS_OFF, closeCommand, EXIT, parseDuration, psCommand, showCommand, spawnRefusal, SPAWN_LIMITS, waitCommand, type ClientDeps } from './commands.js'
+import { addressOf, agentOrigin, AGENTS_START_AGENTS_OFF, closeCommand, DEPTH_REFUSAL, EXIT, parseDuration, psCommand, showCommand, spawnRefusal, SPAWN_LIMITS, waitCommand, type ClientDeps } from './commands.js'
import { AgentClientError, type Daemon, type Where } from './daemon.js'
import { Ledger, ledgerPath } from './ledger.js'
import { writeAgentSettings } from '../lib/agentSettings.js'
@@ -99,6 +99,28 @@ describe('harness ps', () => {
expect(d.closed).toBe(true)
})
+ it('tells an agent how many of its own are open and idle, and to close them once answered', async () => {
+ const d = new FakeDaemon()
+ d.frames = [frame('lead', { engine: 'claude' }), frame('a'), frame('b', { activity: { state: 'working' } }), frame('c', { status: 'stopped' }), frame('theirs')]
+ const h = harness(d)
+ for (const kid of ['a', 'b', 'c']) started(h.ledger, 'lead', kid)
+ expect(await psCommand([], h.deps)).toBe(EXIT.ok)
+ expect(h.out.join('\n').split('\n').at(-1)).toBe('You have 2 harnesses of your own open, 1 idle. Close each once you have its answer: harness close <name>.')
+ // One, and none idle.
+ d.frames = [frame('lead', { engine: 'claude' }), frame('b', { activity: { state: 'working' } })]
+ const one = harness(d, { env: { HARNESS_VIA_MCP: '1' } })
+ expect(await psCommand([], one.deps)).toBe(EXIT.ok)
+ expect(one.out.join('\n').split('\n').at(-1)).toBe('You have 1 harness of your own open. Close each once you have its answer: harness_close (name: "<name>").')
+ // None of its own, and the person: nothing to say.
+ d.frames = [frame('lead', { engine: 'claude' }), frame('theirs')]
+ const none = harness(d)
+ expect(await psCommand([], none.deps)).toBe(EXIT.ok)
+ expect(none.out.join('\n')).not.toMatch(/of your own open/)
+ const person = harness(d, { pane: undefined })
+ expect(await psCommand([], person.deps)).toBe(EXIT.ok)
+ expect(person.out.join('\n')).not.toMatch(/of your own open/)
+ })
+
it('tells an agent outside Harness that it is one', async () => {
const d = new FakeDaemon()
d.frames = [frame('kid')]
@@ -179,12 +201,38 @@ describe('harness show', () => {
'asked: review 3f2a',
'answer (after 1 tool calls):',
'Two bugs.',
+ // Idle and the caller's: a harness left open after its answer is read sits there for nobody.
+ 'Close it once you have what you need: harness close reviewer',
])
expect(d.requests.filter((r) => r.type === 'session_get').map((r) => r.payload)).toEqual([
{ sessionId: 's-reviewer', limit: 80 }, { sessionId: 's-reviewer', limit: 80, before: '1' },
])
})
+ it('ends with a hint to close only an idle harness of the caller\'s, by a name its verbs take, as a tool through MCP', async () => {
+ const d = new FakeDaemon()
+ d.frames = [frame('lead'), frame('kid', { name: 'so we have built harness web,' }), frame('busy'), frame('theirs')]
+ d.screens.set('busy', ['working'])
+ d.pages.set('s-kid', [{ events: [ev('user_message', { content: 'go' }), ev('text_delta', { content: 'done' })], hasMore: false }])
+ const h = harness(d)
+ started(h.ledger, 'lead', 'kid')
+ started(h.ledger, 'lead', 'busy')
+ expect(await showCommand(['kid'], h.deps)).toBe(EXIT.ok)
+ // An unnamed harness is called by its session title, a sentence `close` would not take: the start of its id.
+ expect(h.out.at(-1)).toBe('Close it once you have what you need: harness close kid')
+ expect(await showCommand(['busy'], h.deps)).toBe(EXIT.ok)
+ expect(h.out.at(-1)).not.toMatch(/Close it/)
+ expect(await showCommand(['theirs'], h.deps)).toBe(EXIT.ok)
+ expect(h.out.at(-1)).not.toMatch(/Close it/)
+ const mcp = harness(d, { env: { HARNESS_VIA_MCP: '1' } })
+ expect(await showCommand(['kid'], mcp.deps)).toBe(EXIT.ok)
+ expect(mcp.out.at(-1)).toBe('Close it once you have what you need: harness_close (name: "kid")')
+ // Programs read the state themselves.
+ const json = harness(d)
+ expect(await showCommand(['kid', '--json'], json.deps)).toBe(EXIT.ok)
+ expect(json.out.join('\n')).not.toMatch(/Close it/)
+ })
+
it('clips a long answer and says how to read the rest; says when it waits for the person', async () => {
const d = new FakeDaemon()
d.frames = [frame('lead'), frame('long')]
@@ -373,7 +421,7 @@ describe('harness wait', () => {
s.extra.agents_list = () => ({ agents: n++ < 2 ? s.frames : [{ ...s.frames[0], status: 'stopped' }] })
const st = harness(s)
expect(await waitCommand(['x'], st.deps)).toBe(EXIT.stopped)
- expect(st.out[0]).toBe('x: stopped · it was stopped')
+ expect(st.out[0]).toBe('x: stopped · it was stopped; its answer is kept: harness show x')
let m = 0
s.extra.agents_list = () => ({ agents: m++ < 2 ? s.frames : [] })
const g = harness(s)
@@ -405,14 +453,14 @@ describe('harness wait', () => {
const other = harness(d, { pane: undefined, env: { CODEX_THREAD_ID: 'T1' } })
other.ledger.append({ at: NOW, depth: 1, by: { machineId: 'm', agentId: 'codex:T1', name: 'Codex', engine: 'codex' }, child: { machineId: 'm', agentId: 'kid', name: 'kid' } })
expect(await waitCommand(['kid'], other.deps)).toBe(EXIT.stopped)
- expect(other.out).toEqual(['kid: stopped · it was stopped'])
+ expect(other.out).toEqual(['kid: stopped · it was stopped; its answer is kept: harness show kid'])
const person = harness(d, { pane: undefined })
expect(await waitCommand(['kid'], person.deps)).toBe(EXIT.stopped)
// Reopened after it closed it, worked on and stopped again: not its close any more.
d.frames = [frame('lead'), frame('kid', { status: 'stopped', updatedAt: '2026-10-11T12:05:00Z' })]
const reopened = harness(d)
expect(await waitCommand(['kid'], reopened.deps)).toBe(EXIT.stopped)
- expect(reopened.out).toEqual(['kid: stopped · it was stopped'])
+ expect(reopened.out).toEqual(['kid: stopped · it was stopped; its answer is kept: harness show kid'])
})
it('waits for all of several, listing once a round, or the first with --any; a name twice is one harness', async () => {
@@ -738,7 +786,7 @@ describe('harness new, run by an agent', () => {
const live = new Set(['m/lead'])
expect(spawnRefusal(lead, live, l, 'm', NOW)).toBeNull()
l.append({ at: NOW, depth: SPAWN_LIMITS.depth, by: { ...lead, agentId: 'up' }, child: { machineId: 'm', agentId: 'deep', name: 'deep' } })
- expect(spawnRefusal({ ...lead, agentId: 'deep' }, live, l, 'm', NOW)).toMatch(/this far down/)
+ expect(spawnRefusal({ ...lead, agentId: 'deep' }, live, l, 'm', NOW)).toBe(DEPTH_REFUSAL)
for (let i = 0; i < SPAWN_LIMITS.livePerHarness - 1; i++) {
started(l, 'lead', `k${i}`, 'm', NOW - 3_600_000)
live.add(`m/k${i}`)
@@ -746,7 +794,7 @@ describe('harness new, run by an agent', () => {
started(l, 'lead', 'far', 'mini', NOW - 3_600_000)
expect(spawnRefusal(lead, live, l, 'm', NOW)).toBeNull()
live.add('mini/far')
- expect(spawnRefusal(lead, live, l, 'm', NOW)).toBe('You have 8 harnesses running. Close one you are done with first (harness close <name>).')
+ expect(spawnRefusal(lead, live, l, 'm', NOW)).toBe('You have 4 harnesses running. Close one you are done with first (harness close <name>).')
const crowd = new Ledger(ledgerPath(mkdtempSync(join(dir, 'crowd-'))))
const busy = new Set<string>()
for (let i = 0; i < SPAWN_LIMITS.livePerMachine; i++) {
@@ -759,6 +807,36 @@ describe('harness new, run by an agent', () => {
expect(spawnRefusal(lead, new Set(), quick, 'm', NOW)).toBe('You started 20 harnesses in 10 minutes. Wait, or work with the ones you have.')
})
+ it('keeps a harness another agent started from starting more: four of its own at most, one level down', () => {
+ // Owner, October 12: "be very careful about when to start a new agent so we don't waste user token". They were eight
+ // and two levels: a harness an agent started could start more, and each of those spent the person's tokens.
+ expect(SPAWN_LIMITS).toMatchObject({ livePerHarness: 4, depth: 1, livePerMachine: 40, perWindow: 20 })
+ const l = new Ledger(ledgerPath(dir))
+ const kid = { machineId: 'm', agentId: 'kid', name: 'kid', engine: 'codex' }
+ // The person's harness: no line, no parent.
+ expect(spawnRefusal(kid, new Set(), l, 'm', NOW)).toBeNull()
+ // Its record says another agent started it.
+ started(l, 'lead', 'kid')
+ expect(spawnRefusal(kid, new Set(), l, 'm', NOW)).toBe('A harness another agent started may not start harnesses itself. Do the work yourself, or say in your answer what else is needed.')
+ // The record lost its line, but the core's row names its starter.
+ const lost = new Ledger(ledgerPath(mkdtempSync(join(dir, 'lost-'))))
+ expect(spawnRefusal({ ...kid, startedByAgent: true }, new Set(), lost, 'm', NOW)).toBe(DEPTH_REFUSAL)
+ })
+
+ it('refuses a harness another agent started, from its pane, before anything is asked', async () => {
+ const d = new FakeDaemon()
+ d.frames = [frame('lead', { engine: 'claude' }), frame('kid', { parentAgentId: 'lead' })]
+ const h = harness(d, { pane: '%kid' })
+ const origin = (await agentOrigin(h.deps))!
+ expect(origin.parentAgentId).toBe('kid')
+ await expect(origin.check()).rejects.toThrow(DEPTH_REFUSAL)
+ expect(d.requests.filter((r) => r.type === 'agent_create')).toEqual([])
+ // Started from another machine, or by an agent outside Harness: no starter here, and no line in this machine's
+ // record. The core's row still says another agent started it.
+ d.frames = [frame('far', { startedBy: 'agent' })]
+ await expect((await agentOrigin(harness(d, { pane: '%far' }).deps))!.check()).rejects.toThrow(DEPTH_REFUSAL)
+ })
+
it('counts harnesses asked for and still installing as running, toward every limit', () => {
// Each starts by itself once its engine is installed: asked for meanwhile, they passed every limit and started at once.
const lead = { machineId: 'm', agentId: 'lead', name: 'lead', engine: 'claude' }
@@ -767,7 +845,7 @@ describe('harness new, run by an agent', () => {
const empty = new Ledger(ledgerPath(mkdtempSync(join(dir, 'waiting-'))))
expect(spawnRefusal(lead, new Set(), empty, 'm', NOW, waiting(SPAWN_LIMITS.livePerHarness - 1))).toBeNull()
expect(spawnRefusal(lead, new Set(), empty, 'm', NOW, waiting(SPAWN_LIMITS.livePerHarness)))
- .toBe('You have 8 harnesses running or starting. Close one you are done with first (harness close <name>).')
+ .toBe('You have 4 harnesses running or starting. Close one you are done with first (harness close <name>).')
const crowd = new Ledger(ledgerPath(mkdtempSync(join(dir, 'crowd-waiting-'))))
const busy = new Set<string>()
for (let i = 0; i < SPAWN_LIMITS.livePerMachine - 1; i++) {
diff --git a/cli/src/agentClient/commands.ts b/cli/src/agentClient/commands.ts
index 349a0f9bf..5fcfcdf48 100644
--- a/cli/src/agentClient/commands.ts
+++ b/cli/src/agentClient/commands.ts
@@ -418,12 +418,27 @@ export function psCommand(argv: readonly string[], deps: ClientDeps = {}): Promi
machine: a.row.creation.machineId === o.where.machineId ? o.machine : a.row.creation.machineId.slice(0, 8), state: a.state,
folder: null, branch: null, startedBy: 'you', yours: true, self: false, lastActive: new Date(a.row.at).toISOString(), about: askedLine(s.env, a),
})))
- s.out(flags.has('--json') ? JSON.stringify({ you: o.caller ? o.caller.name : null, machine: o.machine, harnesses: rows }, null, 1) : formatRows(rows, o.caller, s.now()))
+ const note = o.caller ? openNote(s.env, rows) : null
+ s.out(flags.has('--json') ? JSON.stringify({ you: o.caller ? o.caller.name : null, machine: o.machine, harnesses: rows }, null, 1)
+ : [formatRows(rows, o.caller, s.now()), note].filter(Boolean).join('\n'))
return EXIT.ok
} finally { o.daemon.close() }
})
}
+/** What `show` ends with for an idle harness of the caller's. */
+export const closeHint = (env: NodeJS.ProcessEnv, frame: Frame): string =>
+ `Close it once you have what you need: ${step(env, 'close', addressOf({ machineId: '', agentId: frame.id, name: frame.name ?? '', engine: '' }))}`
+
+/** What `ps` ends with for an agent with harnesses of its own open: how many, and how many sit idle. */
+export function openNote(env: NodeJS.ProcessEnv, rows: readonly Row[]): string | null {
+ const open = rows.filter((r) => r.yours && r.state !== 'stopped' && r.state !== 'failed')
+ if (!open.length) return null
+ const idle = open.filter((r) => r.state === 'idle').length
+ return `You have ${open.length} harness${open.length === 1 ? '' : 'es'} of your own open${idle ? `, ${idle} idle` : ''}. `
+ + `Close each once you have its answer: ${step(env, 'close', '<name>')}.`
+}
+
// ── show ───────────────────────────────────────────────────────────────────────────────────────────
export function showCommand(argv: readonly string[], deps: ClientDeps = {}): Promise<number> {
@@ -460,6 +475,8 @@ export function showCommand(argv: readonly string[], deps: ClientDeps = {}): Pro
if (result.truncated) s.out(`… ${answer.length - chars} more characters: ${step(s.env, 'show', result.name, { flag: `--chars ${answer.length}`, arg: `chars: ${answer.length}` })}`)
}
if (state === 'needs-input') s.out(`It is waiting for the person (a question or a permission). Only the person answers it.`)
+ // Idle and the caller's: a harness left open after its answer is read sits there for nobody (owner, October 12).
+ if (state === 'idle' && t.yours) s.out(closeHint(s.env, t.frame))
}
return EXIT.ok
} finally { if (t?.remote) t.daemon.close(); o.daemon.close() }
@@ -606,7 +623,9 @@ export function waitCommand(argv: readonly string[], deps: ClientDeps = {}): Pro
: r.state === 'timeout' ? `still working; wait again or read it: ${step(s.env, 'show', r.name)}`
: r.state === 'installing' || r.state === 'starting' ? `${askedLine(s.env, following.find((a) => a.row.creation.creationId === r.id)!)}`
: r.state === 'failed' ? `it never started: ${r.detail || 'its machine refused it'}`
- : r.state === 'stopped' ? 'it was stopped' : 'it is gone'
+ // Harness closes one another agent started once it has sat idle half an hour (services/agentCleanup.ts): its
+ // answer is still there to read.
+ : r.state === 'stopped' ? `it was stopped; its answer is kept: ${step(s.env, 'show', r.name)}` : 'it is gone'
s.out(`${r.name}: ${r.state} · ${next}`)
}
const states = results.map((r) => r.state)
@@ -685,8 +704,15 @@ export function closeCommand(argv: readonly string[], deps: ClientDeps = {}): Pr
// ── new, run by an agent ───────────────────────────────────────────────────────────────────────────
-/** Limits on starting harnesses (the design's "Limits"): runaway spawning is the risk. */
-export const SPAWN_LIMITS = { livePerHarness: 8, livePerMachine: 40, depth: 2, perWindow: 20, windowMs: 10 * 60_000 } as const
+/**
+ * Limits on starting harnesses (the design's "Limits"): runaway spawning is the risk, and every harness started is
+ * the person's tokens. Four at once and one level down (owner, October 12: "be very careful about when to start a new
+ * agent so we don't waste user token"): a harness another agent started does its work itself. They were eight and two.
+ */
+export const SPAWN_LIMITS = { livePerHarness: 4, livePerMachine: 40, depth: 1, perWindow: 20, windowMs: 10 * 60_000 } as const
+
+/** Said to a harness another agent started when it runs `harness new`. */
+export const DEPTH_REFUSAL = 'A harness another agent started may not start harnesses itself. Do the work yourself, or say in your answer what else is needed.'
/** How long a harness the caller started by a name stops it starting another by that name: a retry, not new work. */
export const DUPLICATE_WITHIN_MS = 10 * 60_000
@@ -699,9 +725,9 @@ const memberKey = (machineId: string, agentId: string): string => `${machineId}/
* machine's, and the caller's on the other machines it reached.
*/
export function spawnRefusal(caller: Caller, live: ReadonlySet<string>, ledger: Ledger, localMachineId: string, now: number, waiting: readonly Pending[] = []): string | null {
- if (ledger.depthOf(caller) >= SPAWN_LIMITS.depth) {
- return 'A harness started this far down may not start more. Do the work yourself, or say so in your answer.'
- }
+ // Its depth by the record, or by the core's row when the record has no line for it: rotated away, a write that failed,
+ // or a start from another machine or from outside Harness, recorded only where the starter ran.
+ if (Math.max(ledger.depthOf(caller), caller.startedByAgent ? 1 : 0) >= SPAWN_LIMITS.depth) return DEPTH_REFUSAL
// Harnesses asked for and not started yet ([waiting]) count as running: each starts by itself once its engine is
// installed, and an agent asking for more meanwhile would otherwise pass every limit until they all started at once.
const mine = ledger.childrenOf(caller)
diff --git a/cli/src/agentClient/mcp.spec.ts b/cli/src/agentClient/mcp.spec.ts
index 8cd54e967..90c409a19 100644
--- a/cli/src/agentClient/mcp.spec.ts
+++ b/cli/src/agentClient/mcp.spec.ts
@@ -38,6 +38,9 @@ describe('harness mcp', () => {
expect(send.description).toMatch(/teammate.*Agent Messages on.*Refused anywhere else, even to a harness you started/)
expect(INSTRUCTIONS).toContain('starting a harness does not let you message it')
expect(INSTRUCTIONS).not.toContain('Send follow-ups for the same work to the harness you started')
+ // When to start one at all, and to close it, as every harness is told as it starts (lib/agentAttribution.ts).
+ expect(INSTRUCTIONS).toContain('Every harness spends the person\'s tokens. Start one (harness_start) only for work that can run in parallel')
+ expect(INSTRUCTIONS).toContain('close it once you have its answer (harness_close). At most 4 of yours at once, and a harness another agent started may not start harnesses itself.')
})
it('says which tools only read or add, so Codex runs them without asking; closing still says it is destructive', async () => {
diff --git a/cli/src/agentClient/mcp.ts b/cli/src/agentClient/mcp.ts
index 1a96446f8..25e9021a8 100644
--- a/cli/src/agentClient/mcp.ts
+++ b/cli/src/agentClient/mcp.ts
@@ -23,11 +23,11 @@ export const PROTOCOL_VERSIONS = ['2025-11-25', '2025-06-18', '2025-03-26', '202
/** What an agent is told once, as it connects: when Harness is the right tool, and whose harnesses are whose. */
export const INSTRUCTIONS = `Harness runs coding agents (Claude Code, Codex) in sessions of their own, on this computer or the person's other machines. The person can watch and join every one in the Harness app.
-Start a harness (harness_start) for work that can run beside yours, needs another agent's view (a review, a second opinion from Codex), takes long, or belongs in its own folder. Do small, quick work yourself. For scratch or new work give it a project name (project), not a folder of your own: Claude Code and Codex ask the person to trust a folder Harness did not make.
+Every harness spends the person's tokens. Start one (harness_start) only for work that can run in parallel with yours, needs another model's view (a review, a second opinion from Codex), takes long, or belongs in its own folder. Do quick work yourself. For scratch or new work give it a project name (project), not a folder of your own: Claude Code and Codex ask the person to trust a folder Harness did not make.
List first (harness_list). Put everything in the task: starting a harness does not let you message it. A follow-up goes by harness_send only to a teammate in a tab where the person turned Agent Messages on; otherwise start a new harness with the fuller task. Never give work to, or close, a harness you did not start: those are the person's.
-Name every harness. Then wait for it (harness_wait), read its answer (harness_read), and close it when you are done (harness_close). Never ask a harness to publish, deploy, delete or push.`
+Name every harness. Then wait for it (harness_wait), read its answer (harness_read), and close it once you have its answer (harness_close). At most 4 of yours at once, and a harness another agent started may not start harnesses itself. Never ask a harness to publish, deploy, delete or push.`
/** How long `harness_wait` waits when not told: within what MCP clients allow a call. Codex ends one at 60 s unless its
* config says otherwise (`tool_timeout_sec`, developers.openai.com/codex/mcp), so 120 s ended every long wait as an
diff --git a/cli/src/agentClient/model.ts b/cli/src/agentClient/model.ts
index 62b8204a4..1526264c9 100644
--- a/cli/src/agentClient/model.ts
+++ b/cli/src/agentClient/model.ts
@@ -18,6 +18,10 @@ export interface Frame {
project?: { cwd?: string; root?: string; branch?: string | null; worktree?: boolean } | null
createdAt?: string
updatedAt?: string
+ /** The harness here whose agent started this one (registry `parentAgentId`); null for the rest. */
+ parentAgentId?: string | null
+ /** `agent` when another agent started it, here or from another machine or outside Harness. */
+ startedBy?: string | null
}
/** The agent running the verb: a harness in its own pane, or an agent outside Harness. Null for a person at a terminal. */
@@ -27,6 +31,8 @@ export interface Caller extends LedgerMember {
/** An agent Harness did not start (in a terminal app, an editor, the person's own tmux): no pane, so nothing can
* message it, and it is told apart by the session its shell names. */
outside?: boolean
+ /** Another agent started this harness, as the core's row says (`startedBy`, `parentAgentId`): one level down. */
+ startedByAgent?: boolean
}
/** `installing`, `starting`, `failed`: a harness asked for that has not started (commands.ts `Asked`). */
@@ -75,7 +81,7 @@ export function callerOf(frames: Frame[], pane: string | undefined, machineId: s
const on = frames.filter((f) => f.tmuxPane === pane)
const frame = on.find((f) => f.status === 'active') ?? on.filter(isAgent).sort((a, b) => (b.updatedAt ?? '').localeCompare(a.updatedAt ?? ''))[0]
if (!frame || !isAgent(frame)) return null
- return { machineId, agentId: frame.id, name: frame.name || frame.id.slice(0, 8), engine: frame.engine ?? '' }
+ return { machineId, agentId: frame.id, name: frame.name || frame.id.slice(0, 8), engine: frame.engine ?? '', ...(frame.startedBy === 'agent' || frame.parentAgentId ? { startedByAgent: true } : {}) }
}
/**
diff --git a/cli/src/core/AGENTS.md b/cli/src/core/AGENTS.md
index 27332bc1d..e366c567c 100644
--- a/cli/src/core/AGENTS.md
+++ b/cli/src/core/AGENTS.md
@@ -62,7 +62,8 @@ questions. Everything else is a service ([../services/AGENTS.md](../services/AGE
`monitorLink.ts`: the monitor's port, asked of its process. `recapsLink.ts`: the recaps in their process,
told each turn's lifecycle and never waited on, and read back from what they last said
(`turns/recaps.ts` is the core's whole side of them). `storeLink.ts`: what the Store in its
- process tells the core. `modelsLink.ts`: what the core keeps of models in its process for frames and
+ process tells the core. `agentCleanupLink.ts`: the agent cleanup in its process, told every stop and the person
+ typing into a harness another agent started, and answered what it may close. `modelsLink.ts`: what the core keeps of models in its process for frames and
keystrokes, and how it asks it the rest. `agentQueries.ts`: what a service in its own process may ask of the agents;
`accountQueries.ts`: of the account (a token, and the fleet's lane's sealing), so it holds no credential.
`experimentQueries.ts`: what an experiment acts on the core through; `experiments.ts`: which are on as the
diff --git a/cli/src/core/agentCleanupLink.spec.ts b/cli/src/core/agentCleanupLink.spec.ts
new file mode 100644
index 000000000..d1b8c4259
--- /dev/null
+++ b/cli/src/core/agentCleanupLink.spec.ts
@@ -0,0 +1,59 @@
+import { describe, expect, it, vi } from 'vitest'
+import type { RegisteredSession } from '../lib/registry.js'
+import { fakeCore } from '../testing/fakeCore.js'
+import { createAgentCleanupLink } from './agentCleanupLink.js'
+
+const row = (agentId: string, over: Partial<RegisteredSession> = {}) => ({ agentId, sessionId: `s-${agentId}`, ...over }) as RegisteredSession
+
+function setup() {
+ const kid = row('kid', { startedBy: 'agent' })
+ const mine = row('mine')
+ const sessions = [kid, mine]
+ const resolve = vi.fn((id: string) => sessions.find((s) => s.agentId === id || s.sessionId === id))
+ const closeStarted = vi.fn(async () => ({ closed: true as const }))
+ const lastActivity = vi.fn(async () => 42)
+ const core = fakeCore({ agents: { resolve, closeStarted, lastActivity, live: vi.fn(() => sessions) } })
+ const notify = vi.fn(() => true)
+ return { link: createAgentCleanupLink(core, notify), notify, closeStarted, lastActivity, core }
+}
+
+describe('the agent cleanup in its own process, as the core sees it', () => {
+ it('tells it every stop, held until it hears it', () => {
+ const { link, notify } = setup()
+ link.port.closed('lead')
+ expect(notify).toHaveBeenCalledWith({ type: 'service_event', payload: { kind: 'closed', agentId: 'lead' } }, { untilDelivered: true })
+ })
+
+ it('tells it once that the person typed into a harness another agent started, by its agent id, and nothing of the rest', () => {
+ const { link, notify } = setup()
+ link.port.typed('s-kid')
+ link.port.typed('kid')
+ // Every keystroke into the person's own harness, or one it does not know, costs a lookup and no frame.
+ link.port.typed('mine')
+ link.port.typed('nobody')
+ expect(notify.mock.calls).toEqual([[{ type: 'service_event', payload: { kind: 'typed', agentId: 'kid' } }, { untilDelivered: true }]])
+ expect(link.port.stop()).toBeUndefined()
+ })
+
+ it('answers when a harness last moved, and closes what the core lets it, as it was asked', async () => {
+ const { link, closeStarted, lastActivity } = setup()
+ expect(await link.answer('lastActivity', { agentId: 'kid' })).toEqual({ at: 42 })
+ expect(lastActivity).toHaveBeenCalledWith('kid')
+ expect(await link.answer('closeStarted', { agentId: 'kid', mode: 'now' })).toEqual({ closed: true })
+ expect(await link.answer('closeStarted', { agentId: 'kid', mode: 'idle' })).toEqual({ closed: true })
+ expect(closeStarted.mock.calls).toEqual([['kid', 'now'], ['kid', 'idle']])
+ // Asked wrongly, nothing is closed.
+ expect(await link.answer('closeStarted', { agentId: 'kid', mode: 'inspect' })).toEqual({ error: 'INVALID_REQUEST' })
+ expect(await link.answer('closeStarted', { mode: 'now' })).toEqual({ error: 'INVALID_REQUEST' })
+ expect(await link.answer('lastActivity', { agentId: 7 })).toEqual({ error: 'INVALID_REQUEST' })
+ expect(closeStarted).toHaveBeenCalledTimes(2)
+ // The live agents, as any light service asks for them.
+ expect(await link.answer('live', {})).toEqual({ agents: [row('kid', { startedBy: 'agent' }), row('mine')] })
+ })
+
+ it('closes nothing, and knows no time, in a core that cannot', async () => {
+ const link = createAgentCleanupLink(fakeCore(), () => true)
+ expect(await link.answer('closeStarted', { agentId: 'kid', mode: 'now' })).toEqual({ error: 'UNSUPPORTED' })
+ expect(await link.answer('lastActivity', { agentId: 'kid' })).toEqual({ at: null })
+ })
+})
diff --git a/cli/src/core/agentCleanupLink.ts b/cli/src/core/agentCleanupLink.ts
new file mode 100644
index 000000000..1475e056d
--- /dev/null
+++ b/cli/src/core/agentCleanupLink.ts
@@ -0,0 +1,46 @@
+/**
+ * The agent cleanup in its own process, as the core sees it (services/agentCleanupProcess.ts; in the edge host by
+ * default).
+ *
+ * The core tells the process what happens and goes on: a harness stopped (`service_event closed`), and the person
+ * typing into a harness another agent started (`typed`), each held until the process hears it, as one it missed while
+ * down would leave a harness open, or close one the person joined. `typed` comes with every keystroke a window sends:
+ * it is said once a harness, and only for a harness another agent started, so the person's typing costs two map
+ * lookups and never a frame to another process.
+ *
+ * The process asks the core what only the core knows: the live agents (`live`), when one's conversation last moved
+ * (`lastActivity`), and to close one (`closeStarted`), which the core does for a harness another agent started and
+ * refuses for any other (core/api.ts `closeStarted`). Whatever the process asks, it closes nothing else.
+ */
+import { answerAgentQuery } from './agentQueries.js'
+import type { AgentCleanupPort, CoreApi } from './api.js'
+import type { ServiceFrame } from './serviceLinks.js'
+
+type Payload = Record<string, unknown>
+
+export function createAgentCleanupLink(core: Pick<CoreApi, 'agents' | 'usage'>, notify: (frame: ServiceFrame, opts?: { untilDelivered?: boolean }) => boolean) {
+ /** The harnesses the process was told the person joined: it keeps them itself (services/agentCleanup.ts). */
+ const told = new Set<string>()
+ const port: AgentCleanupPort = {
+ closed: (agentId) => { notify({ type: 'service_event', payload: { kind: 'closed', agentId } }, { untilDelivered: true }) },
+ typed: (id) => {
+ const session = core.agents.resolve(id)
+ if (session?.startedBy !== 'agent' || told.has(session.agentId)) return
+ told.add(session.agentId)
+ notify({ type: 'service_event', payload: { kind: 'typed', agentId: session.agentId } }, { untilDelivered: true })
+ },
+ stop: () => {},
+ }
+
+ const answer = async (query: string, payload: Payload): Promise<Payload> => {
+ if (query !== 'closeStarted' && query !== 'lastActivity') return answerAgentQuery(core, query, payload)
+ const agentId = typeof payload.agentId === 'string' ? payload.agentId : ''
+ if (!agentId) return { error: 'INVALID_REQUEST' }
+ if (query === 'lastActivity') return { at: await (core.agents.lastActivity?.(agentId) ?? null) }
+ const mode = payload.mode === 'idle' || payload.mode === 'now' ? payload.mode : null
+ if (!mode) return { error: 'INVALID_REQUEST' }
+ return { ...await (core.agents.closeStarted?.(agentId, mode) ?? { error: 'UNSUPPORTED' }) }
+ }
+
+ return { port, answer }
+}
diff --git a/cli/src/core/agents/close.spec.ts b/cli/src/core/agents/close.spec.ts
index 5be68e5c0..f89c2c45e 100644
--- a/cli/src/core/agents/close.spec.ts
+++ b/cli/src/core/agents/close.spec.ts
@@ -49,6 +49,30 @@ describe('closing agents no window shows', () => {
expect(close.changed).toBe(deps.announceSession)
})
+ it('knows a close of one already closed: not live, and kept stopped with the conversation the close names', () => {
+ const live = row({ agentId: 'live' })
+ const kept = row({ agentId: 'kept', sessionId: 's-kept', registeredAt: 5 })
+ const registry = { byAgent: vi.fn((id: string) => (id === 'live' ? live : undefined)) } as unknown as ClosingDeps['registry']
+ const stoppedAgents = { get: vi.fn((id: string) => (id === 'kept' || id === 'live' ? kept : null)), resumeReservedAt: vi.fn((_id: string): number | null => null) }
+ const { close } = setup({ registry, stoppedAgents })
+ const asked = (agentId: string, over: Record<string, string> = {}) => ({ agentId, sessionId: 's-kept', createdAt: new Date(5).toISOString(), mode: 'now' as const, ...over })
+ expect(close.closedAlready!(asked('kept'))).toBe(true)
+ // Asked before its conversation was known: the creation alone.
+ expect(close.closedAlready!(asked('kept', { sessionId: '' }))).toBe(true)
+ expect(close.closedAlready!(asked('kept', { sessionId: 's-other' }))).toBe(false)
+ expect(close.closedAlready!(asked('kept', { createdAt: new Date(6).toISOString() }))).toBe(false)
+ expect(close.closedAlready!(asked('live'))).toBe(false)
+ expect(close.closedAlready!(asked('gone'))).toBe(false)
+ // Being resumed: about to be live again, not closed.
+ stoppedAgents.resumeReservedAt.mockReturnValueOnce(1)
+ expect(close.closedAlready!(asked('kept'))).toBe(false)
+ // A record that cannot be read says nothing.
+ stoppedAgents.get.mockImplementationOnce(() => { throw new Error('unreadable') })
+ expect(close.closedAlready!(asked('kept'))).toBe(false)
+ // A core that keeps no stopped agents knows none closed.
+ expect(setup({ registry }).close.closedAlready!(asked('kept'))).toBe(false)
+ })
+
describe('what an agent is doing', () => {
it('reads its newest lines first, then its screen, its turn and whether it is asking', async () => {
const { deps, close } = setup()
diff --git a/cli/src/core/agents/close.ts b/cli/src/core/agents/close.ts
index 56596a67e..814e3cbdb 100644
--- a/cli/src/core/agents/close.ts
+++ b/cli/src/core/agents/close.ts
@@ -11,6 +11,7 @@ import type { OpenTabProtection } from '../../lib/openTabProtection.js'
import { sid } from '../../lib/log.js'
import { projectDisplayName, type registry } from '../../lib/registry.js'
import type { SessionCheckpointStore } from '../../lib/sessionCheckpoint.js'
+import type { StoppedAgentStore } from '../../lib/stoppedAgents.js'
import type { TerminalBackendCoordinator } from '../../lib/terminalBackendCoordinator.js'
import type { Watcher } from '../../watcher/watcher.js'
@@ -29,6 +30,8 @@ export interface ClosingDeps {
announceSession: CloseAgentServiceDeps['changed']
/** Once the engine's worker is linked again, or the wait is over (core/engines/engineLinks.ts). */
engineReady: (engine: string, ms: number) => Promise<void>
+ /** The agents stopped and kept: a close of one already closed is answered closed. */
+ stoppedAgents?: Pick<StoppedAgentStore, 'get' | 'resumeReservedAt'>
}
/** How long a close waits for an engine worker that restarted under its read, before reading once more. */
@@ -36,7 +39,7 @@ export const CLOSE_READ_RECONNECT_MS = 10_000
export function createAgentClosing({
readScreen, registry, cleanupTabs, watcher, captureTerminal, sessionTurnState, openQuestions, terminals, sessionCheckpoints,
- stopAgent, announceSession, engineReady,
+ stopAgent, announceSession, engineReady, stoppedAgents,
}: ClosingDeps) {
/**
* Read what the transcript has that the tail has not delivered yet. A close sent just as the engine's worker
@@ -74,6 +77,17 @@ export function createAgentClosing({
},
stop: stopAgent,
changed: announceSession,
+ // Not live, not being resumed, and kept as stopped with the conversation the close names: it was closed. A record
+ // that cannot be read says nothing, and the close is answered as changed, as before.
+ closedAlready: request => {
+ if (!stoppedAgents || registry.byAgent(request.agentId)) return false
+ try {
+ if (stoppedAgents.resumeReservedAt(request.agentId) !== null) return false
+ const stopped = stoppedAgents.get(request.agentId)
+ return !!stopped && new Date(stopped.registeredAt).toISOString() === request.createdAt
+ && (!request.sessionId || stopped.sessionId === request.sessionId)
+ } catch { return false }
+ },
})
const cleanupPreview = async (): Promise<Record<string, unknown>> => {
await cleanupTabs.refresh()
diff --git a/cli/src/core/agents/lifecycle.spec.ts b/cli/src/core/agents/lifecycle.spec.ts
index c873fb909..ced86ea01 100644
--- a/cli/src/core/agents/lifecycle.spec.ts
+++ b/cli/src/core/agents/lifecycle.spec.ts
@@ -26,6 +26,7 @@ function setup(over: Partial<LifecycleDeps> = {}) {
const deps: LifecycleDeps = {
registry: {
byAgent: vi.fn(() => row()),
+ resolve: vi.fn((id: string) => (id === 'a1' || id === 's1' ? row() : undefined)),
list: vi.fn(() => [row()]),
deleteSavedNames: vi.fn(),
} as unknown as LifecycleDeps['registry'],
@@ -66,7 +67,6 @@ describe('stopping, purging and resuming an agent', () => {
tmuxBackend: null, agentReconciler: deps.agentReconciler, forgetSession: deps.forgetSession,
markDeleted: deps.markDeleted, clearDeleted: deps.clearDeleted, stopNative: deps.stopNative,
})
- expect(lifecycle.stopAgent).toBe(vi.mocked(createStopAgentService).mock.results[0].value)
expect(createResumeAgentService).toHaveBeenCalledWith({
registry: deps.registry, stoppedAgents: deps.stoppedAgents, tmuxBackend: null, restartJobs: deps.restartJobs,
stopJobs: lifecycle.stopJobs, pinnedControls: deps.pinnedControls, retainExitedSession: deps.retainExitedSession,
@@ -76,6 +76,26 @@ describe('stopping, purging and resuming an agent', () => {
})
})
+ it('stops through the stop service, then tells the cleanup the agent was stopped, never while the stop can fail', async () => {
+ // A harness closed takes the harnesses it started with it (services/agentCleanup.ts): a window's close and Stop
+ // Harness, `harness close` and a purge all stop through here.
+ const stopped = vi.fn()
+ const { lifecycle } = setup({ stopped })
+ const stop = vi.mocked(createStopAgentService).mock.results[0].value as ReturnType<typeof vi.fn>
+ const options = { current: () => true }
+ await lifecycle.stopAgent('s1', options)
+ expect(stop).toHaveBeenCalledWith('s1', options)
+ expect(stopped).toHaveBeenCalledWith('a1')
+ // One already gone is told by the id it was asked by: its harnesses are orphans all the same.
+ await lifecycle.stopAgent('gone')
+ expect(stopped).toHaveBeenLastCalledWith('gone')
+ stop.mockRejectedValueOnce(new AgentStopError('Harness changed while saving its conversation. Try stopping again.'))
+ await expect(lifecycle.stopAgent('a1')).rejects.toThrow('Harness changed')
+ expect(stopped).toHaveBeenCalledTimes(2)
+ // With nobody to tell, a stop is a stop.
+ await expect(setup().lifecycle.stopAgent('a1')).resolves.toBeUndefined()
+ })
+
describe('purge', () => {
it('sees live and stopped agents, stops through the stop service, and checkpoints where stop does', () => {
const { deps, lifecycle, purge } = setup()
diff --git a/cli/src/core/agents/lifecycle.ts b/cli/src/core/agents/lifecycle.ts
index c2da4c939..b771444ae 100644
--- a/cli/src/core/agents/lifecycle.ts
+++ b/cli/src/core/agents/lifecycle.ts
@@ -42,12 +42,15 @@ export interface LifecycleDeps {
attachDsh: ResumeAgentServiceDeps['attachDsh']
attachSession: ResumeAgentServiceDeps['attachSession']
relaunchMarks?: ResumeAgentServiceDeps['relaunchMarks']
+ /** An agent was stopped: closed (`agent_close`), Stop Harness, or purged. The harnesses it started close with it
+ * (services/agentCleanup.ts `closed`). */
+ stopped?: (agentId: string) => void
}
export function createAgentLifecycle({
registry, stoppedAgents, restartJobs, tmuxBackend, agentReconciler, forgetSession, markDeleted, clearDeleted, stopNative, settlePane, cancelExternal,
sessionCheckpoints, mirror, sessionSearch, send, pinnedControls, retainExitedSession, announceSession,
- relaunchOverrides, prepareSessionResume, refreshGridWebSearch, attachDsh, attachSession, relaunchMarks,
+ relaunchOverrides, prepareSessionResume, refreshGridWebSearch, attachDsh, attachSession, relaunchMarks, stopped,
}: LifecycleDeps) {
/**
* Stop Harness (`agent_delete`) archives its conversation and launch settings, removes the live
@@ -55,10 +58,19 @@ export function createAgentLifecycle({
* SIGTERM/SIGKILL fallback. Engine conversation files, recaps and the Harness name remain on disk.
*/
const stopJobs = new Map<string, Promise<void>>()
- const stopAgent = createStopAgentService({
+ const stop = createStopAgentService({
registry, stoppedAgents, restartJobs, stopJobs, tmuxBackend, agentReconciler,
forgetSession, markDeleted, clearDeleted, stopNative, settlePane, cancelExternal,
})
+ /**
+ * Every stop goes through here: a window's close and Stop Harness, `harness close`, a purge. Told once the stop is
+ * done, never while it can still fail: a harness closed takes the harnesses it started with it, so none of them
+ * goes on working, spending the person's tokens, for a starter that will never read it (owner, October 12).
+ */
+ const stopAgent: typeof stop = (target, options) => {
+ const agentId = registry.resolve(target)?.agentId ?? target
+ return stop(target, options).then(() => { stopped?.(agentId) })
+ }
const purgeAgentService = new PurgeAgentService({
live: id => registry.byAgent(id), sessions: () => [...registry.list(), ...stoppedAgents.list()],
stopped: stoppedAgents, checkpoints: sessionCheckpoints, stop: stopAgent,
diff --git a/cli/src/core/api.spec.ts b/cli/src/core/api.spec.ts
index 050cd443f..6be57c8bf 100644
--- a/cli/src/core/api.spec.ts
+++ b/cli/src/core/api.spec.ts
@@ -1,6 +1,6 @@
import { describe, expect, it, vi } from 'vitest'
import { projectDisplayName, type RegisteredSession } from '../lib/registry.js'
-import { CONNECTOR_VAULT_OFF, CONVERSATIONS_OFF, ACCOUNT_BACKEND_OFF, AGENT_ACTIONS_OFF, createCoreApi, DAEMON_UNKNOWN, DELIVERIES_OFF, DEVICES_FALLBACKS, emptyPorts, LANE_OFF, LONG_ANSWERS, MODELS_OFF, STORE_OFF, MODELS_REQUESTS, MONITOR_OFF, OBSERVER_KEY_OFF, ORCHESTRATOR_FALLBACKS, RECAPS_FALLBACKS, resolveAgent, SHARING_FALLBACKS, MESSAGES_FALLBACKS, TERMINALS_OFF, WIFI_FALLBACKS, WIFI_OFF, type CoreApiDeps } from './api.js'
+import { CONNECTOR_VAULT_OFF, CONVERSATIONS_OFF, ACCOUNT_BACKEND_OFF, AGENT_ACTIONS_OFF, AGENT_CLEANUP_FALLBACKS, createCoreApi, DAEMON_UNKNOWN, DELIVERIES_OFF, DEVICES_FALLBACKS, emptyPorts, LANE_OFF, LONG_ANSWERS, MODELS_OFF, STORE_OFF, MODELS_REQUESTS, MONITOR_OFF, OBSERVER_KEY_OFF, ORCHESTRATOR_FALLBACKS, RECAPS_FALLBACKS, resolveAgent, SHARING_FALLBACKS, MESSAGES_FALLBACKS, TERMINALS_OFF, WIFI_FALLBACKS, WIFI_OFF, type CoreApiDeps } from './api.js'
import { FAIL, readFallback } from './serviceHost.js'
const row = (agentId: string) => ({ agentId, sessionId: `s-${agentId}`, engine: 'claude', cwd: '/work/app' }) as RegisteredSession
@@ -147,6 +147,34 @@ describe('the core API services stand on', () => {
expect(core.transcripts.lastTurn).toBe(deps.lastTurn)
})
+ it('closes for the agent cleanup only a harness another agent started, the conversation it is now, and keeps it', async () => {
+ // Whatever the cleanup asks, from its own process or this one, the core decides what it may close
+ // (services/agentCleanup.ts): a harness the person started is never one.
+ const started = { ...row('kid'), startedBy: 'agent' as const, registeredAt: Date.parse('2026-10-12T08:00:00Z') }
+ const mine = { ...row('mine'), registeredAt: 0 }
+ const deps = coreDeps()
+ const byAgent = vi.fn((agentId: string) => (agentId === 'kid' ? started : agentId === 'mine' ? mine : undefined))
+ const closeAgent = vi.fn(async () => ({ closed: true as const }))
+ const activityAt = vi.fn(async () => 1_234)
+ const core = createCoreApi({ ...deps, registry: { ...deps.registry, byAgent } as unknown as CoreApiDeps['registry'], closeAgent, activityAt })
+ expect(await core.agents.closeStarted!('kid', 'idle')).toEqual({ closed: true })
+ expect(closeAgent).toHaveBeenCalledWith({ agentId: 'kid', sessionId: 's-kid', createdAt: '2026-10-12T08:00:00.000Z', mode: 'idle' })
+ expect(await core.agents.closeStarted!('mine', 'now')).toEqual({ error: 'NOT_AGENT_STARTED' })
+ expect(await core.agents.closeStarted!('gone', 'now')).toEqual({ error: 'AGENT_CHANGED' })
+ expect(closeAgent).toHaveBeenCalledTimes(1)
+ expect(await core.agents.lastActivity!('kid')).toBe(1_234)
+ expect(activityAt).toHaveBeenCalledWith(started)
+ expect(await core.agents.lastActivity!('gone')).toBeNull()
+ // A core built without a close or a clock closes nothing and knows no time.
+ const bare = createCoreApi({ ...deps, registry: { ...deps.registry, byAgent } as unknown as CoreApiDeps['registry'] })
+ expect(await bare.agents.closeStarted!('kid', 'now')).toEqual({ error: 'UNSUPPORTED' })
+ expect(await bare.agents.lastActivity!('kid')).toBeNull()
+ })
+
+ it('tells the agent cleanup nothing while it is off: every harness is left open, as before it', () => {
+ expect(AGENT_CLEANUP_FALLBACKS).toEqual({ closed: undefined, typed: undefined, stop: undefined })
+ })
+
it('lets the recaps send a turn\'s cards and recaps, and nothing else, down those doors', () => {
const log = vi.fn()
const turnCard = vi.fn()
@@ -247,7 +275,7 @@ describe('the core API services stand on', () => {
})
it('starts with every port empty: a service fills its own when it starts', () => {
- expect(emptyPorts()).toEqual({ usage: null, store: null, search: null, viewers: null, models: null, workspaces: null, messages: null, devices: null, wifi: null, monitor: null, orchestrator: null, sharing: null, recaps: null })
+ expect(emptyPorts()).toEqual({ usage: null, store: null, search: null, viewers: null, models: null, workspaces: null, messages: null, devices: null, wifi: null, monitor: null, orchestrator: null, sharing: null, recaps: null, agentCleanup: null })
expect(emptyPorts()).not.toBe(emptyPorts())
})
diff --git a/cli/src/core/api.ts b/cli/src/core/api.ts
index fabbc2a73..e90d48764 100644
--- a/cli/src/core/api.ts
+++ b/cli/src/core/api.ts
@@ -32,6 +32,7 @@ import type { NewAgentModel } from '../lib/newAgentModel.js'
import type { LastTurnText, LiveEvent } from '../engines/kit/events.js'
import type { SessionRecaps } from '../lib/recapReads.js'
import type { SessionInputDelivery } from '../lib/sessionInput.js'
+import type { AgentCloseRequest, AgentCloseResult } from '../lib/closeAgentService.js'
import { projectDisplayName, type registry, type RegisteredSession } from '../lib/registry.js'
import type { RuntimeModelOption } from '../lib/runtimeProfile.js'
import { externalUnavailable, type ExternalSessionAnswer, type ExternalSessionRequest } from '../lib/externalSessionWire.js'
@@ -169,6 +170,16 @@ export interface CoreApi {
/** `activityText` with the footer's timer, for a Harness Pro. Absent in a
* core that cannot read them; the dial then gets the footer alone. */
activitySnapshot?(agentId: string): Promise<{ text: string; elapsedSeconds?: number } | null>
+ /** Close a harness another agent started (`startedBy: agent`) as `agent_close` closes one: its screen saved, its
+ * process stopped, its conversation kept. `idle` refuses one that works, waits for the person or holds a draft
+ * (`SESSION_NOT_IDLE`); `now` does not. Any other harness is refused (`NOT_AGENT_STARTED`): the agent cleanup
+ * (services/agentCleanup.ts) closes those, and nothing else. Absent where the core does not let a service close. */
+ closeStarted?(agentId: string, mode: 'idle' | 'now'): Promise<AgentCloseResult>
+ /** When a live agent's conversation last moved, epoch ms: the latest dated work in its transcript
+ * (lib/transcriptActivity.ts), what the apps' `updatedAt` is when there is one. Null for none live, and for an
+ * engine that keeps no dated transcript: a hook's time or the bind's is not when it last answered. Absent where
+ * the core is not asked it. */
+ lastActivity?(agentId: string): Promise<number | null>
}
/** A device's or another machine's turns for an agent on this one: the doors the web and the hooks use. */
turns: {
@@ -691,6 +702,21 @@ export interface RecapsPort {
/** What the core gets when the recaps fail: nothing told, no recap, and no live card. */
export const RECAPS_FALLBACKS: PortFallbacks<RecapsPort> = { lifecycle: undefined, recaps: null, liveCards: later([]) }
+/**
+ * The core's calls into the agent cleanup (services/agentCleanup.ts): what happens to the harnesses agents start.
+ * Notices: the core never waits for them, and one the cleanup misses leaves a harness open, as before it.
+ */
+export interface AgentCleanupPort {
+ /** A harness was closed or stopped (`agent_close`, Stop Harness, a purge): the harnesses it started close too. */
+ closed(agentId: string): void
+ /** The person typed or pasted into a harness, from a window or the phone (an agent id or a session id): one another
+ * agent started is the person's from then on. */
+ typed(id: string): void
+ stop(): void
+}
+/** What the core gets when the cleanup fails: nothing told, and every harness left open, as before it. */
+export const AGENT_CLEANUP_FALLBACKS: PortFallbacks<AgentCleanupPort> = { closed: undefined, typed: undefined, stop: undefined }
+
/** The core's calls into the DSH viewers: each harness agent's viewer server and verdict watch. */
export interface ViewersPort {
/** Start the agent's viewer and verdict watch when it has a DSH. Idempotent: called on every
@@ -1335,10 +1361,11 @@ export interface CorePorts {
orchestrator: OrchestratorPort | null
sharing: SharingPort | null
recaps: RecapsPort | null
+ agentCleanup: AgentCleanupPort | null
}
export function emptyPorts(): CorePorts {
- return { usage: null, store: null, search: null, viewers: null, models: null, workspaces: null, messages: null, devices: null, wifi: null, monitor: null, orchestrator: null, sharing: null, recaps: null }
+ return { usage: null, store: null, search: null, viewers: null, models: null, workspaces: null, messages: null, devices: null, wifi: null, monitor: null, orchestrator: null, sharing: null, recaps: null, agentCleanup: null }
}
export interface CoreApiDeps {
@@ -1377,6 +1404,10 @@ export interface CoreApiDeps {
machine: CoreApi['machine']
activityText: CoreApi['agents']['activityText']
activitySnapshot?: CoreApi['agents']['activitySnapshot']
+ /** One `agent_close` (core/agents/close.ts), for `closeStarted`; absent, a service closes nothing. */
+ closeAgent?: (request: AgentCloseRequest) => Promise<AgentCloseResult>
+ /** When an agent's conversation last moved (lib/transcriptActivity.ts), for `lastActivity`. */
+ activityAt?: (session: RegisteredSession) => Promise<number | null>
signedIn: CoreApi['account']['signedIn']
environment: CoreApi['account']['environment']
machines: CoreApi['account']['machines']
@@ -1400,6 +1431,7 @@ export function createCoreApi({
dataDir, registry, stoppedAgents, databaseHistory, syncSession, runtimeModels, viewerChanged,
gridNamed, gridModelsChanged, dshInstallStatus, mintGridName, accessToken, lane, observerKey, connectorVault, observer, privateGridName, machineName, backend, onNotice,
runtimeProfile, setRuntime, fork, create, dsh, windows, daemon, turns, questions, terminals = TERMINALS_OFF, conversations = CONVERSATIONS_OFF, viewerFrame, machine, activityText, activitySnapshot,
+ closeAgent, activityAt,
signedIn, environment, machines, sendLocal, sendToWindow, hasWindow, devicesChanged, dialWatching, wifi, lastTurn, turnCard, turnSummary,
log = (line) => console.warn(line),
}: CoreApiDeps): CoreApi {
@@ -1432,6 +1464,19 @@ export function createCoreApi({
dsh,
activityText,
activitySnapshot,
+ // Decided here, not in the service: whatever the cleanup asks, from its process or this one, the core closes only
+ // a harness another agent started, the conversation it reviewed (`agent_close`'s identity), and keeps it.
+ closeStarted: async (agentId, mode) => {
+ const session = registry.byAgent(agentId)
+ if (!session) return { error: 'AGENT_CHANGED' }
+ if (session.startedBy !== 'agent') return { error: 'NOT_AGENT_STARTED' }
+ if (!closeAgent) return { error: 'UNSUPPORTED' }
+ return closeAgent({ agentId, sessionId: session.sessionId, createdAt: new Date(session.registeredAt).toISOString(), mode })
+ },
+ lastActivity: async (agentId) => {
+ const session = registry.byAgent(agentId)
+ return session && activityAt ? activityAt(session) : null
+ },
},
turns,
questions,
diff --git a/cli/src/core/input.spec.ts b/cli/src/core/input.spec.ts
index dc38ae191..707dc2e22 100644
--- a/cli/src/core/input.spec.ts
+++ b/cli/src/core/input.spec.ts
@@ -448,11 +448,14 @@ describe('message', () => {
afterEach(() => vi.restoreAllMocks())
it('writes what a person typed into the agent the frame names, and reads no tab an older window still names', () => {
- const inputs = createInput(deps().deps)
+ const typed = vi.fn()
+ const inputs = createInput({ ...deps().deps, typed })
const submit = vi.spyOn(inputs.input, 'submit').mockImplementation(() => {})
vi.spyOn(console, 'log').mockImplementation(() => {})
inputs.messageRequest({ content: 'hello', agentId: 's1', tabId: 'swarm-a' })
expect(submit.mock.calls).toEqual([['a1', 'hello', undefined]])
+ // The person sent it: a harness another agent started is theirs from then on (services/agentCleanup.ts).
+ expect(typed).toHaveBeenCalledWith('s1')
})
it('names the agent by agent id or session id, and drops a frame with no text or no agent', () => {
diff --git a/cli/src/core/input.ts b/cli/src/core/input.ts
index a4382d12f..82f4bf775 100644
--- a/cli/src/core/input.ts
+++ b/cli/src/core/input.ts
@@ -53,6 +53,9 @@ export interface InputDeps {
/** A prompt is about to be typed, and `capture` is the pane as read right before: what is on it belongs
* to the turns before the one it starts (the question watcher's, lib/askQuestion.ts `notePrompt`). */
promptTyped?: (session: RegisteredSession, capture: string | null, screen: ScreenReading) => void
+ /** The person sent text to an agent from a window or the phone (a `message` frame, by agent or session id): one
+ * another agent started is theirs from then on (services/agentCleanup.ts). */
+ typed?: (id: string) => void
}
/**
@@ -225,7 +228,7 @@ export function createInput(deps: InputDeps) {
console.log(`[msg] ${sid(sessionId)} recv · engine=${engine} · bytes=${Buffer.byteLength(adapted, 'utf8')}`)
input.submit(record?.agentId ?? sessionId, adapted, deliveryId)
}
- const messageRequest = createMessageRequest(submitAgent)
+ const messageRequest = createMessageRequest(submitAgent, deps.typed)
return { input, deviceInput, acquireTerminalControl, submitAgent, messageRequest }
}
@@ -236,11 +239,14 @@ export function createInput(deps: InputDeps) {
*
* Moved verbatim out of the socket's request switch (docs/design/2026-10-03-harnessd.md).
*/
-export function createMessageRequest(submit: (id: string, content: string) => void) {
+export function createMessageRequest(submit: (id: string, content: string) => void, typed?: (id: string) => void) {
return (payload: Record<string, unknown>): void => {
const content = payload.content as string | undefined
const target = (payload.agentId as string | undefined) || (payload.sessionId as string | undefined)
if (!content || !target) return
+ // A `message` frame is a person's: the windows' and the phone's composer. Agents and experiments deliver theirs
+ // under a delivery id (`turns.deliver`), which never comes this way.
+ typed?.(target)
// From the relay this is only reached sealed: text typed into an agent is never taken from the relay
// in the clear.
submit(target, content)
diff --git a/cli/src/core/main.ts b/cli/src/core/main.ts
index 907cbab31..6ea60337c 100644
--- a/cli/src/core/main.ts
+++ b/cli/src/core/main.ts
@@ -173,7 +173,8 @@ import { createMonitorLink } from './monitorLink.js'
import { createStoreLink } from './storeLink.js'
import { createModelsLink } from './modelsLink.js'
import { createRecapsLink } from './recapsLink.js'
-import { RECAPS_FALLBACKS, TERMINALS_OFF } from './api.js'
+import { createAgentCleanupLink } from './agentCleanupLink.js'
+import { AGENT_CLEANUP_FALLBACKS, RECAPS_FALLBACKS, TERMINALS_OFF } from './api.js'
import { answerAgentQuery } from './agentQueries.js'
import { createDeliveries } from './deliveries.js'
import { createTerminalWatch } from './terminalWatch.js'
@@ -209,6 +210,7 @@ import { cursorDataDir } from '../engines/cursor/contract.js'
import { TranscriptPager } from '../lib/transcriptPages.js'
import { AgentCreationReceipts } from '../lib/agentCreationReceipt.js'
import { agentFrame, lastActivityAt, type AgentFrame } from '../lib/agentFrame.js'
+import { transcriptActivityAt } from '../lib/transcriptActivity.js'
import { forgetAgentProject } from '../lib/agentProject.js'
import { createUsageLink } from './usageLink.js'
import type { UsageSnapshots } from './usageLink.js'
@@ -825,6 +827,10 @@ async function runForeground(session: AuthSession | null): Promise<void> {
},
// The window's `agent_create` for an experiment's agents (core/experimentQueries.ts).
create: (request) => createForExperiment(backendRef?.onCreateAgent ?? null, request),
+ // The window's `agent_close`, for the agent cleanup (services/agentCleanup.ts): createCoreApi lets it close only
+ // a harness another agent started. Read when it closes one, never now: the close service is built further down.
+ closeAgent: (request) => backendRef?.closeAgentService?.request(request) ?? Promise.resolve({ error: 'UNSUPPORTED' }),
+ activityAt: (session) => transcriptActivityAt(session.transcriptPath, session.engine),
dsh: (session) => backendRef?.dshFrameProvider?.(session) ?? null,
windows: (frame) => backendRef?.sendLocal(frame),
daemon: { command: daemonCommand(), port: env.PORT, machineId: () => backendRef?.machineId ?? '', autonomousEnv: session?.autonomousEnv ?? env.AUTONOMOUS_ENV },
@@ -1001,9 +1007,11 @@ async function runForeground(session: AuthSession | null): Promise<void> {
// The keystroke prewarm (grid-reads-without-waking issue 03): typing into a pane whose agent runs on
// a sleeping grid starts that grid while the person types. Here, in the daemon's own input path, so
// an older desktop and typing from a phone get it too; an agent on its own login has no `grid`.
- onInput: (agentId) => {
+ onInput: (agentId, typed) => {
const grid = registry.resolve(agentId)?.grid
if (grid) ports.models?.prewarm(grid)
+ // The person typing into a harness another agent started makes it theirs (services/agentCleanup.ts).
+ if (typed) ports.agentCleanup?.typed(agentId)
},
})
backend.setTerminalStreamManager(terminalStreams)
@@ -1151,6 +1159,8 @@ async function runForeground(session: AuthSession | null): Promise<void> {
emit: (sessionId, events) => emitSessionEvents(sessionId, events),
// Declared further down: read as a prompt is typed, never now.
promptTyped: (session, capture, screen) => { if (session.sessionId) questionWatcher.notePrompt(session.sessionId, capture, screen.question) },
+ // The person sent a harness text from a window or the phone (services/agentCleanup.ts).
+ typed: (id) => ports.agentCleanup?.typed(id),
})
const input = inputs.input
const deviceInput = inputs.deviceInput
@@ -1292,6 +1302,8 @@ async function runForeground(session: AuthSession | null): Promise<void> {
// What the core tells the recaps in their own process, and reads back from what they said (core/recapsLink.ts).
const recapsLink = createRecapsLink({ notify: (frame, opts) => serviceLinks.notify('recaps', frame, opts), buffered: () => serviceLinks.buffered('recaps'),
call: (type, payload) => serviceLinks.call('recaps', type, payload), clients: coreApi.clients, lastTurn: coreApi.transcripts.lastTurn })
+ // What the core tells the agent cleanup in its own process, and answers it (core/agentCleanupLink.ts).
+ const agentCleanupLink = createAgentCleanupLink(coreApi, (frame, opts) => serviceLinks.notify('agentCleanup', frame, opts))
const usageLink = createUsageLink({ call: payload => ports.usage?.read(payload.target as AgentUsageTarget) ?? Promise.resolve({}), changed: agentEvents.onTokenUsageChanged })
const serviceLinks = createServiceLinks({
token: serviceToken,
@@ -1317,6 +1329,7 @@ async function runForeground(session: AuthSession | null): Promise<void> {
: service === 'wifi' ? wifiLink.answer(query, payload)
: service === 'handoff' ? answerConversationQuery(coreApi, query, payload, authority)
: service === 'shell' ? answerShellQuery(coreApi, query, payload)
+ : service === 'agentCleanup' ? agentCleanupLink.answer(query, payload)
: service === 'recaps' ? recapsLink.answer(query, payload) : answerAgentQuery(coreApi, query, payload)),
// The gateway's own traffic: its remote clients and what they sent, and its comings and goings; and what
// the devices tell the core (a turn, a frame for the windows, a dial on the wire); a viewer stream's answers.
@@ -1428,6 +1441,8 @@ async function runForeground(session: AuthSession | null): Promise<void> {
if (!outOfProcess.has('connectors')) serviceHost.serve('connectors', inline!.startConnectors, coreApi, CONNECTORS_REQUESTS)
// The recaps: in this process, or in the edge host (services/recapsProcess.ts), told each turn's lifecycle.
if (outOfProcess.has('recaps')) ports.recaps = recapsLink.port; else serviceHost.start('recaps', inline!.startRecaps, coreApi, RECAPS_FALLBACKS)
+ // Cleaning up after agents: the harnesses a closed harness started, and those idle half an hour (services/agentCleanup.ts).
+ if (outOfProcess.has('agentCleanup')) ports.agentCleanup = agentCleanupLink.port; else serviceHost.start('agentCleanup', inline!.startAgentCleanup, coreApi, AGENT_CLEANUP_FALLBACKS)
if (!outOfProcess.has('shell')) serviceHost.serve('shell', inline!.startShell, coreApi, SHELL_REQUESTS)
// Models: grid access and its pin, the model pictures on agents' frames, the keystroke prewarm, where an
// agent on a grid model sends its inference, and the models requests the apps send (services/models.ts):
@@ -2424,6 +2439,8 @@ async function runForeground(session: AuthSession | null): Promise<void> {
attachDsh,
attachSession: (session) => attachSession(session),
relaunchMarks,
+ // A harness closed or stopped: the harnesses it started close with it (services/agentCleanup.ts).
+ stopped: (agentId) => ports.agentCleanup?.closed(agentId),
})
const stopJobs = lifecycle.stopJobs
launchCancelled = id => stopJobs.has(id)
@@ -2447,6 +2464,7 @@ async function runForeground(session: AuthSession | null): Promise<void> {
stopAgent,
announceSession,
engineReady: (engine, ms) => engineLinks.ready(engine, ms),
+ stoppedAgents,
})
backend.closeAgentService = closing.closeAgentService
backend.closeAgentService.start()
diff --git a/cli/src/engines/__fixtures__/dsh-launch-shapes.golden.json b/cli/src/engines/__fixtures__/dsh-launch-shapes.golden.json
index 71a00d194..4dcd1d162 100644
--- a/cli/src/engines/__fixtures__/dsh-launch-shapes.golden.json
+++ b/cli/src/engines/__fixtures__/dsh-launch-shapes.golden.json
@@ -1,7 +1,7 @@
{
"create": {
- "claude · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w1","claude","--add-dir","<root>/work/w1/input","--append-system-prompt","Read the harness context at \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w1","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w1","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w1"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w1\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w1\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w1/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w1/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w1/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w1\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w1 · created 1 · kept 0"]},
- "codex · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w2","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w2\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e6131bf59bbe1d4c","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w2","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w2","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w2"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w2\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w2\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w2/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w2/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"600","text":"model = \"gpt-5\"\n\n[projects.\"<root>/work/w2\"]\ntrust_level = \"trusted\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w2 · created 1 · kept 0"]},
+ "claude · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w1","claude","--add-dir","<root>/work/w1/input","--append-system-prompt","Read the harness context at \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w1","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w1","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w1"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w1\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w1\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w1/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w1/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w1/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w1\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w1 · created 1 · kept 0"]},
+ "codex · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w2","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w2\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e6131bf59bbe1d4c","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w2","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w2","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w2"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w2\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w2\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w2/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w2/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"600","text":"model = \"gpt-5\"\n\n[projects.\"<root>/work/w2\"]\ntrust_level = \"trusted\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w2 · created 1 · kept 0"]},
"cursor · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w3","cursor-agent"],"script":"29566af379ec1b57","sessionLabel":"harness-cursor-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w3","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/skills","OTHER_SKILLS":"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/skills","HARNESS_CONTEXT_FILE":"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/skills"},"dsh":"acme/draw","dshRuntime":"harness-cursor-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w3","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w3"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: cursor. Workspace: \"<root>/work/w3\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"cursor\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w3\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w3/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w3/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w3 · created 1 · kept 0"]},
"opencode · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w4","opencode"],"script":"202bb0f766c41d0a","sessionLabel":"harness-opencode-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w4","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills","OTHER_SKILLS":"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills","HARNESS_CONTEXT_FILE":"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills","OPENCODE_CONFIG_CONTENT":"{\"permission\":\"ask\",\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"small_model\":\"opencode/muse-spark-1.3-contributor-free\",\"provider\":{\"opencode\":{\"models\":{\"muse-spark-1.3-contributor-free\":{\"options\":{\"reasoningEffort\":\"xhigh\"}}}}},\"agent\":{\"build\":{\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"variant\":\"xhigh\"},\"plan\":{\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"variant\":\"xhigh\"}}}"},"dsh":"acme/draw","dshRuntime":"harness-opencode-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w4","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w4"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: opencode. Workspace: \"<root>/work/w4\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"opencode\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w4\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w4/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w4/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w4 · created 1 · kept 0"]},
"pi · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w5","pi","--append-system-prompt","<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/CONTEXT.md"],"script":"966cf28f1aee3c21","sessionLabel":"harness-pi-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w5","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/skills","OTHER_SKILLS":"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/skills","HARNESS_CONTEXT_FILE":"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/skills"},"dsh":"acme/draw","dshRuntime":"harness-pi-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w5","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w5"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: pi. Workspace: \"<root>/work/w5\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"pi\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w5\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w5/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w5/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w5 · created 1 · kept 0"]},
@@ -14,14 +14,14 @@
"grok · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w12","grok"],"script":"1d5dedc9aa784b58","sessionLabel":"harness-grok-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w12","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/skills","OTHER_SKILLS":"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/skills","HARNESS_CONTEXT_FILE":"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/skills"},"dsh":"acme/draw","dshRuntime":"harness-grok-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w12","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w12"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: grok. Workspace: \"<root>/work/w12\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"grok\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w12\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w12/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w12/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w12 · created 1 · kept 0"]},
"agy · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w13","agy"],"script":"58960d30c79730a9","sessionLabel":"harness-agy-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w13","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/skills","OTHER_SKILLS":"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/skills","HARNESS_CONTEXT_FILE":"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/skills"},"dsh":"acme/draw","dshRuntime":"harness-agy-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w13","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w13"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: agy. Workspace: \"<root>/work/w13\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"agy\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w13\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w13/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w13/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w13 · created 1 · kept 0"]},
"copilot · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w14","copilot"],"script":"9419445032d815b2","sessionLabel":"harness-copilot-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w14","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/skills","OTHER_SKILLS":"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/skills","HARNESS_CONTEXT_FILE":"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/skills"},"dsh":"acme/draw","dshRuntime":"harness-copilot-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w14","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w14"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: copilot. Workspace: \"<root>/work/w14\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"copilot\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w14\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w14/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w14/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w14 · created 1 · kept 0"]},
- "claude · a folder with work in it": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w15","claude","--add-dir","<root>/work/w15/input","--append-system-prompt","Read the harness context at \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w15","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w15","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w15"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w15\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w15\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w15/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w15/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w15/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"644","text":"# My rules\n\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# Mine\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w15 · created 1 · kept 0"]},
- "codex · a folder already laid out": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w16","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w16\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e6131bf59bbe1d4c","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w16","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w16\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w16\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w16/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w16/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"644","text":"# Ours\n\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w16 · created 0 · kept 1"]},
- "codex · a package with nothing but an engine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w17","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/bare\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/bare\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w17\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e6131bf59bbe1d4c","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/bare","HARNESS_DSH_DIR":"<root>/dsh/acme/bare","HARNESS_WORKSPACE":"<root>/work/w17","HARNESS_CONTEXT_FILE":"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/bare","dshRuntime":"harness-codex-golden","label":"Bare","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Bare (acme/bare)\nEngine: codex. Workspace: \"<root>/work/w17\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/bare\",\n \"engine\": \"codex\",\n \"instructions\": \"\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/bare\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/bare\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w17\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/bare materialized <root>/work/w17 · created 0 · kept 0"]},
- "claude · the account's private grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w18","claude","--add-dir","<root>/work/w18/input","--append-system-prompt","Read the harness context at \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"a198cd4214b98e78","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w18","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_PRIVATE_GRID":"grid-of-mine","HARNESS_CONTEXT_FILE":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w18","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w18","HARNESS_PRIVATE_GRID":"grid-of-mine"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w18\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w18\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w18/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w18/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w18/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w18\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w18 · created 1 · kept 0"]},
- "claude · on a grid too": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w19","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--add-dir","<root>/work/w19/input","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"f3a086bb498ac082","sessionLabel":"harness-claude-golden","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w19","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"}},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w19","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w19"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w19\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w19\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w19/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w19/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w19/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w19\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w19 · created 1 · kept 0","[grid] claude -> Home (net-home) · qwen3-coder · web search unavailable"]},
- "codex · the Model Manager": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w20","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"autonomous/autonomous-grid\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/autonomous/autonomous-grid\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w20\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e6131bf59bbe1d4c","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"autonomous/autonomous-grid","HARNESS_DSH_DIR":"<root>/dsh/autonomous/autonomous-grid","HARNESS_WORKSPACE":"<root>/work/w20","HARNESS_CONTEXT_FILE":"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"autonomous/autonomous-grid","dshRuntime":"harness-codex-golden","label":"Grid","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":["ensure ownGrid=true"],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Grid (autonomous/autonomous-grid)\nEngine: codex. Workspace: \"<root>/work/w20\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Grid\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"autonomous/autonomous-grid\",\n \"engine\": \"codex\",\n \"instructions\": \"# Grid\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"autonomous/autonomous-grid\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/autonomous/autonomous-grid\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w20\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] autonomous/autonomous-grid materialized <root>/work/w20 · created 0 · kept 0"]},
- "claude · an init that fails": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w21","claude","--add-dir","<root>/work/w21/input","--append-system-prompt","Read the harness context at \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w21","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w21","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w21"},"timeoutMs":300000}],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w21\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w21\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w21/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w21/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w21/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w21\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialize · init exited 2","[dsh] acme/draw materialized <root>/work/w21 · created 1 · kept 0"]},
- "claude · an init that runs out of time": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w22","claude","--add-dir","<root>/work/w22/input","--append-system-prompt","Read the harness context at \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w22","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w22","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w22"},"timeoutMs":300000}],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w22\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w22\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w22/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w22/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w22/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w22\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialize · init exited by timeout","[dsh] acme/draw materialized <root>/work/w22 · created 1 · kept 0"]},
+ "claude · a folder with work in it": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w15","claude","--add-dir","<root>/work/w15/input","--append-system-prompt","Read the harness context at \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w15","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w15","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w15"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w15\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w15\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w15/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w15/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w15/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"644","text":"# My rules\n\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# Mine\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w15 · created 1 · kept 0"]},
+ "codex · a folder already laid out": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w16","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w16\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e6131bf59bbe1d4c","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w16","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w16\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w16\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w16/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w16/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"644","text":"# Ours\n\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w16 · created 0 · kept 1"]},
+ "codex · a package with nothing but an engine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w17","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/bare\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/bare\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w17\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e6131bf59bbe1d4c","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/bare","HARNESS_DSH_DIR":"<root>/dsh/acme/bare","HARNESS_WORKSPACE":"<root>/work/w17","HARNESS_CONTEXT_FILE":"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/bare","dshRuntime":"harness-codex-golden","label":"Bare","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Bare (acme/bare)\nEngine: codex. Workspace: \"<root>/work/w17\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/bare\",\n \"engine\": \"codex\",\n \"instructions\": \"\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/bare\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/bare\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w17\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/bare materialized <root>/work/w17 · created 0 · kept 0"]},
+ "claude · the account's private grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w18","claude","--add-dir","<root>/work/w18/input","--append-system-prompt","Read the harness context at \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"a198cd4214b98e78","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w18","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_PRIVATE_GRID":"grid-of-mine","HARNESS_CONTEXT_FILE":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w18","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w18","HARNESS_PRIVATE_GRID":"grid-of-mine"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w18\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w18\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w18/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w18/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w18/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w18\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w18 · created 1 · kept 0"]},
+ "claude · on a grid too": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w19","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--add-dir","<root>/work/w19/input","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"f3a086bb498ac082","sessionLabel":"harness-claude-golden","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w19","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"}},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w19","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w19"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w19\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w19\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w19/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w19/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w19/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w19\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w19 · created 1 · kept 0","[grid] claude -> Home (net-home) · qwen3-coder · web search unavailable"]},
+ "codex · the Model Manager": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w20","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"autonomous/autonomous-grid\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/autonomous/autonomous-grid\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w20\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e6131bf59bbe1d4c","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"autonomous/autonomous-grid","HARNESS_DSH_DIR":"<root>/dsh/autonomous/autonomous-grid","HARNESS_WORKSPACE":"<root>/work/w20","HARNESS_CONTEXT_FILE":"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"autonomous/autonomous-grid","dshRuntime":"harness-codex-golden","label":"Grid","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":["ensure ownGrid=true"],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Grid (autonomous/autonomous-grid)\nEngine: codex. Workspace: \"<root>/work/w20\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Grid\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"autonomous/autonomous-grid\",\n \"engine\": \"codex\",\n \"instructions\": \"# Grid\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"autonomous/autonomous-grid\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/autonomous/autonomous-grid\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w20\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] autonomous/autonomous-grid materialized <root>/work/w20 · created 0 · kept 0"]},
+ "claude · an init that fails": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w21","claude","--add-dir","<root>/work/w21/input","--append-system-prompt","Read the harness context at \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w21","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w21","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w21"},"timeoutMs":300000}],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w21\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w21\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w21/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w21/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w21/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w21\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialize · init exited 2","[dsh] acme/draw materialized <root>/work/w21 · created 1 · kept 0"]},
+ "claude · an init that runs out of time": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w22","claude","--add-dir","<root>/work/w22/input","--append-system-prompt","Read the harness context at \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w22","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w22","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w22"},"timeoutMs":300000}],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w22\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w22\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w22/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w22/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w22/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w22\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialize · init exited by timeout","[dsh] acme/draw materialized <root>/work/w22 · created 1 · kept 0"]},
"claude · a skill folder with no skill": {"result":{"ok":false,"error":"DSH_RUNTIME_FAILED","detail":"No SKILL.md under <root>/dsh/acme/broken/nothing-here; repair the harness installation."},"pane":null,"init":[],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{}}}}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/broken materialized <root>/work/w23 · created 0 · kept 0"]},
"claude · a viewer package": {"result":{"ok":false,"error":"INVALID_DSH","detail":"acme/viewer is a viewer package, not an agent"},"pane":null,"init":[],"gridSetup":[],"folder":{},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":[]},
"claude · a package not installed": {"result":{"ok":false,"error":"INVALID_DSH","detail":"acme/missing is not installed on this machine"},"pane":null,"init":[],"gridSetup":[],"folder":{},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":[]},
@@ -29,8 +29,8 @@
"claude · an old tmux": {"result":{"ok":false,"error":"TMUX_TOO_OLD_FOR_DSH","detail":"this machine's tmux is older than 3.2, which is the first version that can give a new session its own environment — so Drawing could not tell claude which harness it is. Upgrade tmux."},"pane":null,"init":[],"gridSetup":[],"folder":{},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[agent] create acme/draw refused · this machine's tmux is older than 3.2, which is the first version that can give a new session its own environment — so Drawing could not tell claude which harness it is. Upgrade tmux."]}
},
"relaunch": {
- "claude · restart a harness agent": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w28","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","OTHER_SKILLS":"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","HARNESS_CONTEXT_FILE":"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills"},"extraArgs":["--add-dir","<root>/work/w28/input","--append-system-prompt","Read the harness context at \"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w28\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w28\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w28/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w28/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w28/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "codex · restart a harness agent": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w29","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills","OTHER_SKILLS":"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills","HARNESS_CONTEXT_FILE":"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills"},"extraArgs":["-c","model_provider=\"openai\"","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w29\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w29\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w29\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w29/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w29/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · restart a harness agent": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w28","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","OTHER_SKILLS":"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","HARNESS_CONTEXT_FILE":"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills"},"extraArgs":["--add-dir","<root>/work/w28/input","--append-system-prompt","Read the harness context at \"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w28\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w28/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w28\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w28/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w28/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w28/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "codex · restart a harness agent": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w29","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills","OTHER_SKILLS":"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills","HARNESS_CONTEXT_FILE":"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills"},"extraArgs":["-c","model_provider=\"openai\"","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w29\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w29\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w29/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w29\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w29/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w29/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"cursor · restart a harness agent": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w30","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w30/.harness/runtime/f561e1f77a40b020cde2f1d7e4d82e588e86e0e081ee7cd8aae96000894e410b/skills","OTHER_SKILLS":"<root>/work/w30/.harness/runtime/f561e1f77a40b020cde2f1d7e4d82e588e86e0e081ee7cd8aae96000894e410b/skills","HARNESS_CONTEXT_FILE":"<root>/work/w30/.harness/runtime/f561e1f77a40b020cde2f1d7e4d82e588e86e0e081ee7cd8aae96000894e410b/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w30/.harness/runtime/f561e1f77a40b020cde2f1d7e4d82e588e86e0e081ee7cd8aae96000894e410b/skills"},"extraArgs":[],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"f561e1f77a40b020cde2f1d7e4d82e588e86e0e081ee7cd8aae96000894e410b":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: cursor. Workspace: \"<root>/work/w30\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w30/.harness/runtime/f561e1f77a40b020cde2f1d7e4d82e588e86e0e081ee7cd8aae96000894e410b/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w30/.harness/runtime/f561e1f77a40b020cde2f1d7e4d82e588e86e0e081ee7cd8aae96000894e410b/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"cursor\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w30\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w30/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w30/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"opencode · restart a harness agent": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w31","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w31/.harness/runtime/4cdc33e93e07a3b3594ad8d55f0bf893e589d3beff3ccf8c3abc8c9b053b5564/skills","OTHER_SKILLS":"<root>/work/w31/.harness/runtime/4cdc33e93e07a3b3594ad8d55f0bf893e589d3beff3ccf8c3abc8c9b053b5564/skills","HARNESS_CONTEXT_FILE":"<root>/work/w31/.harness/runtime/4cdc33e93e07a3b3594ad8d55f0bf893e589d3beff3ccf8c3abc8c9b053b5564/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w31/.harness/runtime/4cdc33e93e07a3b3594ad8d55f0bf893e589d3beff3ccf8c3abc8c9b053b5564/skills"},"extraArgs":[],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"4cdc33e93e07a3b3594ad8d55f0bf893e589d3beff3ccf8c3abc8c9b053b5564":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: opencode. Workspace: \"<root>/work/w31\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w31/.harness/runtime/4cdc33e93e07a3b3594ad8d55f0bf893e589d3beff3ccf8c3abc8c9b053b5564/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w31/.harness/runtime/4cdc33e93e07a3b3594ad8d55f0bf893e589d3beff3ccf8c3abc8c9b053b5564/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"opencode\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w31\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w31/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w31/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"pi · restart a harness agent": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w32","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w32/.harness/runtime/b4d403b95a1e3e79f7dcb27021157c740bb0f4a7249c18c77f742e09e1151964/skills","OTHER_SKILLS":"<root>/work/w32/.harness/runtime/b4d403b95a1e3e79f7dcb27021157c740bb0f4a7249c18c77f742e09e1151964/skills","HARNESS_CONTEXT_FILE":"<root>/work/w32/.harness/runtime/b4d403b95a1e3e79f7dcb27021157c740bb0f4a7249c18c77f742e09e1151964/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w32/.harness/runtime/b4d403b95a1e3e79f7dcb27021157c740bb0f4a7249c18c77f742e09e1151964/skills"},"extraArgs":["--append-system-prompt","<root>/work/w32/.harness/runtime/b4d403b95a1e3e79f7dcb27021157c740bb0f4a7249c18c77f742e09e1151964/CONTEXT.md"],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b4d403b95a1e3e79f7dcb27021157c740bb0f4a7249c18c77f742e09e1151964":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: pi. Workspace: \"<root>/work/w32\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w32/.harness/runtime/b4d403b95a1e3e79f7dcb27021157c740bb0f4a7249c18c77f742e09e1151964/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w32/.harness/runtime/b4d403b95a1e3e79f7dcb27021157c740bb0f4a7249c18c77f742e09e1151964/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"pi\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w32\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w32/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w32/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
@@ -43,33 +43,33 @@
"grok · restart a harness agent": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w39","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w39/.harness/runtime/2fd14a8ac18ac048a9f1cb43300b3182a27d16dd9befc5db093de001d1f6406b/skills","OTHER_SKILLS":"<root>/work/w39/.harness/runtime/2fd14a8ac18ac048a9f1cb43300b3182a27d16dd9befc5db093de001d1f6406b/skills","HARNESS_CONTEXT_FILE":"<root>/work/w39/.harness/runtime/2fd14a8ac18ac048a9f1cb43300b3182a27d16dd9befc5db093de001d1f6406b/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w39/.harness/runtime/2fd14a8ac18ac048a9f1cb43300b3182a27d16dd9befc5db093de001d1f6406b/skills"},"extraArgs":[],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"2fd14a8ac18ac048a9f1cb43300b3182a27d16dd9befc5db093de001d1f6406b":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: grok. Workspace: \"<root>/work/w39\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w39/.harness/runtime/2fd14a8ac18ac048a9f1cb43300b3182a27d16dd9befc5db093de001d1f6406b/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w39/.harness/runtime/2fd14a8ac18ac048a9f1cb43300b3182a27d16dd9befc5db093de001d1f6406b/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"grok\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w39\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w39/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w39/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"agy · restart a harness agent": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w40","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w40/.harness/runtime/6da00afb140bae88243685a25960d49770e082fcef0fc5441ab24307d7a5343a/skills","OTHER_SKILLS":"<root>/work/w40/.harness/runtime/6da00afb140bae88243685a25960d49770e082fcef0fc5441ab24307d7a5343a/skills","HARNESS_CONTEXT_FILE":"<root>/work/w40/.harness/runtime/6da00afb140bae88243685a25960d49770e082fcef0fc5441ab24307d7a5343a/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w40/.harness/runtime/6da00afb140bae88243685a25960d49770e082fcef0fc5441ab24307d7a5343a/skills"},"extraArgs":[],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"6da00afb140bae88243685a25960d49770e082fcef0fc5441ab24307d7a5343a":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: agy. Workspace: \"<root>/work/w40\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w40/.harness/runtime/6da00afb140bae88243685a25960d49770e082fcef0fc5441ab24307d7a5343a/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w40/.harness/runtime/6da00afb140bae88243685a25960d49770e082fcef0fc5441ab24307d7a5343a/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"agy\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w40\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w40/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w40/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"copilot · restart a harness agent": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w41","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w41/.harness/runtime/166641b9365407905d2bcc20c58056b69c94a181a2b58082b2da41c91353644c/skills","OTHER_SKILLS":"<root>/work/w41/.harness/runtime/166641b9365407905d2bcc20c58056b69c94a181a2b58082b2da41c91353644c/skills","HARNESS_CONTEXT_FILE":"<root>/work/w41/.harness/runtime/166641b9365407905d2bcc20c58056b69c94a181a2b58082b2da41c91353644c/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w41/.harness/runtime/166641b9365407905d2bcc20c58056b69c94a181a2b58082b2da41c91353644c/skills"},"extraArgs":[],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"166641b9365407905d2bcc20c58056b69c94a181a2b58082b2da41c91353644c":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: copilot. Workspace: \"<root>/work/w41\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w41/.harness/runtime/166641b9365407905d2bcc20c58056b69c94a181a2b58082b2da41c91353644c/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w41/.harness/runtime/166641b9365407905d2bcc20c58056b69c94a181a2b58082b2da41c91353644c/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"copilot\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w41\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w41/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w41/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "claude · restart after the package changed": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w42","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/skills","OTHER_SKILLS":"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/skills","HARNESS_CONTEXT_FILE":"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/skills"},"extraArgs":["--add-dir","<root>/work/w42/input","--append-system-prompt","Read the harness context at \"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w42\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w42\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w42/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w42/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w42/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "claude · the account's private grid": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w43","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","OTHER_SKILLS":"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","HARNESS_PRIVATE_GRID":"grid-of-mine","HARNESS_CONTEXT_FILE":"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills"},"extraArgs":["--add-dir","<root>/work/w43/input","--append-system-prompt","Read the harness context at \"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":[]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w43\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w43\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w43/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w43/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w43/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "claude · on a grid too": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w44","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","OTHER_SKILLS":"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","HARNESS_CONTEXT_FILE":"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--add-dir","<root>/work/w44/input","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder"},"webSearch":"unavailable"}}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w44\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w44\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w44/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w44/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w44/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "claude · a package that was renamed": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/renamed","HARNESS_DSH_DIR":"<root>/dsh/acme/renamed","HARNESS_WORKSPACE":"<root>/work/w45","HARNESS_CONTEXT_FILE":"<root>/work/w45/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w45/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills"},"extraArgs":["--append-system-prompt","Read the harness context at \"<root>/work/w45/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Renamed (acme/renamed)\nEngine: claude. Workspace: \"<root>/work/w45\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Renamed\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/renamed\",\n \"engine\": \"claude\",\n \"instructions\": \"# Renamed\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/renamed\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/renamed\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w45\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "codex · a package with nothing but an engine": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/bare","HARNESS_DSH_DIR":"<root>/dsh/acme/bare","HARNESS_WORKSPACE":"<root>/work/w46","HARNESS_CONTEXT_FILE":"<root>/work/w46/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w46/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills"},"extraArgs":["-c","model_provider=\"openai\"","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/bare\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/bare\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w46\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w46/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w46/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Bare (acme/bare)\nEngine: codex. Workspace: \"<root>/work/w46\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/bare\",\n \"engine\": \"codex\",\n \"instructions\": \"\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/bare\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/bare\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w46\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · restart after the package changed": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w42","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/skills","OTHER_SKILLS":"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/skills","HARNESS_CONTEXT_FILE":"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/skills"},"extraArgs":["--add-dir","<root>/work/w42/input","--append-system-prompt","Read the harness context at \"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w42\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w42/.harness/runtime/e45e5b1cbfe4a35d3e687d80e33ce802781dce7b1aa4cd4d9d8c6ccdaf86b200/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w42\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w42/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w42/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w42/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · the account's private grid": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w43","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","OTHER_SKILLS":"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","HARNESS_PRIVATE_GRID":"grid-of-mine","HARNESS_CONTEXT_FILE":"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills"},"extraArgs":["--add-dir","<root>/work/w43/input","--append-system-prompt","Read the harness context at \"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":[]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w43\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w43/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w43\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w43/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w43/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w43/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · on a grid too": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w44","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","OTHER_SKILLS":"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","HARNESS_CONTEXT_FILE":"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--add-dir","<root>/work/w44/input","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder"},"webSearch":"unavailable"}}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w44\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w44/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w44\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w44/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w44/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w44/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · a package that was renamed": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/renamed","HARNESS_DSH_DIR":"<root>/dsh/acme/renamed","HARNESS_WORKSPACE":"<root>/work/w45","HARNESS_CONTEXT_FILE":"<root>/work/w45/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w45/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills"},"extraArgs":["--append-system-prompt","Read the harness context at \"<root>/work/w45/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Renamed (acme/renamed)\nEngine: claude. Workspace: \"<root>/work/w45\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Renamed\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/renamed\",\n \"engine\": \"claude\",\n \"instructions\": \"# Renamed\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/renamed\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/renamed\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w45\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "codex · a package with nothing but an engine": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/bare","HARNESS_DSH_DIR":"<root>/dsh/acme/bare","HARNESS_WORKSPACE":"<root>/work/w46","HARNESS_CONTEXT_FILE":"<root>/work/w46/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w46/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills"},"extraArgs":["-c","model_provider=\"openai\"","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/bare\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/bare\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w46\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w46/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w46/.harness/runtime/4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_PRIVATE_GRID"]}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"4c1bdd5c3c799706891c91d495bd149c5f329a8c39ef5d389aa5108e18e4f0bd":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Bare (acme/bare)\nEngine: codex. Workspace: \"<root>/work/w46\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/bare\",\n \"engine\": \"codex\",\n \"instructions\": \"\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/bare\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/bare\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w46\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"claude · a package not installed": {"result":{"ok":false,"error":"DSH_NOT_INSTALLED","detail":"acme/missing is not installed on this machine"},"folder":{},"logged":["[dsh] acme/missing is not installed on this machine · cannot restore its harness context"]},
"claude · no workspace on record": {"result":{"ok":false,"error":"DSH_WORKSPACE_MISSING","detail":"acme/draw has no saved workspace"},"folder":null,"logged":[]},
"claude · a bootstrap edited by hand": {"result":{"ok":false,"error":"DSH_RUNTIME_FAILED","detail":"Error: The Harness bootstrap in <root>/work/w49/CLAUDE.md was edited; restore it before launching."},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w49\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w49/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w49/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w49\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w49/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w49/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w49/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime edited -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"codex · a runtime that belongs to another engine": {"result":{"ok":false,"error":"DSH_RUNTIME_FAILED","detail":"Error: Harness runtime belongs to acme/draw / claude, not acme/draw / codex"},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"a4d26868017c0ccffe2efe50944ef4211834660cca834c6e9f86dec6a88246fa":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w50\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w50/.harness/runtime/a4d26868017c0ccffe2efe50944ef4211834660cca834c6e9f86dec6a88246fa/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w50/.harness/runtime/a4d26868017c0ccffe2efe50944ef4211834660cca834c6e9f86dec6a88246fa/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w50\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w50/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w50/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w50/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"claude · a skill folder with no skill": {"result":{"ok":false,"error":"DSH_RUNTIME_FAILED","detail":"Error: No SKILL.md under <root>/dsh/acme/broken/nothing-here; repair the harness installation."},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{}}}}}}},"logged":[]},
- "claude · retarget onto a grid": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w54","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","OTHER_SKILLS":"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","HARNESS_CONTEXT_FILE":"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--add-dir","<root>/work/w54/input","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder"},"webSearch":"unavailable"}}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w54\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w54\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w54/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w54/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w54/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]}
+ "claude · retarget onto a grid": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w54","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","OTHER_SKILLS":"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills","HARNESS_CONTEXT_FILE":"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--add-dir","<root>/work/w54/input","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder"},"webSearch":"unavailable"}}},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w54\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w54/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w54\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w54/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w54/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w54/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]}
},
"fork": {
- "claude · a fork of a harness agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w55","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.","--add-dir","<root>/work/w55/input","--append-system-prompt","Read the harness context at \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w55","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w55\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w55\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w55/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w55/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w55/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w55\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w55/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w55/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w55\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w55/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w55/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w55/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "codex · a fork of a harness agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w56","codex","fork","s","-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\"","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w56\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\""],"script":"e6131bf59bbe1d4c","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w56","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w56\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w56\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w56/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w56/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w56\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w56/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w56/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w56\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w56/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w56/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · a fork of a harness agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w55","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.","--add-dir","<root>/work/w55/input","--append-system-prompt","Read the harness context at \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w55","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w55\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w55\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w55/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w55/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w55/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w55\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w55/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w55/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w55\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w55/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w55/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w55/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "codex · a fork of a harness agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w56","codex","fork","s","-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\"","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w56\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\""],"script":"e6131bf59bbe1d4c","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w56","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w56\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w56\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w56/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w56/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w56\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w56/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w56/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w56\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w56/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w56/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"opencode · a fork of a harness agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w57","opencode","--prompt","You are a fork of the agent \"w57 · s\" in <root>/work/w57. It kept working; you are its copy, picking up from what it knew. Here is its recent history, oldest first.\n\nContinue from there. Before acting, say in one line what you understand the current state to be."],"script":"27ad120c69130190","sessionLabel":"harness-opencode-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w57","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills","OTHER_SKILLS":"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills","HARNESS_CONTEXT_FILE":"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills"},"dsh":"acme/draw","dshRuntime":"harness-opencode-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: opencode. Workspace: \"<root>/work/w57\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"opencode\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w57\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w57/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w57/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: opencode. Workspace: \"<root>/work/w57\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w57/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w57/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"opencode\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w57\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w57/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w57/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"cursor · a fork of a harness agent": {"result":{"ok":false,"error":"FORK_UNSUPPORTED","detail":"cursor can neither fork a session nor open with a first message, so a fork of w58 · s would start empty."},"pane":null,"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: cursor. Workspace: \"<root>/work/w58\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w58/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w58/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"cursor\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w58\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w58/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w58/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "claude · a source whose runtime is under its agent id": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w59","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.","--add-dir","<root>/work/w59/input","--append-system-prompt","Read the harness context at \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w59","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w59\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w59/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w59/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w59\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w59/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w59/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w59/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w59\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w59\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w59/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w59/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w59/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "claude · a source with no runtime yet": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w60","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.","--add-dir","<root>/work/w60/input","--append-system-prompt","Read the harness context at \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w60","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w60\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w60\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w60/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w60/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w60/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · a source whose runtime is under its agent id": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w59","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.","--add-dir","<root>/work/w59/input","--append-system-prompt","Read the harness context at \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w59","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w59\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w59/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w59/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w59\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w59/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w59/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w59/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w59\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w59\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w59/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w59/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w59/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · a source with no runtime yet": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w60","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.","--add-dir","<root>/work/w60/input","--append-system-prompt","Read the harness context at \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"8d3fee3272f24d93","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w60","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w60\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w60\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w60/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w60/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w60/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"claude · a runtime the row names that is gone": {"result":{"ok":false,"error":"DSH_RUNTIME_FAILED","detail":"The original harness runtime is missing: <root>/work/w61/.harness/runtime/283bb9deef02e6843abfb538efa1eca70801bd8a701c3f98191e123496339247/runtime.json"},"pane":null,"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"283bb9deef02e6843abfb538efa1eca70801bd8a701c3f98191e123496339247":{"dir":{}},"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{}}}}}}},"logged":[]},
- "claude · the account's private grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w62","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.","--add-dir","<root>/work/w62/input","--append-system-prompt","Read the harness context at \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"a198cd4214b98e78","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w62","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_PRIVATE_GRID":"grid-of-mine","HARNESS_CONTEXT_FILE":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w62\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w62/.harness/runtime/ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w62/.harness/runtime/ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w62\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w62/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w62/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w62/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w62\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w62\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w62/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w62/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w62/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · the account's private grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w62","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.","--add-dir","<root>/work/w62/input","--append-system-prompt","Read the harness context at \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"a198cd4214b98e78","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w62","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_PRIVATE_GRID":"grid-of-mine","HARNESS_CONTEXT_FILE":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w62\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w62/.harness/runtime/ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w62/.harness/runtime/ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w62\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w62/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w62/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w62/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w62\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w62\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w62/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w62/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w62/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"claude · a package no longer installed": {"result":{"ok":false,"error":"INVALID_DSH","detail":"acme/missing is no longer installed on this machine"},"pane":null,"folder":{},"logged":[]}
},
"linux": {
"create": {
- "claude · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w1","claude","--add-dir","<root>/work/w1/input","--append-system-prompt","Read the harness context at \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w1","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w1","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w1"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w1\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w1\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w1/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w1/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w1/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w1\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w1 · created 1 · kept 0"]},
- "codex · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w2","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w2\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"2f6ca20200afddb4","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w2","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w2","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w2"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w2\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w2\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w2/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w2/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"600","text":"model = \"gpt-5\"\n\n[projects.\"<root>/work/w2\"]\ntrust_level = \"trusted\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w2 · created 1 · kept 0"]},
+ "claude · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w1","claude","--add-dir","<root>/work/w1/input","--append-system-prompt","Read the harness context at \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w1","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w1","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w1"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w1\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w1/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w1\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w1/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w1/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w1/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w1\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w1 · created 1 · kept 0"]},
+ "codex · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w2","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w2\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"2f6ca20200afddb4","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w2","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w2","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w2"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w2\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w2/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w2\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w2/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w2/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"600","text":"model = \"gpt-5\"\n\n[projects.\"<root>/work/w2\"]\ntrust_level = \"trusted\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w2 · created 1 · kept 0"]},
"cursor · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w3","cursor-agent"],"script":"b17146c64db645b4","sessionLabel":"harness-cursor-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w3","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/skills","OTHER_SKILLS":"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/skills","HARNESS_CONTEXT_FILE":"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/skills"},"dsh":"acme/draw","dshRuntime":"harness-cursor-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w3","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w3"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: cursor. Workspace: \"<root>/work/w3\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w3/.harness/runtime/68b6346e43d9e6f4d05127853ba54c774d8dca9d072d7c66f7b02a8a8a1e904d/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"cursor\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w3\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w3/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w3/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w3 · created 1 · kept 0"]},
"opencode · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w4","opencode"],"script":"12501aaa12cf7372","sessionLabel":"harness-opencode-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w4","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills","OTHER_SKILLS":"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills","HARNESS_CONTEXT_FILE":"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills","OPENCODE_CONFIG_CONTENT":"{\"permission\":\"ask\",\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"small_model\":\"opencode/muse-spark-1.3-contributor-free\",\"provider\":{\"opencode\":{\"models\":{\"muse-spark-1.3-contributor-free\":{\"options\":{\"reasoningEffort\":\"xhigh\"}}}}},\"agent\":{\"build\":{\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"variant\":\"xhigh\"},\"plan\":{\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"variant\":\"xhigh\"}}}"},"dsh":"acme/draw","dshRuntime":"harness-opencode-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w4","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w4"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: opencode. Workspace: \"<root>/work/w4\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w4/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"opencode\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w4\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w4/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w4/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w4 · created 1 · kept 0"]},
"pi · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w5","pi","--append-system-prompt","<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/CONTEXT.md"],"script":"267e89d19341217a","sessionLabel":"harness-pi-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w5","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/skills","OTHER_SKILLS":"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/skills","HARNESS_CONTEXT_FILE":"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/skills"},"dsh":"acme/draw","dshRuntime":"harness-pi-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w5","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w5"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: pi. Workspace: \"<root>/work/w5\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w5/.harness/runtime/b28af9354f73755a5bc9cd8f3c0a9cacb711f0edd27d694dd0c4ea51599e5685/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"pi\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w5\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w5/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w5/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w5 · created 1 · kept 0"]},
@@ -82,25 +82,25 @@
"grok · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w12","grok"],"script":"5c74d58044cb175e","sessionLabel":"harness-grok-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w12","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/skills","OTHER_SKILLS":"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/skills","HARNESS_CONTEXT_FILE":"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/skills"},"dsh":"acme/draw","dshRuntime":"harness-grok-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w12","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w12"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: grok. Workspace: \"<root>/work/w12\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w12/.harness/runtime/a7ce90251623745a2216eb5da83d91a9faf93cbbfca575f50c08d49adc65b30e/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"grok\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w12\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w12/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w12/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w12 · created 1 · kept 0"]},
"agy · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w13","agy"],"script":"2c9edff773ead6d2","sessionLabel":"harness-agy-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w13","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/skills","OTHER_SKILLS":"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/skills","HARNESS_CONTEXT_FILE":"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/skills"},"dsh":"acme/draw","dshRuntime":"harness-agy-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w13","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w13"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: agy. Workspace: \"<root>/work/w13\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w13/.harness/runtime/9ebfddd51f700e618680618ef32159201ca4527adbb4610ccf77e0c02f744780/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"agy\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w13\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w13/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w13/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w13 · created 1 · kept 0"]},
"copilot · a harness in an empty folder": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w14","copilot"],"script":"abc058ae400e1189","sessionLabel":"harness-copilot-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w14","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/skills","OTHER_SKILLS":"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/skills","HARNESS_CONTEXT_FILE":"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/skills"},"dsh":"acme/draw","dshRuntime":"harness-copilot-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w14","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w14"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: copilot. Workspace: \"<root>/work/w14\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w14/.harness/runtime/b03c397db2aa2e1e120d1b3fe9f5054c1c7b3202054adb82dfd90e73d9bb2d58/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"copilot\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w14\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w14/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w14/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w14 · created 1 · kept 0"]},
- "claude · a folder with work in it": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w15","claude","--add-dir","<root>/work/w15/input","--append-system-prompt","Read the harness context at \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w15","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w15","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w15"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w15\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w15\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w15/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w15/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w15/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"644","text":"# My rules\n\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# Mine\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w15 · created 1 · kept 0"]},
- "codex · a folder already laid out": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w16","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w16\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"2f6ca20200afddb4","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w16","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w16\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w16\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w16/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w16/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"644","text":"# Ours\n\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w16 · created 0 · kept 1"]},
- "codex · a package with nothing but an engine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w17","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/bare\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/bare\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w17\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"2f6ca20200afddb4","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/bare","HARNESS_DSH_DIR":"<root>/dsh/acme/bare","HARNESS_WORKSPACE":"<root>/work/w17","HARNESS_CONTEXT_FILE":"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/bare","dshRuntime":"harness-codex-golden","label":"Bare","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Bare (acme/bare)\nEngine: codex. Workspace: \"<root>/work/w17\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/bare\",\n \"engine\": \"codex\",\n \"instructions\": \"\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/bare\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/bare\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w17\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/bare materialized <root>/work/w17 · created 0 · kept 0"]},
- "claude · the account's private grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w18","claude","--add-dir","<root>/work/w18/input","--append-system-prompt","Read the harness context at \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"bd93588e9774003b","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w18","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_PRIVATE_GRID":"grid-of-mine","HARNESS_CONTEXT_FILE":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w18","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w18","HARNESS_PRIVATE_GRID":"grid-of-mine"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w18\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w18\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w18/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w18/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w18/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w18\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w18 · created 1 · kept 0"]},
- "claude · on a grid too": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w19","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--add-dir","<root>/work/w19/input","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"91f18d1a70d63493","sessionLabel":"harness-claude-golden","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w19","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"}},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w19","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w19"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w19\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w19\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w19/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w19/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w19/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w19\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w19 · created 1 · kept 0","[grid] claude -> Home (net-home) · qwen3-coder · web search unavailable"]},
- "codex · the Model Manager": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w20","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"autonomous/autonomous-grid\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/autonomous/autonomous-grid\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w20\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"2f6ca20200afddb4","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"autonomous/autonomous-grid","HARNESS_DSH_DIR":"<root>/dsh/autonomous/autonomous-grid","HARNESS_WORKSPACE":"<root>/work/w20","HARNESS_CONTEXT_FILE":"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"autonomous/autonomous-grid","dshRuntime":"harness-codex-golden","label":"Grid","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":["ensure ownGrid=true"],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Grid (autonomous/autonomous-grid)\nEngine: codex. Workspace: \"<root>/work/w20\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Grid\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"autonomous/autonomous-grid\",\n \"engine\": \"codex\",\n \"instructions\": \"# Grid\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"autonomous/autonomous-grid\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/autonomous/autonomous-grid\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w20\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] autonomous/autonomous-grid materialized <root>/work/w20 · created 0 · kept 0"]},
- "claude · an init that fails": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w21","claude","--add-dir","<root>/work/w21/input","--append-system-prompt","Read the harness context at \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w21","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w21","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w21"},"timeoutMs":300000}],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w21\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w21\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w21/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w21/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w21/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w21\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialize · init exited 2","[dsh] acme/draw materialized <root>/work/w21 · created 1 · kept 0"]},
- "claude · an init that runs out of time": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w22","claude","--add-dir","<root>/work/w22/input","--append-system-prompt","Read the harness context at \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w22","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w22","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w22"},"timeoutMs":300000}],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w22\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w22\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w22/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w22/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w22/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w22\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialize · init exited by timeout","[dsh] acme/draw materialized <root>/work/w22 · created 1 · kept 0"]}
+ "claude · a folder with work in it": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w15","claude","--add-dir","<root>/work/w15/input","--append-system-prompt","Read the harness context at \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w15","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w15","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w15"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w15\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w15/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w15\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w15/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w15/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w15/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"644","text":"# My rules\n\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# Mine\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w15 · created 1 · kept 0"]},
+ "codex · a folder already laid out": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w16","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w16\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"2f6ca20200afddb4","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w16","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w16\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w16/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w16\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w16/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w16/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"644","text":"# Ours\n\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w16 · created 0 · kept 1"]},
+ "codex · a package with nothing but an engine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w17","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/bare\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/bare\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w17\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"2f6ca20200afddb4","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/bare","HARNESS_DSH_DIR":"<root>/dsh/acme/bare","HARNESS_WORKSPACE":"<root>/work/w17","HARNESS_CONTEXT_FILE":"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w17/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/bare","dshRuntime":"harness-codex-golden","label":"Bare","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Bare (acme/bare)\nEngine: codex. Workspace: \"<root>/work/w17\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/bare\",\n \"engine\": \"codex\",\n \"instructions\": \"\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/bare\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/bare\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w17\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/bare materialized <root>/work/w17 · created 0 · kept 0"]},
+ "claude · the account's private grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w18","claude","--add-dir","<root>/work/w18/input","--append-system-prompt","Read the harness context at \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"bd93588e9774003b","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w18","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_PRIVATE_GRID":"grid-of-mine","HARNESS_CONTEXT_FILE":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w18","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w18","HARNESS_PRIVATE_GRID":"grid-of-mine"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w18\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w18/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w18\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w18/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w18/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w18/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w18\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w18 · created 1 · kept 0"]},
+ "claude · on a grid too": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w19","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--add-dir","<root>/work/w19/input","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"91f18d1a70d63493","sessionLabel":"harness-claude-golden","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w19","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"}},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w19","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w19"},"timeoutMs":300000}],"gridSetup":[],"folder":{".draw-ready":{"mode":"644","text":"ready\n"},".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w19\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w19/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w19\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w19/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w19/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w19/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w19\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialized <root>/work/w19 · created 1 · kept 0","[grid] claude -> Home (net-home) · qwen3-coder · web search unavailable"]},
+ "codex · the Model Manager": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w20","codex","-c","shell_environment_policy.set.HARNESS_DSH=\"autonomous/autonomous-grid\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/autonomous/autonomous-grid\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w20\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"2f6ca20200afddb4","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"autonomous/autonomous-grid","HARNESS_DSH_DIR":"<root>/dsh/autonomous/autonomous-grid","HARNESS_WORKSPACE":"<root>/work/w20","HARNESS_CONTEXT_FILE":"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w20/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"autonomous/autonomous-grid","dshRuntime":"harness-codex-golden","label":"Grid","permissionMode":null,"bypassPermission":false,"grid":null},"init":[],"gridSetup":["ensure ownGrid=true"],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Grid (autonomous/autonomous-grid)\nEngine: codex. Workspace: \"<root>/work/w20\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Grid\n\n## Harness skills\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"autonomous/autonomous-grid\",\n \"engine\": \"codex\",\n \"instructions\": \"# Grid\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"autonomous/autonomous-grid\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/autonomous/autonomous-grid\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w20\"\n },\n \"args\": [],\n \"skills\": []\n}\n"},"skills":{"dir":{}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"home":{".claude.json":{"mode":"644","text":"{\n \"projects\": {}\n}\n"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] autonomous/autonomous-grid materialized <root>/work/w20 · created 0 · kept 0"]},
+ "claude · an init that fails": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w21","claude","--add-dir","<root>/work/w21/input","--append-system-prompt","Read the harness context at \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w21","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w21","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w21"},"timeoutMs":300000}],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w21\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w21/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w21\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w21/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w21/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w21/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w21\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialize · init exited 2","[dsh] acme/draw materialized <root>/work/w21 · created 1 · kept 0"]},
+ "claude · an init that runs out of time": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w22","claude","--add-dir","<root>/work/w22/input","--append-system-prompt","Read the harness context at \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w22","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"init":[{"command":"'<root>/dsh/acme/draw/harness/init.sh'","cwd":"<root>/work/w22","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w22"},"timeoutMs":300000}],"gridSetup":[],"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w22\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w22/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w22\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w22/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w22/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w22/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"},"README.md":{"mode":"644","text":"# A drawing\n"},"input":{"dir":{"brief.md":{"mode":"644","text":"Draw a house.\n"}}}},"home":{".claude.json":{"mode":"600","text":"{\n \"projects\": {\n \"<root>/work/w22\": {\n \"allowedTools\": [],\n \"hasTrustDialogAccepted\": true\n }\n }\n}"},".codex":{"dir":{"config.toml":{"mode":"644","text":"model = \"gpt-5\"\n"}}}},"logged":["[dsh] acme/draw materialize · init exited by timeout","[dsh] acme/draw materialized <root>/work/w22 · created 1 · kept 0"]}
},
"relaunch": {
},
"fork": {
- "claude · a fork of a harness agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w55","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.","--add-dir","<root>/work/w55/input","--append-system-prompt","Read the harness context at \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w55","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w55\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w55\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w55/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w55/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w55/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w55\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w55/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w55/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w55\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w55/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w55/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w55/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "codex · a fork of a harness agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w56","codex","fork","s","-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\"","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w56\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\""],"script":"2f6ca20200afddb4","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w56","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w56\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w56\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w56/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w56/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w56\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w56/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w56/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w56\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w56/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w56/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · a fork of a harness agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w55","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.","--add-dir","<root>/work/w55/input","--append-system-prompt","Read the harness context at \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w55","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w55\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w55/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w55\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w55/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w55/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w55/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w55\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w55/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w55/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w55\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w55/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w55/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w55/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "codex · a fork of a harness agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w56","codex","fork","s","-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\"","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/draw\"","-c","shell_environment_policy.set.HARNESS_DSH_DIR=\"<root>/dsh/acme/draw\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"<root>/work/w56\"","-c","shell_environment_policy.set.TOOLCHAIN=\"<root>/dsh/acme/draw/tools\"","-c","shell_environment_policy.set.OLD_SKILLS=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.OTHER_SKILLS=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md\"","-c","shell_environment_policy.set.HARNESS_SKILLS_DIR=\"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills\""],"script":"2f6ca20200afddb4","sessionLabel":"harness-codex-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w56","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","OTHER_SKILLS":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills","HARNESS_CONTEXT_FILE":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills"},"dsh":"acme/draw","dshRuntime":"harness-codex-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w56\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w56/.harness/runtime/9909a78e070b42b52f6b498d1dc4a49224ca3b22c95a50c402f47d9abd99a671/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w56\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w56/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w56/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: codex. Workspace: \"<root>/work/w56\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w56/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w56/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"codex\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w56\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w56/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w56/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
"opencode · a fork of a harness agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w57","opencode","--prompt","You are a fork of the agent \"w57 · s\" in <root>/work/w57. It kept working; you are its copy, picking up from what it knew. Here is its recent history, oldest first.\n\nContinue from there. Before acting, say in one line what you understand the current state to be."],"script":"1bc01f95d698692e","sessionLabel":"harness-opencode-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w57","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills","OTHER_SKILLS":"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills","HARNESS_CONTEXT_FILE":"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills"},"dsh":"acme/draw","dshRuntime":"harness-opencode-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: opencode. Workspace: \"<root>/work/w57\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w57/.harness/runtime/93b54ba496862c3ba2937702c3fb1a4ec6c7f5933e90d881736af21fc4012b93/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"opencode\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w57\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w57/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w57/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: opencode. Workspace: \"<root>/work/w57\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w57/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w57/.harness/runtime/ffa6a744d78d28386438b4a8e8ee32f56718633735057b1b5ddfb5d224d45d98/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"opencode\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w57\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w57/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w57/.agents/skills\"\n },\n \"args\": [],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"AGENTS.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "claude · a source whose runtime is under its agent id": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w59","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.","--add-dir","<root>/work/w59/input","--append-system-prompt","Read the harness context at \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w59","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w59\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w59/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w59/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w59\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w59/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w59/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w59/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w59\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w59\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w59/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w59/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w59/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "claude · a source with no runtime yet": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w60","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.","--add-dir","<root>/work/w60/input","--append-system-prompt","Read the harness context at \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w60","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w60\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w60\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w60/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w60/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w60/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
- "claude · the account's private grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w62","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.","--add-dir","<root>/work/w62/input","--append-system-prompt","Read the harness context at \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"bd93588e9774003b","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w62","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_PRIVATE_GRID":"grid-of-mine","HARNESS_CONTEXT_FILE":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w62\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w62/.harness/runtime/ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w62/.harness/runtime/ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w62\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w62/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w62/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w62/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w62\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w62\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w62/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w62/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w62/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]}
+ "claude · a source whose runtime is under its agent id": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w59","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.","--add-dir","<root>/work/w59/input","--append-system-prompt","Read the harness context at \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w59","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w59\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w59/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w59/.harness/runtime/b53c366c4e4cb65e4639b3c9e7a9d4f56d5b1a764749dc88ba8327d2cebedca3/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w59\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w59/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w59/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w59/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w59\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w59/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w59\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w59/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w59/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w59/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · a source with no runtime yet": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w60","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.","--add-dir","<root>/work/w60/input","--append-system-prompt","Read the harness context at \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"fdcdae9c18146da0","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w60","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_CONTEXT_FILE":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w60\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w60/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w60\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w60/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w60/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w60/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]},
+ "claude · the account's private grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w62","claude","--resume","s","--fork-session","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.","--add-dir","<root>/work/w62/input","--append-system-prompt","Read the harness context at \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md\" before working."],"script":"bd93588e9774003b","sessionLabel":"harness-claude-golden","env":{"HARNESS_DSH":"acme/draw","HARNESS_DSH_DIR":"<root>/dsh/acme/draw","HARNESS_WORKSPACE":"<root>/work/w62","TOOLCHAIN":"<root>/dsh/acme/draw/tools","OLD_SKILLS":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","OTHER_SKILLS":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills","HARNESS_PRIVATE_GRID":"grid-of-mine","HARNESS_CONTEXT_FILE":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/CONTEXT.md","HARNESS_SKILLS_DIR":"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills"},"dsh":"acme/draw","dshRuntime":"harness-claude-golden","label":"Drawing","permissionMode":null,"bypassPermission":false,"grid":null},"folder":{".harness":{"dir":{".gitignore":{"mode":"600","text":"runtime/\nlegacy-AGENTS.md\n"},"runtime":{"dir":{"ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w62\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w62/.harness/runtime/ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w62/.harness/runtime/ada987408d4654e36fc691a5d4383af83cdaf396c088eb6370a5a18235786882/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w62\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w62/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w62/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w62/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}},"ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5":{"dir":{"CONTEXT.md":{"mode":"600","text":"# Drawing (acme/draw)\nEngine: claude. Workspace: \"<root>/work/w62\".\nUse your file and shell tools for these instructions, skills, and toolchain commands.\nLegacy references to .claude/skills or .agents/skills mean HARNESS_SKILLS_DIR for this session.\n\n# Make a drawing\nYou are an agent in this workspace.\nRead `\"$HARNESS_SKILLS_DIR\"/draw/SKILL.md` first.\n\n## Harness skills\n- draw: \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/draw/SKILL.md\"\n- paint: \"<root>/work/w62/.harness/runtime/ff05148322bf53a234e71f1b5fec5dd97865453e32fa8c8be7f58ca6a81493c5/skills/paint/SKILL.md\"\n"},"runtime.json":{"mode":"600","text":"{\n \"version\": 1,\n \"id\": \"acme/draw\",\n \"engine\": \"claude\",\n \"instructions\": \"# Make a drawing\\nYou are Claude Code in this workspace.\\nRead `.claude/skills/draw/SKILL.md` first.\\n\",\n \"env\": {\n \"HARNESS_DSH\": \"acme/draw\",\n \"HARNESS_DSH_DIR\": \"<root>/dsh/acme/draw\",\n \"HARNESS_WORKSPACE\": \"<root>/work/w62\",\n \"TOOLCHAIN\": \"<root>/dsh/acme/draw/tools\",\n \"OLD_SKILLS\": \"<root>/work/w62/.claude/skills\",\n \"OTHER_SKILLS\": \"<root>/work/w62/.agents/skills\"\n },\n \"args\": [\n \"--add-dir\",\n \"<root>/work/w62/input\"\n ],\n \"skills\": [\n {\n \"name\": \"draw\",\n \"dir\": \"<root>/dsh/acme/draw/skills/draw\"\n },\n {\n \"name\": \"paint\",\n \"dir\": \"<root>/dsh/acme/draw/skills/paint\"\n }\n ]\n}\n"},"skills":{"dir":{"draw":{"link":"<root>/dsh/acme/draw/skills/draw"},"paint":{"link":"<root>/dsh/acme/draw/skills/paint"}}}}}}}}},"CLAUDE.md":{"mode":"600","text":"\n<!-- harness:runtime v1 -->\n## Harness session context\n\nIf HARNESS_CONTEXT_FILE is set in your environment, read that file before working:\n\n```sh\nif [ -n \"$HARNESS_CONTEXT_FILE\" ]; then cat \"$HARNESS_CONTEXT_FILE\"; fi\n```\n\nIt contains the instructions and skill index for this session's selected harness.\nIf the variable is absent, this is an ordinary coding session with no selected harness.\n<!-- /harness:runtime -->\n"}},"logged":[]}
}
},
"scripts": {
diff --git a/cli/src/engines/__fixtures__/launch-argv.golden.json b/cli/src/engines/__fixtures__/launch-argv.golden.json
index 29d36bf42..8b6ed9027 100644
--- a/cli/src/engines/__fixtures__/launch-argv.golden.json
+++ b/cli/src/engines/__fixtures__/launch-argv.golden.json
@@ -1070,22 +1070,22 @@
"terminal · new tile · none": {"argv":["/bin/sh","-c","<script 09f6eade56ba2b9e>","harness-terminal","/work/my project's folder","/bin/sh"],"script":"09f6eade56ba2b9e","command":[""]}
},
"overrides": {
- "claude · own login": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
- "claude · own login, its model back": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_MODEL":"opus"},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
- "claude · grid": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/grid-abc/relay","ANTHROPIC_AUTH_TOKEN":"gridkey-abc123","ANTHROPIC_MODEL":"gpt-5"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
- "claude · grid with web tools and a window": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/grid-abc/relay","ANTHROPIC_AUTH_TOKEN":"gridkey-abc123","ANTHROPIC_MODEL":"gpt-5","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"gridkey-abc123"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"calls":[]},
- "claude · saved API": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://openrouter.ai/api","ANTHROPIC_AUTH_TOKEN":"sk-or-1","ANTHROPIC_MODEL":"qwen/qwen3-coder"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"api:openrouter","networkName":"OpenRouter","baseUrl":"https://openrouter.ai/api/v1","apiKey":"sk-or-1","model":"qwen/qwen3-coder"},"webSearch":"unavailable"}}},"calls":[]},
- "claude · harness": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/harness-runtime-1/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["--studio","--append-system-prompt","Read the harness context at \"/work/studio/.harness/runtime/harness-runtime-1/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["dshLaunch acme/studio /work/studio claude harness-runtime-1"]},
+ "claude · own login": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
+ "claude · own login, its model back": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_MODEL":"opus"},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
+ "claude · grid": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/grid-abc/relay","ANTHROPIC_AUTH_TOKEN":"gridkey-abc123","ANTHROPIC_MODEL":"gpt-5"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
+ "claude · grid with web tools and a window": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/grid-abc/relay","ANTHROPIC_AUTH_TOKEN":"gridkey-abc123","ANTHROPIC_MODEL":"gpt-5","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"gridkey-abc123"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"calls":[]},
+ "claude · saved API": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://openrouter.ai/api","ANTHROPIC_AUTH_TOKEN":"sk-or-1","ANTHROPIC_MODEL":"qwen/qwen3-coder"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"api:openrouter","networkName":"OpenRouter","baseUrl":"https://openrouter.ai/api/v1","apiKey":"sk-or-1","model":"qwen/qwen3-coder"},"webSearch":"unavailable"}}},"calls":[]},
+ "claude · harness": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/harness-runtime-1/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["--studio","--append-system-prompt","Read the harness context at \"/work/studio/.harness/runtime/harness-runtime-1/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["dshLaunch acme/studio /work/studio claude harness-runtime-1"]},
"claude · harness without a workspace": {"result":{"ok":false,"error":"DSH_WORKSPACE_MISSING","detail":"acme/studio has no saved workspace"},"calls":[]},
- "claude · harness on a grid, as git, as a named agent": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/grid-abc/relay","ANTHROPIC_AUTH_TOKEN":"gridkey-abc123","ANTHROPIC_MODEL":"gpt-5","HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/agent-0001/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--studio","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"/work/studio/.harness/runtime/agent-0001/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":["dshLaunch acme/studio /work/studio claude agent-0001"]},
- "codex · own login": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · own login, its model back": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
- "codex · grid with web tools and a window": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123","GRID_MCP_AUTHORIZATION":"Bearer gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","gpt-5","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"calls":[]},
- "codex · saved API": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"sk-or-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://openrouter.ai/api/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","qwen/qwen3-coder","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"api:openrouter","networkName":"OpenRouter","baseUrl":"https://openrouter.ai/api/v1","apiKey":"sk-or-1","model":"qwen/qwen3-coder"},"webSearch":"unavailable"}}},"calls":[]},
- "codex · harness": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/harness-runtime-1/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["-c","model_provider=\"openai\"","--studio","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/studio\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"/work/studio\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"/work/studio/.harness/runtime/harness-runtime-1/CONTEXT.md\"","-c","shell_environment_policy.set.STUDIO_MODE=\"pro \\\"x\\\"\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml","dshLaunch acme/studio /work/studio codex harness-runtime-1"]},
+ "claude · harness on a grid, as git, as a named agent": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/grid-abc/relay","ANTHROPIC_AUTH_TOKEN":"gridkey-abc123","ANTHROPIC_MODEL":"gpt-5","HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/agent-0001/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--studio","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"/work/studio/.harness/runtime/agent-0001/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":["dshLaunch acme/studio /work/studio claude agent-0001"]},
+ "codex · own login": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · own login, its model back": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
+ "codex · grid with web tools and a window": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123","GRID_MCP_AUTHORIZATION":"Bearer gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","gpt-5","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"calls":[]},
+ "codex · saved API": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"sk-or-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://openrouter.ai/api/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","qwen/qwen3-coder","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"api:openrouter","networkName":"OpenRouter","baseUrl":"https://openrouter.ai/api/v1","apiKey":"sk-or-1","model":"qwen/qwen3-coder"},"webSearch":"unavailable"}}},"calls":[]},
+ "codex · harness": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/harness-runtime-1/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["-c","model_provider=\"openai\"","--studio","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/studio\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"/work/studio\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"/work/studio/.harness/runtime/harness-runtime-1/CONTEXT.md\"","-c","shell_environment_policy.set.STUDIO_MODE=\"pro \\\"x\\\"\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml","dshLaunch acme/studio /work/studio codex harness-runtime-1"]},
"codex · harness without a workspace": {"result":{"ok":false,"error":"DSH_WORKSPACE_MISSING","detail":"acme/studio has no saved workspace"},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · harness on a grid, as git, as a named agent": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123","HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/agent-0001/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","--studio","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/studio\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"/work/studio\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"/work/studio/.harness/runtime/agent-0001/CONTEXT.md\"","-c","shell_environment_policy.set.STUDIO_MODE=\"pro \\\"x\\\"\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":["dshLaunch acme/studio /work/studio codex agent-0001"]},
+ "codex · harness on a grid, as git, as a named agent": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123","HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/agent-0001/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","--studio","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/studio\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"/work/studio\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"/work/studio/.harness/runtime/agent-0001/CONTEXT.md\"","-c","shell_environment_policy.set.STUDIO_MODE=\"pro \\\"x\\\"\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":["dshLaunch acme/studio /work/studio codex agent-0001"]},
"opencode · own login": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":[],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
"opencode · own login, its model back": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":[],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
"opencode · grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123","OPENCODE_CONFIG":"/state/grid-engine-config/agent-0001/opencode.json"},"extraArgs":["-m","team-grid/gpt-5"],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"},"sessionModel":"team-grid/gpt-5"}},"calls":["writeGridConfigDir agent-0001 opencode.json"]},
@@ -1102,56 +1102,56 @@
"pi · harness": {"result":{"ok":true,"overrides":{"env":{"HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/harness-runtime-1/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["--studio","--append-system-prompt","/work/studio/.harness/runtime/harness-runtime-1/CONTEXT.md"],"clearEnv":["HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["dshLaunch acme/studio /work/studio pi harness-runtime-1"]},
"pi · harness without a workspace": {"result":{"ok":false,"error":"DSH_WORKSPACE_MISSING","detail":"acme/studio has no saved workspace"},"calls":[]},
"pi · harness on a grid, as git, as a named agent": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123","PI_CODING_AGENT_DIR":"/state/grid-engine-config/agent-0001","HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/agent-0001/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["--model","grid/gpt-5","--studio","--append-system-prompt","/work/studio/.harness/runtime/agent-0001/CONTEXT.md"],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unsupported"}}},"calls":["writeGridConfigDir agent-0001 models.json,settings.json","dshLaunch acme/studio /work/studio pi agent-0001"]},
- "codex · own login · no config": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · profile · no config": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile, its model back · no config": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile on a grid · no config": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
- "codex · own login · its own provider": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"azure\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · profile · its own provider": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"azure\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile, its model back · its own provider": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"azure\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile on a grid · its own provider": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
- "codex · own login · single quotes and a comment": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"local\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · profile · single quotes and a comment": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"local\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile, its model back · single quotes and a comment": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"local\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile on a grid · single quotes and a comment": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
- "codex · own login · only under a profile": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · profile · only under a profile": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile, its model back · only under a profile": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile on a grid · only under a profile": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
- "codex · own login · commented out": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · profile · commented out": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile, its model back · commented out": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile on a grid · commented out": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
- "codex · own login · empty value": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · profile · empty value": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile, its model back · empty value": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile on a grid · empty value": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
- "codex · own login · a quote in single quotes": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"a\"b\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · profile · a quote in single quotes": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"a\"b\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile, its model back · a quote in single quotes": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"a\"b\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · profile on a grid · a quote in single quotes": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
- "codex · own login · moved home": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"moved\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /moved/codex/config.toml"]},
- "codex · profile · moved home": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · own login · moved home, relative": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"moved\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · profile · moved home, relative": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · own login · moved home, tilde": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"moved\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · profile · moved home, tilde": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · own login · moved home, trailing slash": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"moved\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /moved/codex/config.toml"]},
- "codex · profile · moved home, trailing slash": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "codex · own login · the daemon home's file": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"from-file\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
- "codex · profile · the profile's file": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"<root>/profile-home"},"extraArgs":["-c","model_provider=\"profile-file\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["installCodexHooks <root>/profile-home"]},
- "codex · profile without a file": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"<root>/no-such-profile"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["installCodexHooks <root>/no-such-profile"]},
- "claude · started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
- "claude · started by an agent, a harness on a grid as a named agent": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/grid-abc/relay","ANTHROPIC_AUTH_TOKEN":"gridkey-abc123","ANTHROPIC_MODEL":"gpt-5","HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/agent-0001/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--studio","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"/work/studio/.harness/runtime/agent-0001/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":["dshLaunch acme/studio /work/studio claude agent-0001"]},
- "codex · started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
- "codex · started by an agent, a harness on a grid as a named agent": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123","HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/agent-0001/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","--studio","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/studio\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"/work/studio\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"/work/studio/.harness/runtime/agent-0001/CONTEXT.md\"","-c","shell_environment_policy.set.STUDIO_MODE=\"pro \\\"x\\\"\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":["dshLaunch acme/studio /work/studio codex agent-0001"]},
+ "codex · own login · no config": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · profile · no config": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile, its model back · no config": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile on a grid · no config": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
+ "codex · own login · its own provider": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"azure\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · profile · its own provider": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"azure\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile, its model back · its own provider": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"azure\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile on a grid · its own provider": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
+ "codex · own login · single quotes and a comment": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"local\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · profile · single quotes and a comment": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"local\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile, its model back · single quotes and a comment": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"local\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile on a grid · single quotes and a comment": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
+ "codex · own login · only under a profile": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · profile · only under a profile": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile, its model back · only under a profile": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile on a grid · only under a profile": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
+ "codex · own login · commented out": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · profile · commented out": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile, its model back · commented out": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile on a grid · commented out": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
+ "codex · own login · empty value": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · profile · empty value": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile, its model back · empty value": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile on a grid · empty value": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
+ "codex · own login · a quote in single quotes": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"a\"b\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · profile · a quote in single quotes": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"a\"b\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile, its model back · a quote in single quotes": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"a\"b\"","-m","gpt-5.5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · profile on a grid · a quote in single quotes": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":[]},
+ "codex · own login · moved home": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"moved\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /moved/codex/config.toml"]},
+ "codex · profile · moved home": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · own login · moved home, relative": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"moved\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · profile · moved home, relative": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · own login · moved home, tilde": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"moved\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · profile · moved home, tilde": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · own login · moved home, trailing slash": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"moved\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /moved/codex/config.toml"]},
+ "codex · profile · moved home, trailing slash": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
+ "codex · own login · the daemon home's file": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"from-file\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
+ "codex · profile · the profile's file": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"<root>/profile-home"},"extraArgs":["-c","model_provider=\"profile-file\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["installCodexHooks <root>/profile-home"]},
+ "codex · profile without a file": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"<root>/no-such-profile"},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["installCodexHooks <root>/no-such-profile"]},
+ "claude · started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
+ "claude · started by an agent, a harness on a grid as a named agent": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/grid-abc/relay","ANTHROPIC_AUTH_TOKEN":"gridkey-abc123","ANTHROPIC_MODEL":"gpt-5","HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/agent-0001/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--studio","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nRead the harness context at \"/work/studio/.harness/runtime/agent-0001/CONTEXT.md\" before working.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":["dshLaunch acme/studio /work/studio claude agent-0001"]},
+ "codex · started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "codex · started by an agent, a harness on a grid as a named agent": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123","HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/agent-0001/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/grid-abc/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","gpt-5","--studio","-c","shell_environment_policy.set.HARNESS_DSH=\"acme/studio\"","-c","shell_environment_policy.set.HARNESS_WORKSPACE=\"/work/studio\"","-c","shell_environment_policy.set.HARNESS_CONTEXT_FILE=\"/work/studio/.harness/runtime/agent-0001/CONTEXT.md\"","-c","shell_environment_policy.set.STUDIO_MODE=\"pro \\\"x\\\"\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"}}},"calls":["dshLaunch acme/studio /work/studio codex agent-0001"]},
"opencode · started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":[],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
"opencode · started by an agent, a harness on a grid as a named agent": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"gridkey-abc123","OPENCODE_CONFIG":"/state/grid-engine-config/agent-0001/opencode.json","HARNESS_DSH":"acme/studio","HARNESS_WORKSPACE":"/work/studio","HARNESS_CONTEXT_FILE":"/work/studio/.harness/runtime/agent-0001/CONTEXT.md","STUDIO_MODE":"pro \"x\""},"extraArgs":["-m","team-grid/gpt-5","--studio","--agent","harness-compute"],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH_DIR","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"grid-abc","networkName":"Team grid","baseUrl":"https://grid.example/grid-abc/relay/v1","apiKey":"gridkey-abc123","model":"gpt-5"},"webSearch":"unavailable"},"sessionModel":"team-grid/gpt-5"}},"calls":["dshLaunch acme/studio /work/studio opencode agent-0001","writeGridConfigDir agent-0001 opencode.json"]},
- "claude · started by an agent, the folder's own pull request text": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--settings","{\"attribution\":{\"pr\":\"Made in the app\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
- "claude · started by an agent, attribution hidden": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
- "codex · started by an agent, the person's own instructions": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"Be terse.\\n\\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
+ "claude · started by an agent, the folder's own pull request text": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--settings","{\"attribution\":{\"pr\":\"Made in the app\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
+ "claude · started by an agent, attribution hidden": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]},
+ "codex · started by an agent, the person's own instructions": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"Be terse.\\n\\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig <home>/.codex/config.toml"]},
"codex · profile, started by an agent, instructions it cannot keep exactly": {"result":{"ok":true,"overrides":{"env":{"CODEX_HOME":"/profiles/work"},"extraArgs":["-c","model_provider=\"openai\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":["readCodexConfig /profiles/work/config.toml","installCodexHooks /profiles/work"]},
- "claude · not started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]}
+ "claude · not started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"calls":[]}
},
"adapters": {
"claude": {"instructionFiles":["CLAUDE.md"],"contextArgs":["--append-system-prompt","Read the harness context at \"/work/my \\\"studio\\\"/.harness/runtime/agent-1/CONTEXT.md\" before working."],"envArgs":null,"envArgsEmpty":null},
diff --git a/cli/src/engines/__fixtures__/launch-shapes.golden.json b/cli/src/engines/__fixtures__/launch-shapes.golden.json
index f4de7868c..19bb0c877 100644
--- a/cli/src/engines/__fixtures__/launch-shapes.golden.json
+++ b/cli/src/engines/__fixtures__/launch-shapes.golden.json
@@ -1,7 +1,7 @@
{
"create": {
- "claude · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w1","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"0fd89f463aa4e0e9","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]},
- "codex · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w2","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e121d4648505dbf7","env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · qwen3-coder · web search on"]},
+ "claude · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w1","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"0fd89f463aa4e0e9","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]},
+ "codex · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w2","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e121d4648505dbf7","env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · qwen3-coder · web search on"]},
"cursor · grid": {"result":{"ok":false,"error":"GRID_ENGINE_UNSUPPORTED","detail":"cursor cannot run on grid Home: Cursor Agent can only be re-pointed at another Cursor API (CURSOR_API_ENDPOINT), not at an OpenAI-compatible relay. It would have run on its own login instead. Engines that can: claude, codex, opencode, hermes, grok, pi, copilot."},"pane":null,"config":null,"folder":{},"logged":["[agent] create cursor refused · cursor cannot run on grid Home: Cursor Agent can only be re-pointed at another Cursor API (CURSOR_API_ENDPOINT), not at an OpenAI-compatible relay. It would have run on its own login instead. Engines that can: claude, codex, opencode, hermes, grok, pi, copilot."]},
"opencode · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w4","opencode","--standalone"],"script":"acd3863e35e5c432","env":{"GRID_API_KEY":"grid-key-1","OPENCODE_CONFIG":"<root>/data/grid-engine-config/<label>/opencode.json"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":{"<label>":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"model\": \"home/qwen3-coder\",\n \"provider\": {\n \"home\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Home\",\n \"options\": {\n \"baseURL\": \"https://grid.example/g/home/relay/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"qwen3-coder\": {\n \"name\": \"qwen3-coder\",\n \"limit\": {\n \"context\": 131072,\n \"output\": 32000\n }\n }\n }\n }\n },\n \"mcp\": {\n \"harness\": {\n \"type\": \"remote\",\n \"url\": \"https://control.example/v1/grid/web-mcp/\",\n \"enabled\": true,\n \"headers\": {\n \"Authorization\": \"Bearer {env:GRID_API_KEY}\"\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] opencode -> Home (net-home) · qwen3-coder · web search on"]},
"pi · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w5","pi","--model","grid/qwen3-coder"],"script":"8bb2d44e0c77d3b8","env":{"GRID_API_KEY":"grid-key-1","PI_CODING_AGENT_DIR":"<root>/data/grid-engine-config/<label>"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"unsupported"}},"config":{"<label>":{"dir":{"models.json":{"mode":"600","text":"{\n \"providers\": {\n \"grid\": {\n \"name\": \"Autonomous Grid\",\n \"baseUrl\": \"https://grid.example/g/home/relay/v1\",\n \"api\": \"openai-completions\",\n \"apiKey\": \"$GRID_API_KEY\",\n \"models\": [\n {\n \"id\": \"qwen3-coder\",\n \"name\": \"qwen3-coder\",\n \"reasoning\": false,\n \"input\": [\n \"text\"\n ],\n \"contextWindow\": 131072,\n \"maxTokens\": 8192,\n \"cost\": {\n \"input\": 0,\n \"output\": 0,\n \"cacheRead\": 0,\n \"cacheWrite\": 0\n }\n }\n ]\n }\n }\n}"},"settings.json":{"mode":"600","text":"{\n \"skills\": [\n \"<root>/home/.pi/agent/skills\"\n ]\n}"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] pi -> Home (net-home) · qwen3-coder · web search unsupported (Pi has no MCP client)"]},
@@ -15,12 +15,12 @@
"agy · grid": {"result":{"ok":false,"error":"GRID_ENGINE_UNSUPPORTED","detail":"agy cannot run on grid Home: Antigravity speaks the Gemini API, which this relay does not serve. It would have run on its own login instead. Engines that can: claude, codex, opencode, hermes, grok, pi, copilot."},"pane":null,"config":null,"folder":{},"logged":["[agent] create agy refused · agy cannot run on grid Home: Antigravity speaks the Gemini API, which this relay does not serve. It would have run on its own login instead. Engines that can: claude, codex, opencode, hermes, grok, pi, copilot."]},
"copilot · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w14","copilot","--additional-mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}"],"script":"0e97ffe705e9905c","env":{"COPILOT_PROVIDER_BASE_URL":"https://grid.example/g/home/relay/v1","COPILOT_PROVIDER_API_KEY":"grid-key-1","COPILOT_MODEL":"qwen3-coder","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] copilot -> Home (net-home) · qwen3-coder · web search on"]},
"terminal · grid": {"result":{"ok":false,"error":"GRID_ENGINE_UNSUPPORTED","detail":"terminal cannot run on grid Home: a terminal runs no engine to point at a grid — start one inside it and it will use its own login. It would have run on its own login instead. Engines that can: claude, codex, opencode, hermes, grok, pi, copilot."},"pane":null,"config":null,"folder":{},"logged":["[agent] create terminal refused · terminal cannot run on grid Home: a terminal runs no engine to point at a grid — start one inside it and it will use its own login. It would have run on its own login instead. Engines that can: claude, codex, opencode, hermes, grok, pi, copilot."]},
- "claude · grid without a model, web tools or window": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w16","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"bed4505fbce71ce3","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-2"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":null},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · model chosen by the engine · web search unavailable"]},
- "claude · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w17","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"0fd89f463aa4e0e9","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]},
- "claude · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w18","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"b444c6de9ed73b6b","env":{"ANTHROPIC_BASE_URL":"http://127.0.0.1:<port>","ANTHROPIC_AUTH_TOKEN":"stale-key","ANTHROPIC_MODEL":"vendor/model-a","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"200000"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Router (api:router) · vendor/model-a · web search unavailable"]},
- "codex · grid without a model, web tools or window": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w19","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"8f855b51ab6b35e6","env":{"GRID_API_KEY":"grid-key-2"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":null},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · model chosen by the engine · web search unavailable"]},
- "codex · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w20","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e121d4648505dbf7","env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · qwen3-coder · web search on"]},
- "codex · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w21","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"http://127.0.0.1:<port>/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","vendor/model-a","-c","model_context_window=200000","-c","model_auto_compact_token_limit=180000","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"8f855b51ab6b35e6","env":{"GRID_API_KEY":"stale-key"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Router (api:router) · vendor/model-a · web search unavailable"]},
+ "claude · grid without a model, web tools or window": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w16","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"bed4505fbce71ce3","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-2"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":null},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · model chosen by the engine · web search unavailable"]},
+ "claude · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w17","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"0fd89f463aa4e0e9","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]},
+ "claude · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w18","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"b444c6de9ed73b6b","env":{"ANTHROPIC_BASE_URL":"http://127.0.0.1:<port>","ANTHROPIC_AUTH_TOKEN":"stale-key","ANTHROPIC_MODEL":"vendor/model-a","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"200000"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Router (api:router) · vendor/model-a · web search unavailable"]},
+ "codex · grid without a model, web tools or window": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w19","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"8f855b51ab6b35e6","env":{"GRID_API_KEY":"grid-key-2"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":null},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · model chosen by the engine · web search unavailable"]},
+ "codex · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w20","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"e121d4648505dbf7","env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · qwen3-coder · web search on"]},
+ "codex · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w21","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"http://127.0.0.1:<port>/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","vendor/model-a","-c","model_context_window=200000","-c","model_auto_compact_token_limit=180000","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"8f855b51ab6b35e6","env":{"GRID_API_KEY":"stale-key"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Router (api:router) · vendor/model-a · web search unavailable"]},
"opencode · grid without a model, web tools or window": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w22","opencode","--standalone"],"script":"acd3863e35e5c432","env":{"GRID_API_KEY":"grid-key-2","OPENCODE_CONFIG":"<root>/data/grid-engine-config/<label>/opencode.json"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":null},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}},"config":{"<label>":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"model\": \"home/Auto\",\n \"provider\": {\n \"home\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Home\",\n \"options\": {\n \"baseURL\": \"https://grid.example/g/home/relay/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"Auto\": {\n \"name\": \"Auto\"\n }\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] opencode -> Home (net-home) · model chosen by the engine · web search unavailable"]},
"opencode · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w23","opencode","-m","home/qwen3-coder"],"script":"acd3863e35e5c432","env":{"GRID_API_KEY":"grid-key-1","OPENCODE_CONFIG":"<root>/data/grid-engine-config/<label>/opencode.json"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":{"<label>":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"provider\": {\n \"home\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Home\",\n \"options\": {\n \"baseURL\": \"https://grid.example/g/home/relay/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"qwen3-coder\": {\n \"name\": \"qwen3-coder\",\n \"limit\": {\n \"context\": 131072,\n \"output\": 32000\n }\n }\n }\n }\n },\n \"mcp\": {\n \"harness\": {\n \"type\": \"remote\",\n \"url\": \"https://control.example/v1/grid/web-mcp/\",\n \"enabled\": true,\n \"headers\": {\n \"Authorization\": \"Bearer {env:GRID_API_KEY}\"\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] opencode -> Home (net-home) · qwen3-coder · web search on"]},
"opencode · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w24","opencode","--standalone"],"script":"acd3863e35e5c432","env":{"GRID_API_KEY":"stale-key","OPENCODE_CONFIG":"<root>/data/grid-engine-config/<label>/opencode.json"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":{"<label>":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"model\": \"api-router/vendor/model-a\",\n \"provider\": {\n \"api-router\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Router\",\n \"options\": {\n \"baseURL\": \"http://127.0.0.1:<port>/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"vendor/model-a\": {\n \"name\": \"vendor/model-a\",\n \"limit\": {\n \"context\": 200000,\n \"output\": 32000\n }\n }\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] opencode -> Router (api:router) · vendor/model-a · web search unavailable"]},
@@ -37,16 +37,16 @@
"copilot · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w35","copilot","--additional-mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}"],"script":"0e97ffe705e9905c","env":{"COPILOT_PROVIDER_BASE_URL":"https://grid.example/g/home/relay/v1","COPILOT_PROVIDER_API_KEY":"grid-key-1","COPILOT_MODEL":"qwen3-coder","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] copilot -> Home (net-home) · qwen3-coder · web search on"]},
"copilot · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w36","copilot"],"script":"cf114508370b186a","env":{"COPILOT_PROVIDER_BASE_URL":"http://127.0.0.1:<port>/v1","COPILOT_PROVIDER_API_KEY":"stale-key","COPILOT_MODEL":"vendor/model-a"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] copilot -> Router (api:router) · vendor/model-a · web search unavailable"]},
"claude · grid · an old tmux": {"result":{"ok":false,"error":"TMUX_TOO_OLD_FOR_GRID","detail":"this machine's tmux is older than 3.2, which is the first version that can give a new session its own environment — so claude could not have been pointed at grid Home. Upgrade tmux, or create this agent without a grid selected."},"pane":null,"config":null,"folder":{},"logged":["[agent] create claude refused · this machine's tmux is older than 3.2, which is the first version that can give a new session its own environment — so claude could not have been pointed at grid Home. Upgrade tmux, or create this agent without a grid selected."]},
- "claude · own login": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w38","claude","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"d3f8b8a35d872741","env":{},"grid":null,"gridLaunchRecord":null},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · own login": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w39","codex","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"51edaf2d9021f9c1","env":{},"grid":null,"gridLaunchRecord":null},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · own login · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w40","claude","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"script":"d3f8b8a35d872741","env":{},"grid":null,"gridLaunchRecord":null,"startedBy":"agent"},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · own login · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w41","codex","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"script":"51edaf2d9021f9c1","env":{},"grid":null,"gridLaunchRecord":null,"startedBy":"agent"},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · own login": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w38","claude","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"d3f8b8a35d872741","env":{},"grid":null,"gridLaunchRecord":null},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · own login": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w39","codex","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"51edaf2d9021f9c1","env":{},"grid":null,"gridLaunchRecord":null},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · own login · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w40","claude","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"script":"d3f8b8a35d872741","env":{},"grid":null,"gridLaunchRecord":null,"startedBy":"agent"},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · own login · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w41","codex","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"script":"51edaf2d9021f9c1","env":{},"grid":null,"gridLaunchRecord":null,"startedBy":"agent"},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"opencode · own login · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w42","opencode"],"script":"7ff052bd4280633e","env":{"OPENCODE_CONFIG_CONTENT":"{\"permission\":\"ask\",\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"small_model\":\"opencode/muse-spark-1.3-contributor-free\",\"provider\":{\"opencode\":{\"models\":{\"muse-spark-1.3-contributor-free\":{\"options\":{\"reasoningEffort\":\"xhigh\"}}}}},\"agent\":{\"build\":{\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"variant\":\"xhigh\"},\"plan\":{\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"variant\":\"xhigh\"}}}"},"grid":null,"gridLaunchRecord":null,"startedBy":"agent"},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · grid · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w43","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"script":"0fd89f463aa4e0e9","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"},"startedBy":"agent"},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]}
+ "claude · grid · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w43","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"script":"0fd89f463aa4e0e9","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"},"startedBy":"agent"},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]}
},
"relaunch": {
- "claude · restart on a grid": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] claude -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · restart on a grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] codex -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · restart on a grid": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] claude -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · restart on a grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] codex -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"cursor · restart on a grid": {"result":{"ok":false,"error":"GRID_ENGINE_UNSUPPORTED","detail":"cursor cannot run on grid Home: Cursor Agent can only be re-pointed at another Cursor API (CURSOR_API_ENDPOINT), not at an OpenAI-compatible relay. It would have run on its own login instead. Engines that can: claude, codex, opencode, hermes, grok, pi, copilot."},"retargetLine":null,"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"opencode · restart on a grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","OPENCODE_CONFIG":"<root>/data/grid-engine-config/agent-opencode/opencode.json"},"extraArgs":["--standalone"],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"},"sessionModel":"home/qwen3-coder"}},"retargetLine":"[grid] opencode -> Home (net-home) · qwen3-coder · web search on","config":{"agent-opencode":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"model\": \"home/qwen3-coder\",\n \"provider\": {\n \"home\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Home\",\n \"options\": {\n \"baseURL\": \"https://grid.example/g/home/relay/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"qwen3-coder\": {\n \"name\": \"qwen3-coder\",\n \"limit\": {\n \"context\": 131072,\n \"output\": 32000\n }\n }\n }\n }\n },\n \"mcp\": {\n \"harness\": {\n \"type\": \"remote\",\n \"url\": \"https://control.example/v1/grid/web-mcp/\",\n \"enabled\": true,\n \"headers\": {\n \"Authorization\": \"Bearer {env:GRID_API_KEY}\"\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"pi · restart on a grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","PI_CODING_AGENT_DIR":"<root>/data/grid-engine-config/agent-pi"},"extraArgs":["--model","grid/qwen3-coder"],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"unsupported"}}},"retargetLine":"[grid] pi -> Home (net-home) · qwen3-coder · web search unsupported (Pi has no MCP client)","config":{"agent-pi":{"dir":{"models.json":{"mode":"600","text":"{\n \"providers\": {\n \"grid\": {\n \"name\": \"Autonomous Grid\",\n \"baseUrl\": \"https://grid.example/g/home/relay/v1\",\n \"api\": \"openai-completions\",\n \"apiKey\": \"$GRID_API_KEY\",\n \"models\": [\n {\n \"id\": \"qwen3-coder\",\n \"name\": \"qwen3-coder\",\n \"reasoning\": false,\n \"input\": [\n \"text\"\n ],\n \"contextWindow\": 131072,\n \"maxTokens\": 8192,\n \"cost\": {\n \"input\": 0,\n \"output\": 0,\n \"cacheRead\": 0,\n \"cacheWrite\": 0\n }\n }\n ]\n }\n }\n}"},"settings.json":{"mode":"600","text":"{\n \"skills\": [\n \"<root>/home/.pi/agent/skills\"\n ]\n}"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
@@ -60,18 +60,18 @@
"agy · restart on a grid": {"result":{"ok":false,"error":"GRID_ENGINE_UNSUPPORTED","detail":"agy cannot run on grid Home: Antigravity speaks the Gemini API, which this relay does not serve. It would have run on its own login instead. Engines that can: claude, codex, opencode, hermes, grok, pi, copilot."},"retargetLine":null,"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"copilot · restart on a grid": {"result":{"ok":true,"overrides":{"env":{"COPILOT_PROVIDER_BASE_URL":"https://grid.example/g/home/relay/v1","COPILOT_PROVIDER_API_KEY":"grid-key-1","COPILOT_MODEL":"qwen3-coder","GRID_API_KEY":"grid-key-1"},"extraArgs":["--additional-mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}"],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] copilot -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"terminal · restart on a grid": {"result":{"ok":false,"error":"GRID_ENGINE_UNSUPPORTED","detail":"terminal cannot run on grid Home: a terminal runs no engine to point at a grid — start one inside it and it will use its own login. It would have run on its own login instead. Engines that can: claude, codex, opencode, hermes, grok, pi, copilot."},"retargetLine":null,"config":null,"folder":{},"logged":[]},
- "claude · restart on a grid without a model": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-2"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}}},"retargetLine":"[grid] claude -> Home (net-home) · model chosen by the engine · web search unavailable","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · restart on a grid · another machine": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] claude -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · restart on a saved API, its key as saved now": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"http://127.0.0.1:<port>","ANTHROPIC_AUTH_TOKEN":"fresh-key","ANTHROPIC_MODEL":"vendor/model-a","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"200000"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"fresh-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}}},"retargetLine":"[grid] claude -> Router (api:router) · vendor/model-a · web search unavailable","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · retarget onto a grid": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] claude -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · retarget back to the own login": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_MODEL":"own-model"},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · fork on a grid": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] claude -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · restart on a grid without a model": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-2"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}}},"retargetLine":"[grid] codex -> Home (net-home) · model chosen by the engine · web search unavailable","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · restart on a grid · another machine": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] codex -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · restart on a saved API, its key as saved now": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"fresh-key"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"http://127.0.0.1:<port>/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","vendor/model-a","-c","model_context_window=200000","-c","model_auto_compact_token_limit=180000","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"fresh-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}}},"retargetLine":"[grid] codex -> Router (api:router) · vendor/model-a · web search unavailable","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · retarget onto a grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] codex -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · retarget back to the own login": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-m","own-model","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · fork on a grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] codex -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · restart on a grid without a model": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-2"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}}},"retargetLine":"[grid] claude -> Home (net-home) · model chosen by the engine · web search unavailable","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · restart on a grid · another machine": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] claude -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · restart on a saved API, its key as saved now": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"http://127.0.0.1:<port>","ANTHROPIC_AUTH_TOKEN":"fresh-key","ANTHROPIC_MODEL":"vendor/model-a","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"200000"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_API_KEY","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"fresh-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}}},"retargetLine":"[grid] claude -> Router (api:router) · vendor/model-a · web search unavailable","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · retarget onto a grid": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] claude -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · retarget back to the own login": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_MODEL":"own-model"},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · fork on a grid": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] claude -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · restart on a grid without a model": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-2"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}}},"retargetLine":"[grid] codex -> Home (net-home) · model chosen by the engine · web search unavailable","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · restart on a grid · another machine": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] codex -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · restart on a saved API, its key as saved now": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"fresh-key"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"http://127.0.0.1:<port>/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","vendor/model-a","-c","model_context_window=200000","-c","model_auto_compact_token_limit=180000","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"fresh-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}}},"retargetLine":"[grid] codex -> Router (api:router) · vendor/model-a · web search unavailable","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · retarget onto a grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] codex -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · retarget back to the own login": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-m","own-model","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · fork on a grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] codex -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"opencode · restart on a grid without a model": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-2","OPENCODE_CONFIG":"<root>/data/grid-engine-config/agent-opencode/opencode.json"},"extraArgs":["--standalone"],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"},"sessionModel":"home/Auto"}},"retargetLine":"[grid] opencode -> Home (net-home) · model chosen by the engine · web search unavailable","config":{"agent-opencode":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"model\": \"home/Auto\",\n \"provider\": {\n \"home\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Home\",\n \"options\": {\n \"baseURL\": \"https://grid.example/g/home/relay/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"Auto\": {\n \"name\": \"Auto\"\n }\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"opencode · restart on a grid · another machine": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","OPENCODE_CONFIG":"<root>/data/grid-engine-config/agent-opencode/opencode.json"},"extraArgs":["-m","home/qwen3-coder"],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"},"sessionModel":"home/qwen3-coder"}},"retargetLine":"[grid] opencode -> Home (net-home) · qwen3-coder · web search on","config":{"agent-opencode":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"provider\": {\n \"home\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Home\",\n \"options\": {\n \"baseURL\": \"https://grid.example/g/home/relay/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"qwen3-coder\": {\n \"name\": \"qwen3-coder\",\n \"limit\": {\n \"context\": 131072,\n \"output\": 32000\n }\n }\n }\n }\n },\n \"mcp\": {\n \"harness\": {\n \"type\": \"remote\",\n \"url\": \"https://control.example/v1/grid/web-mcp/\",\n \"enabled\": true,\n \"headers\": {\n \"Authorization\": \"Bearer {env:GRID_API_KEY}\"\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"opencode · restart on a saved API, its key as saved now": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"fresh-key","OPENCODE_CONFIG":"<root>/data/grid-engine-config/agent-opencode/opencode.json"},"extraArgs":["--standalone"],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"fresh-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"},"sessionModel":"api-router/vendor/model-a"}},"retargetLine":"[grid] opencode -> Router (api:router) · vendor/model-a · web search unavailable","config":{"agent-opencode":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"model\": \"api-router/vendor/model-a\",\n \"provider\": {\n \"api-router\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Router\",\n \"options\": {\n \"baseURL\": \"http://127.0.0.1:<port>/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"vendor/model-a\": {\n \"name\": \"vendor/model-a\",\n \"limit\": {\n \"context\": 200000,\n \"output\": 32000\n }\n }\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
@@ -106,12 +106,12 @@
"codex · restart on an API with no Bearer key": {"result":{"ok":false,"error":"API_UNAVAILABLE","detail":"Keyed does not take a Bearer key, so coding agents cannot run on it."},"retargetLine":null,"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"codex · restart on a grid · an old tmux": {"result":{"ok":false,"error":"TMUX_TOO_OLD_FOR_GRID","detail":"this machine's tmux is older than 3.2, which is the first version that can give a pane its own environment — so codex could not be pointed at grid Home."},"retargetLine":null,"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"opencode · restart as a named agent on a grid": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","OPENCODE_CONFIG":"<root>/data/grid-engine-config/agent-opencode/opencode.json"},"extraArgs":["--standalone"],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"},"sessionModel":"home/qwen3-coder"}},"retargetLine":"[grid] opencode -> Home (net-home) · qwen3-coder · web search on","config":{"agent-opencode":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"model\": \"home/qwen3-coder\",\n \"provider\": {\n \"home\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Home\",\n \"options\": {\n \"baseURL\": \"https://grid.example/g/home/relay/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"qwen3-coder\": {\n \"name\": \"qwen3-coder\",\n \"limit\": {\n \"context\": 131072,\n \"output\": 32000\n }\n }\n }\n }\n },\n \"mcp\": {\n \"harness\": {\n \"type\": \"remote\",\n \"url\": \"https://control.example/v1/grid/web-mcp/\",\n \"enabled\": true,\n \"headers\": {\n \"Authorization\": \"Bearer {env:GRID_API_KEY}\"\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · restart, started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · retarget onto a grid, started by an agent": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] claude -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · fork of one an agent started": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · restart, started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · retarget onto a grid, started by an agent": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] codex -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · fork of one an agent started": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]}
+ "claude · restart, started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · retarget onto a grid, started by an agent": {"result":{"ok":true,"overrides":{"env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"extraArgs":["--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"clearEnv":["ANTHROPIC_API_KEY","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","GRID_MCP_AUTHORIZATION","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] claude -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · fork of one an agent started": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · restart, started by an agent": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · retarget onto a grid, started by an agent": {"result":{"ok":true,"overrides":{"env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"extraArgs":["-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"clearEnv":["ANTHROPIC_API_KEY","ANTHROPIC_AUTH_TOKEN","ANTHROPIC_BASE_URL","ANTHROPIC_MODEL","OPENAI_API_KEY","OPENAI_BASE_URL","XAI_API_KEY","GROK_MODELS_BASE_URL","COPILOT_PROVIDER_API_KEY","COPILOT_PROVIDER_BASE_URL","COPILOT_MODEL","HERMES_INFERENCE_MODEL","HERMES_MANAGED_DIR","GROK_HOME","HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"],"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}}},"retargetLine":"[grid] codex -> Home (net-home) · qwen3-coder · web search on","config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · fork of one an agent started": {"result":{"ok":true,"overrides":{"env":{},"extraArgs":["-c","model_provider=\"openai\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"clearEnv":["HARNESS_DSH","HARNESS_DSH_DIR","HARNESS_WORKSPACE","HARNESS_CONTEXT_FILE","HARNESS_SKILLS_DIR","HARNESS_PRIVATE_GRID"]}},"retargetLine":null,"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]}
},
"other": {
"validate · claude · grid": {"ok":true},
@@ -179,19 +179,19 @@
},
"linux": {
"create": {
- "claude · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w1","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"72fe6c9b6937e258","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]},
- "codex · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w2","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"607ab05ca64c3b45","env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · qwen3-coder · web search on"]},
+ "claude · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w1","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"72fe6c9b6937e258","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]},
+ "codex · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w2","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"607ab05ca64c3b45","env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · qwen3-coder · web search on"]},
"opencode · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w4","opencode","--standalone"],"script":"0ef3217567b213cf","env":{"GRID_API_KEY":"grid-key-1","OPENCODE_CONFIG":"<root>/data/grid-engine-config/<label>/opencode.json"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":{"<label>":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"model\": \"home/qwen3-coder\",\n \"provider\": {\n \"home\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Home\",\n \"options\": {\n \"baseURL\": \"https://grid.example/g/home/relay/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"qwen3-coder\": {\n \"name\": \"qwen3-coder\",\n \"limit\": {\n \"context\": 131072,\n \"output\": 32000\n }\n }\n }\n }\n },\n \"mcp\": {\n \"harness\": {\n \"type\": \"remote\",\n \"url\": \"https://control.example/v1/grid/web-mcp/\",\n \"enabled\": true,\n \"headers\": {\n \"Authorization\": \"Bearer {env:GRID_API_KEY}\"\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] opencode -> Home (net-home) · qwen3-coder · web search on"]},
"pi · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w5","pi","--model","grid/qwen3-coder"],"script":"c59f64c9c100f673","env":{"GRID_API_KEY":"grid-key-1","PI_CODING_AGENT_DIR":"<root>/data/grid-engine-config/<label>"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"unsupported"}},"config":{"<label>":{"dir":{"models.json":{"mode":"600","text":"{\n \"providers\": {\n \"grid\": {\n \"name\": \"Autonomous Grid\",\n \"baseUrl\": \"https://grid.example/g/home/relay/v1\",\n \"api\": \"openai-completions\",\n \"apiKey\": \"$GRID_API_KEY\",\n \"models\": [\n {\n \"id\": \"qwen3-coder\",\n \"name\": \"qwen3-coder\",\n \"reasoning\": false,\n \"input\": [\n \"text\"\n ],\n \"contextWindow\": 131072,\n \"maxTokens\": 8192,\n \"cost\": {\n \"input\": 0,\n \"output\": 0,\n \"cacheRead\": 0,\n \"cacheWrite\": 0\n }\n }\n ]\n }\n }\n}"},"settings.json":{"mode":"600","text":"{\n \"skills\": [\n \"<root>/home/.pi/agent/skills\"\n ]\n}"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] pi -> Home (net-home) · qwen3-coder · web search unsupported (Pi has no MCP client)"]},
"hermes · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w6","hermes","-m","qwen3-coder"],"script":"80b2123b0f8552a1","env":{"OPENAI_BASE_URL":"https://grid.example/g/home/relay/v1","OPENAI_API_KEY":"grid-key-1","HERMES_INFERENCE_MODEL":"qwen3-coder","GRID_API_KEY":"grid-key-1","HERMES_MANAGED_DIR":"<root>/data/grid-engine-config/<label>"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":{"<label>":{"dir":{"config.yaml":{"mode":"600","text":"{\n \"mcp_servers\": {\n \"harness\": {\n \"url\": \"https://control.example/v1/grid/web-mcp/\",\n \"headers\": {\n \"Authorization\": \"Bearer ${GRID_API_KEY}\"\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] hermes -> Home (net-home) · qwen3-coder · web search on"]},
"grok · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w12","grok","-m","qwen3-coder"],"script":"3638ad48ae5fea74","env":{"GROK_MODELS_BASE_URL":"https://grid.example/g/home/relay/v1","XAI_API_KEY":"grid-key-1","GRID_API_KEY":"grid-key-1","GROK_HOME":"<root>/data/grid-engine-config/<label>"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":{"<label>":{"dir":{"config.toml":{"mode":"600","text":"[model.\"qwen3-coder\"]\nmodel = \"qwen3-coder\"\nbase_url = \"https://grid.example/g/home/relay/v1\"\nenv_key = \"GRID_API_KEY\"\n\n[mcp_servers.harness]\nurl = \"https://control.example/v1/grid/web-mcp/\"\nheaders = { Authorization = \"Bearer ${GRID_API_KEY}\" }\n"},"sessions":{"link":"<root>/home/.grok/sessions"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] grok -> Home (net-home) · qwen3-coder · web search on"]},
"copilot · grid": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w14","copilot","--additional-mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}"],"script":"1e83a0f95c1669fd","env":{"COPILOT_PROVIDER_BASE_URL":"https://grid.example/g/home/relay/v1","COPILOT_PROVIDER_API_KEY":"grid-key-1","COPILOT_MODEL":"qwen3-coder","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] copilot -> Home (net-home) · qwen3-coder · web search on"]},
- "claude · grid without a model, web tools or window": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w16","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"a73d03074982c240","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-2"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":null},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · model chosen by the engine · web search unavailable"]},
- "claude · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w17","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"72fe6c9b6937e258","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]},
- "claude · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w18","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"309b9bf7d77902e3","env":{"ANTHROPIC_BASE_URL":"http://127.0.0.1:<port>","ANTHROPIC_AUTH_TOKEN":"stale-key","ANTHROPIC_MODEL":"vendor/model-a","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"200000"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Router (api:router) · vendor/model-a · web search unavailable"]},
- "codex · grid without a model, web tools or window": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w19","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"99c3dfb625954926","env":{"GRID_API_KEY":"grid-key-2"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":null},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · model chosen by the engine · web search unavailable"]},
- "codex · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w20","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"607ab05ca64c3b45","env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · qwen3-coder · web search on"]},
- "codex · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w21","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"http://127.0.0.1:<port>/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","vendor/model-a","-c","model_context_window=200000","-c","model_auto_compact_token_limit=180000","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"99c3dfb625954926","env":{"GRID_API_KEY":"stale-key"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Router (api:router) · vendor/model-a · web search unavailable"]},
+ "claude · grid without a model, web tools or window": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w16","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"a73d03074982c240","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-2"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":null},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · model chosen by the engine · web search unavailable"]},
+ "claude · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w17","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"72fe6c9b6937e258","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]},
+ "claude · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w18","claude","--disallowedTools=WebSearch,WebFetch","--system-prompt-snapshot","off","--append-system-prompt","This session has no web tools. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; tell the user the web cannot be searched or read from here instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"309b9bf7d77902e3","env":{"ANTHROPIC_BASE_URL":"http://127.0.0.1:<port>","ANTHROPIC_AUTH_TOKEN":"stale-key","ANTHROPIC_MODEL":"vendor/model-a","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"200000"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Router (api:router) · vendor/model-a · web search unavailable"]},
+ "codex · grid without a model, web tools or window": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w19","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"99c3dfb625954926","env":{"GRID_API_KEY":"grid-key-2"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":null},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · model chosen by the engine · web search unavailable"]},
+ "codex · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w20","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"https://grid.example/g/home/relay/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-c","mcp_servers.harness.url=\"https://control.example/v1/grid/web-mcp/\"","-c","mcp_servers.harness.env_http_headers.Authorization=\"GRID_MCP_AUTHORIZATION\"","-m","qwen3-coder","-c","model_context_window=131072","-c","model_auto_compact_token_limit=117964","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"607ab05ca64c3b45","env":{"GRID_API_KEY":"grid-key-1","GRID_MCP_AUTHORIZATION":"Bearer grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Home (net-home) · qwen3-coder · web search on"]},
+ "codex · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w21","codex","-c","model_provider=\"grid\"","-c","model_providers.grid.name=\"Autonomous Grid\"","-c","model_providers.grid.base_url=\"http://127.0.0.1:<port>/v1\"","-c","model_providers.grid.env_key=\"GRID_API_KEY\"","-c","model_providers.grid.wire_api=\"responses\"","-c","model_providers.grid.supports_websockets=false","-c","web_search=\"disabled\"","-m","vendor/model-a","-c","model_context_window=200000","-c","model_auto_compact_token_limit=180000","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"99c3dfb625954926","env":{"GRID_API_KEY":"stale-key"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] codex -> Router (api:router) · vendor/model-a · web search unavailable"]},
"opencode · grid without a model, web tools or window": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w22","opencode","--standalone"],"script":"0ef3217567b213cf","env":{"GRID_API_KEY":"grid-key-2","OPENCODE_CONFIG":"<root>/data/grid-engine-config/<label>/opencode.json"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":null},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-2"},"webSearch":"unavailable"}},"config":{"<label>":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"model\": \"home/Auto\",\n \"provider\": {\n \"home\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Home\",\n \"options\": {\n \"baseURL\": \"https://grid.example/g/home/relay/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"Auto\": {\n \"name\": \"Auto\"\n }\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] opencode -> Home (net-home) · model chosen by the engine · web search unavailable"]},
"opencode · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w23","opencode","-m","home/qwen3-coder"],"script":"0ef3217567b213cf","env":{"GRID_API_KEY":"grid-key-1","OPENCODE_CONFIG":"<root>/data/grid-engine-config/<label>/opencode.json"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":{"<label>":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"provider\": {\n \"home\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Home\",\n \"options\": {\n \"baseURL\": \"https://grid.example/g/home/relay/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"qwen3-coder\": {\n \"name\": \"qwen3-coder\",\n \"limit\": {\n \"context\": 131072,\n \"output\": 32000\n }\n }\n }\n }\n },\n \"mcp\": {\n \"harness\": {\n \"type\": \"remote\",\n \"url\": \"https://control.example/v1/grid/web-mcp/\",\n \"enabled\": true,\n \"headers\": {\n \"Authorization\": \"Bearer {env:GRID_API_KEY}\"\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] opencode -> Home (net-home) · qwen3-coder · web search on"]},
"opencode · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w24","opencode","--standalone"],"script":"0ef3217567b213cf","env":{"GRID_API_KEY":"stale-key","OPENCODE_CONFIG":"<root>/data/grid-engine-config/<label>/opencode.json"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":{"<label>":{"dir":{"opencode.json":{"mode":"600","text":"{\n \"$schema\": \"https://opencode.ai/config.json\",\n \"model\": \"api-router/vendor/model-a\",\n \"provider\": {\n \"api-router\": {\n \"npm\": \"@ai-sdk/openai-compatible\",\n \"name\": \"Router\",\n \"options\": {\n \"baseURL\": \"http://127.0.0.1:<port>/v1\",\n \"apiKey\": \"{env:GRID_API_KEY}\"\n },\n \"models\": {\n \"vendor/model-a\": {\n \"name\": \"vendor/model-a\",\n \"limit\": {\n \"context\": 200000,\n \"output\": 32000\n }\n }\n }\n }\n }\n}\n"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] opencode -> Router (api:router) · vendor/model-a · web search unavailable"]},
@@ -206,12 +206,12 @@
"grok · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w33","grok","-m","vendor/model-a"],"script":"3638ad48ae5fea74","env":{"GROK_MODELS_BASE_URL":"http://127.0.0.1:<port>/v1","XAI_API_KEY":"stale-key","GRID_API_KEY":"stale-key","GROK_HOME":"<root>/data/grid-engine-config/<label>"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":{"<label>":{"dir":{"config.toml":{"mode":"600","text":"[model.\"vendor/model-a\"]\nmodel = \"vendor/model-a\"\nbase_url = \"http://127.0.0.1:<port>/v1\"\nenv_key = \"GRID_API_KEY\"\n"},"sessions":{"link":"<root>/home/.grok/sessions"}}}},"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] grok -> Router (api:router) · vendor/model-a · web search unavailable"]},
"copilot · grid · another machine": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w35","copilot","--additional-mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}"],"script":"1e83a0f95c1669fd","env":{"COPILOT_PROVIDER_BASE_URL":"https://grid.example/g/home/relay/v1","COPILOT_PROVIDER_API_KEY":"grid-key-1","COPILOT_MODEL":"qwen3-coder","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] copilot -> Home (net-home) · qwen3-coder · web search on"]},
"copilot · saved API": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w36","copilot"],"script":"bdbea8aa1d3e4b33","env":{"COPILOT_PROVIDER_BASE_URL":"http://127.0.0.1:<port>/v1","COPILOT_PROVIDER_API_KEY":"stale-key","COPILOT_MODEL":"vendor/model-a"},"grid":{"baseUrl":"http://127.0.0.1:<port>/v1","model":"vendor/model-a"},"gridLaunchRecord":{"override":{"networkId":"api:router","networkName":"Router","baseUrl":"http://127.0.0.1:<port>/v1","apiKey":"stale-key","model":"vendor/model-a","contextWindow":200000},"webSearch":"unavailable"}},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] copilot -> Router (api:router) · vendor/model-a · web search unavailable"]},
- "claude · own login": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w38","claude","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started."],"script":"6dc5fd2b979b6e45","env":{},"grid":null,"gridLaunchRecord":null},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · own login": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w39","codex","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"1ff40fd30d1478c3","env":{},"grid":null,"gridLaunchRecord":null},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · own login · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w40","claude","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"script":"6dc5fd2b979b6e45","env":{},"grid":null,"gridLaunchRecord":null,"startedBy":"agent"},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "codex · own login · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w41","codex","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"script":"1ff40fd30d1478c3","env":{},"grid":null,"gridLaunchRecord":null,"startedBy":"agent"},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · own login": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w38","claude","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started."],"script":"6dc5fd2b979b6e45","env":{},"grid":null,"gridLaunchRecord":null},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · own login": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w39","codex","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\""],"script":"1ff40fd30d1478c3","env":{},"grid":null,"gridLaunchRecord":null},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "claude · own login · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w40","claude","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"script":"6dc5fd2b979b6e45","env":{},"grid":null,"gridLaunchRecord":null,"startedBy":"agent"},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
+ "codex · own login · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w41","codex","-c","developer_instructions=\"You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started. Codex's sandbox cannot reach Harness: run each harness command outside it, with escalated permissions.\\n\\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages.\""],"script":"1ff40fd30d1478c3","env":{},"grid":null,"gridLaunchRecord":null,"startedBy":"agent"},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
"opencode · own login · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w42","opencode"],"script":"ba08f0832f2d5eeb","env":{"OPENCODE_CONFIG_CONTENT":"{\"permission\":\"ask\",\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"small_model\":\"opencode/muse-spark-1.3-contributor-free\",\"provider\":{\"opencode\":{\"models\":{\"muse-spark-1.3-contributor-free\":{\"options\":{\"reasoningEffort\":\"xhigh\"}}}}},\"agent\":{\"build\":{\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"variant\":\"xhigh\"},\"plan\":{\"model\":\"opencode/muse-spark-1.3-contributor-free\",\"variant\":\"xhigh\"}}}"},"grid":null,"gridLaunchRecord":null,"startedBy":"agent"},"config":null,"folder":{"AGENTS.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":[]},
- "claude · grid · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w43","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"script":"72fe6c9b6937e258","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"},"startedBy":"agent"},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]}
+ "claude · grid · started by an agent": {"result":{"ok":false,"error":"RECORDED","detail":"the pane is not opened by this spec"},"pane":{"argv":["/usr/bin/env","DISABLE_AUTO_UPDATE=true","/bin/zsh","-lic",". '<root>/data/launch/<script>.sh'","harness-engine","<root>/work/w43","claude","--disallowedTools=WebSearch,WebFetch","--mcp-config","{\"mcpServers\":{\"harness\":{\"type\":\"http\",\"url\":\"https://control.example/v1/grid/web-mcp/\",\"headers\":{\"Authorization\":\"Bearer ${GRID_API_KEY}\"}}}}","--allowedTools=mcp__harness__web_search,mcp__harness__web_read","--system-prompt-snapshot","off","--settings","{\"attribution\":{\"pr\":\"🤖 Generated with [Claude Code](https://claude.com/claude-code)\\n🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.\"}}","--append-system-prompt","Web tools in this session come from the `harness` MCP server only: `mcp__harness__web_search` searches the web and `mcp__harness__web_read` reads pages. The built-in `WebSearch` and `WebFetch` tools are not available in this session, even where earlier turns of this conversation used them — do not call them; call the `harness` tools instead.\n\nYou are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.\n\nWhen you create or update a pull request, its body must include this exact line once, near the end and after any other attribution line: `🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.` It says another agent started this session in Harness. Do not add it to commit messages."],"script":"72fe6c9b6937e258","env":{"ANTHROPIC_BASE_URL":"https://grid.example/g/home/relay","ANTHROPIC_AUTH_TOKEN":"grid-key-1","ANTHROPIC_MODEL":"qwen3-coder","CLAUDE_CODE_MAX_CONTEXT_TOKENS":"131072","GRID_API_KEY":"grid-key-1"},"grid":{"baseUrl":"https://grid.example/g/home/relay/v1","model":"qwen3-coder"},"gridLaunchRecord":{"override":{"networkId":"net-home","networkName":"Home","baseUrl":"https://grid.example/g/home/relay/v1","apiKey":"grid-key-1","model":"qwen3-coder","mcpUrl":"https://control.example/v1/grid/web-mcp/","contextWindow":131072},"webSearch":"on"},"startedBy":"agent"},"config":null,"folder":{"CLAUDE.md":{"mode":"644","text":"<!-- harness:apis -->\n## Saved APIs\n\nThe user can save API connections in Harness → Models → APIs on this computer.\nRun `harness api list --json` to discover the current connections; the list contains no keys.\nUse `harness api request <id> <path> --method POST --data @request.json` for JSON APIs,\nor `harness api run <id> -- <command> [args]` for a provider SDK/tool. The latter supplies\nthe connection's documented key variable and HARNESS_API_BASE_URL to that tool only.\nUse these APIs when needed for the user's task. Never print, paste, or save API keys in a project.\nAdding an API does not change this harness's model or subscription.\n"}},"logged":["[grid] claude -> Home (net-home) · qwen3-coder · web search on"]}
},
"relaunch": {
diff --git a/cli/src/harnessd/services.spec.ts b/cli/src/harnessd/services.spec.ts
index 7199c722c..b460756a7 100644
--- a/cli/src/harnessd/services.spec.ts
+++ b/cli/src/harnessd/services.spec.ts
@@ -462,7 +462,7 @@ describe('ServiceSupervisor', () => {
const hosted = Object.values(SERVICE_HOSTS).flatMap((host) => host.services)
expect(KNOWN_SERVICES).toEqual(hosted)
expect(new Set(hosted).size).toBe(hosted.length)
- expect(SERVICE_HOSTS.edge.services).toEqual(['workspaces', 'usage', 'monitor', 'projects', 'handoff', 'recaps', 'windowNames', 'shell', 'connectors'])
+ expect(SERVICE_HOSTS.edge.services).toEqual(['workspaces', 'usage', 'monitor', 'projects', 'handoff', 'recaps', 'windowNames', 'shell', 'connectors', 'agentCleanup'])
})
})
@@ -478,7 +478,7 @@ describe('which services the core leaves to its master', () => {
expect(serviceProcessesEnv(serviceSpecs({ HARNESSD_SERVICES: 'store' }, SERVICE_HOSTS), 42)).toMatchObject({ [SERVICE_PROCESSES_ENV]: 'store' })
// By service, not by process: a core from before the edge host routes the services it knows of it.
const hosted = serviceSpecs({ HARNESSD_SERVICES: 'edge' }, SERVICE_HOSTS)
- expect(serviceProcessesEnv(hosted, 42)).toEqual({ [SERVICE_PROCESSES_ENV]: 'workspaces,usage,monitor,projects,handoff,recaps,windowNames,shell,connectors', HARNESSD_SERVICES: 'workspaces,usage,monitor,projects,handoff,recaps,windowNames,shell,connectors', [ENGINE_LIVE_ENV]: '42:1', [ENGINE_RUNTIME_ENV]: '42:1', [ENGINE_SCREEN_ENV]: '42:1', [ENGINE_MODEL_CONTROL_ENV]: '42:1', [ENGINE_QUESTION_CONTROL_ENV]: '42:1', [ENGINE_SUBMISSION_ENV]: '42:1', [ENGINE_NATIVE_CONTROL_ENV]: '42:1' })
+ expect(serviceProcessesEnv(hosted, 42)).toEqual({ [SERVICE_PROCESSES_ENV]: 'workspaces,usage,monitor,projects,handoff,recaps,windowNames,shell,connectors,agentCleanup', HARNESSD_SERVICES: 'workspaces,usage,monitor,projects,handoff,recaps,windowNames,shell,connectors,agentCleanup', [ENGINE_LIVE_ENV]: '42:1', [ENGINE_RUNTIME_ENV]: '42:1', [ENGINE_SCREEN_ENV]: '42:1', [ENGINE_MODEL_CONTROL_ENV]: '42:1', [ENGINE_QUESTION_CONTROL_ENV]: '42:1', [ENGINE_SUBMISSION_ENV]: '42:1', [ENGINE_NATIVE_CONTROL_ENV]: '42:1' })
expect([...servicesTheMasterRuns({ ...supervised, ...serviceProcessesEnv(hosted, 42) }, known)]).toEqual(['workspaces'])
expect(serviceProcessesEnv([], 42)).toEqual({ [SERVICE_PROCESSES_ENV]: '', HARNESSD_SERVICES: 'none', [ENGINE_LIVE_ENV]: '42:1', [ENGINE_RUNTIME_ENV]: '42:1', [ENGINE_SCREEN_ENV]: '42:1', [ENGINE_MODEL_CONTROL_ENV]: '42:1', [ENGINE_QUESTION_CONTROL_ENV]: '42:1', [ENGINE_SUBMISSION_ENV]: '42:1', [ENGINE_NATIVE_CONTROL_ENV]: '42:1' })
expect([...servicesTheMasterRuns({ ...supervised, ...serviceProcessesEnv(specs, 42) }, known)]).toEqual(['search', 'workspaces'])
diff --git a/cli/src/harnessd/services.ts b/cli/src/harnessd/services.ts
index 0c3859dff..365e19021 100644
--- a/cli/src/harnessd/services.ts
+++ b/cli/src/harnessd/services.ts
@@ -380,8 +380,9 @@ export const SERVICE_HOSTS: Readonly<Record<string, ServiceHostSpec>> = {
// The monitor parses up to 8 MB of ioreg output per Monitor poll on a Mac with a GPU, hence more
// than workspaces alone had. The recaps hold each session's last three recaps, answers (8 KiB each at
// most) and asks, as the core did while they ran in it, and a few timers per open turn. The window
- // names hold at most 400 short names; their model runs in its own process (lib/oneshot.ts).
- edge: { services: ['workspaces', 'usage', 'monitor', 'projects', 'handoff', 'recaps', 'windowNames', 'shell', 'connectors'], heapLimitMiB: 384, rssLimitMiB: 768 },
+ // names hold at most 400 short names; their model runs in its own process (lib/oneshot.ts). The agent cleanup
+ // holds the harnesses the person joined (a thousand ids at most) and looks at the live ones once a minute.
+ edge: { services: ['workspaces', 'usage', 'monitor', 'projects', 'handoff', 'recaps', 'windowNames', 'shell', 'connectors', 'agentCleanup'], heapLimitMiB: 384, rssLimitMiB: 768 },
// The orchestrator (services/orchestratorProcess.ts), an experiment: started only once it is on, for a
// saved project or a request (core/api.ts `EXPERIMENTS`). Its projects' files and the frames of their
// Directors; the agents it runs are the core's.
diff --git a/cli/src/lib/agentAttribution.spec.ts b/cli/src/lib/agentAttribution.spec.ts
index a5c4352b6..4cd65d9fd 100644
--- a/cli/src/lib/agentAttribution.spec.ts
+++ b/cli/src/lib/agentAttribution.spec.ts
@@ -22,7 +22,7 @@ const codexValue = (...texts: string[]) => ['-c', `developer_instructions=${JSON
describe('what Harness tells an agent as it starts', () => {
it('is the line the owner chose, and the instructions every harness gets', () => {
expect(AGENT_ATTRIBUTION_LINE).toBe('🤝 Started by another agent in [Harness](https://harness.autonomous.ai) — agents that run agents.')
- expect(HARNESS_INSTRUCTIONS).toBe('You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Only use or close harnesses you started.')
+ expect(HARNESS_INSTRUCTIONS).toBe('You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.')
})
it('tells every harness how to use Harness, and adds the pull request line only to one another agent started', () => {
diff --git a/cli/src/lib/agentAttribution.ts b/cli/src/lib/agentAttribution.ts
index a41fc930d..3b27da5d0 100644
--- a/cli/src/lib/agentAttribution.ts
+++ b/cli/src/lib/agentAttribution.ts
@@ -20,12 +20,15 @@ export type StartedBy = 'agent'
/**
* What every harness is told as it starts, while the person lets agents start agents (Settings ▸ Agents): that it runs in
- * Harness, and the verbs that hand work to another agent. An agent that does not know Harness is there never uses it,
- * and every harness Harness starts is an agent that could (docs/design/2026-10-11-agent-client.md).
+ * Harness, the verbs that hand work to another agent, and when to: every harness started spends the person's tokens, so
+ * only for work that gains from it, and closed once answered (owner, October 12). An agent that does not know Harness is
+ * there never uses it, and every harness Harness starts is an agent that could (docs/design/2026-10-11-agent-client.md).
+ * It is in every system prompt: keep it short.
*/
export const HARNESS_INSTRUCTIONS = 'You are running in Harness, and the person can watch you. You can hand work to other agents '
+ '(Claude Code, Codex) that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` '
- + '(see `harness --help`). Only use or close harnesses you started.'
+ + '(see `harness --help`). Start one only for work that can run in parallel, needs another model, or takes long; '
+ + 'do quick work yourself, and close each one once you have its answer. Only use or close harnesses you started.'
/**
* A launch's argv for what Harness tells the agent: the instructions above while agents may start agents, and the pull
diff --git a/cli/src/lib/agentFrame.spec.ts b/cli/src/lib/agentFrame.spec.ts
index e3dd951d6..8bc0f016f 100644
--- a/cli/src/lib/agentFrame.spec.ts
+++ b/cli/src/lib/agentFrame.spec.ts
@@ -78,6 +78,13 @@ describe('agentFrame', () => {
expect((await agentFrame(row, { selectedModel: null, terminalAvailable: true })).parentAgentId).toBe('lead-1')
})
+ it('says another agent started it, wherever that agent runs, and null for the person\'s', async () => {
+ const row = session(null)
+ expect(await agentFrame(row, { selectedModel: null, terminalAvailable: true })).toHaveProperty('startedBy', null)
+ row.startedBy = 'agent'
+ expect((await agentFrame(row, { selectedModel: null, terminalAvailable: true })).startedBy).toBe('agent')
+ })
+
// The SCM seam (scm/scmProjects.ts) names the project's SCM and changes nothing else on it: every
// field a client reads is exactly what agentProject reports. A folder no SCM claims is `none`.
it('names the project\'s SCM beside exactly what agentProject reports, and none outside any repository', async () => {
diff --git a/cli/src/lib/agentFrame.ts b/cli/src/lib/agentFrame.ts
index 25e4e98f4..912dfb17d 100644
--- a/cli/src/lib/agentFrame.ts
+++ b/cli/src/lib/agentFrame.ts
@@ -110,6 +110,10 @@ export type AgentFrame = {
* starter's current name (desktop `Agent.parentAgentId`).
*/
parentAgentId: string | null
+ /** `agent` when another agent started this harness, wherever that agent runs (registry `startedBy`); null when the
+ * person did. The agent verbs read it to keep such a harness from starting harnesses itself (agentClient/model.ts
+ * `callerOf`): `parentAgentId` names only a starter on this machine. */
+ startedBy: 'agent' | null
/** Whether `agent_fork` can do anything for this engine (lib/forkAgent.ts) — natively, or by a
* handoff. A client hides the Fork action on a false rather than offering a button that refuses. */
forkable: boolean
@@ -264,6 +268,7 @@ export async function agentFrame(
verdict: dsh?.verdict ?? null,
forkedFrom: s.forkedFrom ? { agentId: s.forkedFrom.agentId, name: s.forkedFrom.name } : null,
parentAgentId: s.parentAgentId ?? null,
+ startedBy: s.startedBy === 'agent' ? 'agent' : null,
forkable: engineCanFork(s.engine),
resumeMode: resumeMode(s.engine),
permissionMode: s.permissionMode ?? null,
diff --git a/cli/src/lib/closeAgentService.spec.ts b/cli/src/lib/closeAgentService.spec.ts
index c5e44cbe4..b38dc567f 100644
--- a/cli/src/lib/closeAgentService.spec.ts
+++ b/cli/src/lib/closeAgentService.spec.ts
@@ -185,6 +185,29 @@ it('an inspect asked for while a close is saving waits for it, and is never answ
expect(await closing).toEqual({ closed: true })
expect(await inspecting).toEqual({ error: 'AGENT_CHANGED' })
})
+it('a close of one already closed, the same conversation, is closed: asked after it, or waiting behind the close that did it', async () => {
+ // A window closing a tab closes its panes one at a time; the harnesses a pane's harness started close with it
+ // (services/agentCleanup.ts), and the window's own close of one of them was answered AGENT_CHANGED.
+ const asked = request()
+ const closedAlready = vi.fn((r: AgentCloseRequest) => r.agentId === asked.agentId && r.sessionId === asked.sessionId && !registry.byAgent(r.agentId))
+ service.dispose()
+ service = new CloseAgentService({ ...deps, closedAlready })
+ let finish!: () => void
+ vi.mocked(deps.checkpoint).mockImplementationOnce(() => new Promise(resolve => { finish = resolve }))
+ const cleanup = service.request(request('now'))
+ await vi.waitFor(() => expect(finish).toBeTypeOf('function'))
+ const window = service.request(request('idle'))
+ finish()
+ expect(await cleanup).toEqual({ closed: true })
+ expect(await window).toEqual({ closed: true })
+ expect(await service.request(request('now'))).toEqual({ closed: true })
+ expect(deps.stop).toHaveBeenCalledOnce()
+ // Asking what it is doing, or to close it after its task, is about a live one: it changed.
+ expect(await service.request(request('inspect'))).toEqual({ error: 'AGENT_CHANGED' })
+ expect(await service.request(request('after_task'))).toEqual({ error: 'AGENT_CHANGED' })
+ // Another conversation by that id was not closed.
+ expect(await service.request({ ...request('now'), sessionId: 'other' })).toEqual({ error: 'AGENT_CHANGED' })
+})
it('a close now asked for while a close after the task is being planned closes now', async () => {
let read!: (activity: CloseActivity) => void
vi.mocked(deps.activity).mockImplementationOnce(() => new Promise(resolve => { read = resolve }))
diff --git a/cli/src/lib/closeAgentService.ts b/cli/src/lib/closeAgentService.ts
index 334b78404..302679f21 100644
--- a/cli/src/lib/closeAgentService.ts
+++ b/cli/src/lib/closeAgentService.ts
@@ -107,6 +107,8 @@ export interface CloseAgentServiceDeps {
stop(agentId: string, options: StopAgentOptions): Promise<void>
changed(s: RegisteredSession): void
openTabs?: { assertHidden(s: RegisteredSession): Promise<void>; isHidden(s: RegisteredSession): boolean }
+ /** Whether the conversation [request] names is closed already: no longer live, and kept as stopped. */
+ closedAlready?(request: AgentCloseRequest): boolean
now?: () => number
sleep?: (ms: number) => Promise<void>
}
@@ -150,7 +152,7 @@ export class CloseAgentService {
request(request: AgentCloseRequest): Promise<AgentCloseResult> {
const current = this.deps.registry.byAgent(request.agentId)
- if (!matches(current, request, this.now())) return Promise.resolve({ error: 'AGENT_CHANGED' })
+ if (!matches(current, request, this.now())) return Promise.resolve(this.changed(request))
if (request.mode === 'cancel') { this.cancel(request.agentId); return Promise.resolve({ cancelled: true }) }
// A job answers only a request of its own kind. When any job running for the agent answered every
// request, a close asked for while another window (or the cleanup preview) was inspecting the agent
@@ -174,6 +176,16 @@ export class CloseAgentService {
return job
}
+ /**
+ * The answer to a close of an agent that is not the one asked about. One already closed, the same conversation, is
+ * closed: a window closing a tab after a pane in it, its starter, was closed takes the harnesses that starter started
+ * with it (services/agentCleanup.ts), and the window's own close of them, asked a moment later, was answered
+ * AGENT_CHANGED and shown as "Could not confirm that it stopped". Anything else is AGENT_CHANGED.
+ */
+ private changed(request: AgentCloseRequest): AgentCloseResult {
+ return (request.mode === 'idle' || request.mode === 'now') && this.deps.closedAlready?.(request) ? { closed: true } : { error: 'AGENT_CHANGED' }
+ }
+
private async execute(request: AgentCloseRequest, target: string, revision: number): Promise<AgentCloseResult> {
const askedAt = this.now()
// Read before the try: a row can be changed in place while it runs.
@@ -199,7 +211,7 @@ export class CloseAgentService {
let tried = target, result = first, waiting = false
for (;;) {
const s = same()
- if (!s) return { error: 'AGENT_CHANGED' }
+ if (!s) return this.changed(request)
const bound = identity(s) !== tried
// Neither unreadable nor newly bound: a refusal of its own (working, a failed save), answered as it is.
if (!bound && result.error !== 'IDENTITY_UNAVAILABLE') return result
@@ -214,7 +226,7 @@ export class CloseAgentService {
waiting = true
if (!bound) await this.sleep(Math.min(CLOSE_BIND_RETRY_MS, deadline - this.now()))
const latest = same()
- if (!latest) return { error: 'AGENT_CHANGED' }
+ if (!latest) return this.changed(request)
tried = identity(latest)
result = await this.attempt(request, tried, revision)
if (result.closed) return result
@@ -237,7 +249,8 @@ export class CloseAgentService {
await this.deps.openTabs.assertHidden(s)
}
const s = current()
- if (!s) return { error: 'AGENT_CHANGED' }
+ // Closed by the job ahead of this one (another window's close, or the cleanup's): closed all the same.
+ if (!s) return this.changed(request)
const observed = await this.deps.activity(s)
// Close belongs to the global workspace. Other viewers do not change
// whether the session has unfinished work.
diff --git a/cli/src/lib/terminalStreamManager.spec.ts b/cli/src/lib/terminalStreamManager.spec.ts
index 627a33330..af2c6427b 100644
--- a/cli/src/lib/terminalStreamManager.spec.ts
+++ b/cli/src/lib/terminalStreamManager.spec.ts
@@ -185,8 +185,9 @@ describe('TerminalStreamManager', () => {
// agent runs on a sleeping grid starts that grid. A read-only watcher's input reaches no pty, so it
// must start nothing either.
const typed: string[] = []
+ const person: boolean[] = []
await manager.stop()
- manager = newManager({ onInput: (agentId) => { typed.push(agentId) } })
+ manager = newManager({ onInput: (agentId, byPerson) => { typed.push(agentId); person.push(byPerson) } })
await manager.handleFrame('web-1', 'terminal_open', { requestId: 'open-1', protocolVersion: 3, agentId: 'agent-1', cols: 120, rows: 40 })
const controller = sent.find((frame) => frame.type === 'terminal_ready' && frame.connId === 'web-1')!.payload.streamId as string
await manager.handleFrame('web-2', 'terminal_open', {
@@ -205,7 +206,11 @@ describe('TerminalStreamManager', () => {
await manager.handleBinary('web-1', { kind: TerminalBinaryKind.input, streamId: controller, seq: 0, compressed: false, bytes: Buffer.from('x') })
await manager.handleBinary('web-1', { kind: TerminalBinaryKind.paste, streamId: controller, seq: 0, compressed: false, bytes: Buffer.from('hello') })
- expect(typed).toEqual(['agent-1', 'agent-1'])
+ // What the window's terminal answers by itself (focus, its attributes) is input too, but not the person's: every
+ // harness another agent starts opens as a pane beside its starter, and its reports made each one the person's.
+ await manager.handleBinary('web-1', { kind: TerminalBinaryKind.input, streamId: controller, seq: 1, compressed: false, bytes: Buffer.from('\x1b[I\x1b[?1;2c') })
+ expect(typed).toEqual(['agent-1', 'agent-1', 'agent-1'])
+ expect(person).toEqual([true, true, false])
})
it('opens with a keyframe, streams coalesced output, and writes ordered raw input', async () => {
diff --git a/cli/src/lib/terminalStreamManager.ts b/cli/src/lib/terminalStreamManager.ts
index c4e6a5d1a..b4117b93a 100644
--- a/cli/src/lib/terminalStreamManager.ts
+++ b/cli/src/lib/terminalStreamManager.ts
@@ -1,4 +1,5 @@
import { randomUUID } from 'node:crypto'
+import { typedByPerson } from './typedInput.js'
import { unlink } from 'node:fs/promises'
import { deflateSync } from 'node:zlib'
import { ENGINES } from '../engines/types.js'
@@ -171,9 +172,11 @@ export interface TerminalStreamManagerDeps {
* An agent's terminal just took input (a keystroke or a paste) from the stream that holds it — never
* from a watcher, whose input is refused before it gets here. Called on every accepted frame, so it
* must be cheap and must not throw; the daemon starts a sleeping grid from it (the keystroke prewarm,
- * grid-reads-without-waking issue 03), which is why it lives here, below every client that types.
+ * grid-reads-without-waking issue 03), which is why it lives here, below every client that types. `typed`: the frame
+ * holds what the person typed or pasted, not only what the terminal sends by itself (lib/typedInput.ts), which the
+ * agent cleanup reads as the person joining a harness another agent started (services/agentCleanup.ts).
*/
- onInput?: (agentId: string) => void
+ onInput?: (agentId: string, typed: boolean) => void
}
function sizeFrom(payload: FramePayload): TerminalStreamSize | null {
@@ -644,7 +647,7 @@ export class TerminalStreamManager {
}
state.lastInputSeq = inputSeq
state.expiresAt = this.now() + HEARTBEAT_TIMEOUT_MS
- this.tookInput(state)
+ this.tookInput(state, typedByPerson(bytes))
// Not awaited. `writeRaw` hands its `send-keys` to the control client's FIFO synchronously, so
// keystroke order is already fixed by the time it returns its promise — and the seq above is
// spent, so the next frame cannot race this one. Awaiting the reply held the whole local
@@ -672,8 +675,8 @@ export class TerminalStreamManager {
}
/** See `TerminalStreamManagerDeps.onInput`. A listener's failure is its own: input is never held by it. */
- private tookInput(state: ActiveStream): void {
- try { this.deps.onInput?.(state.agentId) } catch (error) {
+ private tookInput(state: ActiveStream, typed: boolean): void {
+ try { this.deps.onInput?.(state.agentId, typed) } catch (error) {
this.diagnostic(state, 'input_listener_failed', { reason: error instanceof Error ? error.message : String(error) })
}
}
@@ -718,7 +721,7 @@ export class TerminalStreamManager {
return
}
state.expiresAt = this.now() + HEARTBEAT_TIMEOUT_MS
- this.tookInput(state)
+ this.tookInput(state, true)
const result = await state.handle.pasteRaw(text)
if (result.state !== 'succeeded') {
this.sendError(state.connId, 'TERMINAL_PASTE_FAILED', { streamId: state.streamId, message: result.reason })
diff --git a/cli/src/lib/typedInput.spec.ts b/cli/src/lib/typedInput.spec.ts
new file mode 100644
index 000000000..891c060b8
--- /dev/null
+++ b/cli/src/lib/typedInput.spec.ts
@@ -0,0 +1,21 @@
+import { describe, expect, it } from 'vitest'
+import { typedByPerson } from './typedInput.js'
+
+const bytes = (text: string): Uint8Array => new TextEncoder().encode(text)
+
+describe('whether a terminal input frame is the person typing', () => {
+ it('counts what a person types: letters, Enter, Backspace, Ctrl-C, Escape, Alt with a key, a paste', () => {
+ for (const typed of ['a', 'hello', '\r', '\x7f', '\x03', '\x1b', '\x1bb', '\x1b1', '\x1b[200~fix it\x1b[201~', '\x1b[Ax']) {
+ expect(typedByPerson(bytes(typed)), JSON.stringify(typed)).toBe(true)
+ }
+ })
+
+ it('does not count what the terminal sends by itself, nor keys that only move around', () => {
+ for (const reply of [
+ '', '\x1b[?1;2c', '\x1b[>0;276;0c', '\x1b[12;40R', '\x1b[I', '\x1b[O', // attributes, cursor, focus
+ '\x1b]11;rgb:0000/0000/0000\x1b\\', '\x1b]10;rgb:ffff/ffff/ffff\x07', '\x1bP1$r0m\x1b\\', '\x1b]52;c;', // colours, a DCS reply, unended
+ '\x1b[<0;10;5M', '\x1b[<0;10;5m', '\x1b[M #!', // the mouse, in both encodings
+ '\x1b[A', '\x1b[1;5C', '\x1bOA', '\x1b[3~', '\x1b[15~', '\x1b\x1b[A', '\x1b=', '\x1b[', // keys that move, and halves
+ ]) expect(typedByPerson(bytes(reply)), JSON.stringify(reply)).toBe(false)
+ })
+})
diff --git a/cli/src/lib/typedInput.ts b/cli/src/lib/typedInput.ts
new file mode 100644
index 000000000..0dd7ff40d
--- /dev/null
+++ b/cli/src/lib/typedInput.ts
@@ -0,0 +1,54 @@
+/**
+ * Whether a terminal's input frame is the person typing, or what the terminal sends by itself.
+ *
+ * A window's or the phone's terminal answers the engine's queries and reports what happens to it, unasked: a device
+ * attribute report (`ESC [ ? 1 ; 2 c`), the cursor's position, the colours (`ESC ] 11 ; rgb:… ESC \`), focus in and
+ * out (`ESC [ I`, `ESC [ O`), the mouse over a pane that tracks it. Every harness another agent starts now opens as a
+ * pane beside its starter (docs/design/2026-10-11-agent-client.md), so those reports reach almost every one of them:
+ * taken for typing, they made every such harness the person's, and the agent cleanup never closed one
+ * (services/agentCleanup.ts). Typing is what is left once they are taken out: a letter, Enter, Backspace, Ctrl-C.
+ * Arrow keys and other keys sent as escape sequences are not counted: moving around a pane is not giving it work.
+ */
+
+const ESC = 0x1b
+
+const isAlnum = (byte: number): boolean => (byte >= 0x30 && byte <= 0x39) || (byte >= 0x41 && byte <= 0x5a) || (byte >= 0x61 && byte <= 0x7a)
+
+/** The end of the escape sequence that starts at [i] (an ESC with a byte after it), the index after it. */
+function sequenceEnd(bytes: Uint8Array, i: number): number {
+ const kind = bytes[i + 1]
+ // CSI: parameters, intermediates, one final byte. `ESC [ M` with three bytes after it is the old mouse report,
+ // whose three bytes can be letters.
+ if (kind === 0x5b) {
+ let j = i + 2
+ while (j < bytes.length && bytes[j] >= 0x30 && bytes[j] <= 0x3f) j++
+ while (j < bytes.length && bytes[j] >= 0x20 && bytes[j] <= 0x2f) j++
+ return bytes[j] === 0x4d && j === i + 2 ? j + 4 : j + 1
+ }
+ // OSC, DCS, SOS, PM, APC: a string to BEL or ESC \.
+ if (kind === 0x5d || kind === 0x50 || kind === 0x58 || kind === 0x5e || kind === 0x5f) {
+ for (let j = i + 2; j < bytes.length; j++) {
+ if (bytes[j] === 0x07) return j + 1
+ if (bytes[j] === ESC && bytes[j + 1] === 0x5c) return j + 2
+ }
+ return bytes.length
+ }
+ // SS3 (`ESC O A`, an arrow key in application mode): one byte after it.
+ if (kind === 0x4f) return i + 3
+ // Alt with a letter or a digit is typing: the letter is read next, as is a second ESC (Alt with an arrow key). Any
+ // other pair is the terminal's.
+ return isAlnum(kind) || kind === ESC ? i + 1 : i + 2
+}
+
+/**
+ * Whether [bytes] hold anything the person typed: a byte outside every escape sequence the terminal sends itself.
+ * Escape alone, the key, is the person's: it interrupts the engine's turn.
+ */
+export function typedByPerson(bytes: Uint8Array): boolean {
+ if (bytes.length === 1 && bytes[0] === ESC) return true
+ for (let i = 0; i < bytes.length;) {
+ if (bytes[i] !== ESC) return true
+ i = sequenceEnd(bytes, i)
+ }
+ return false
+}
diff --git a/cli/src/serviceProcess.ts b/cli/src/serviceProcess.ts
index f41897560..87a025ccb 100644
--- a/cli/src/serviceProcess.ts
+++ b/cli/src/serviceProcess.ts
@@ -42,6 +42,7 @@ export const SERVICE_RUNNERS: ReadonlyMap<string, () => Promise<Runner>> = new M
['handoff', async () => (await import('./services/handoffProcess.js')).runHandoffService],
['connectors', async () => (await import('./services/connectorsProcess.js')).runConnectorsService],
['recaps', async () => (await import('./services/recapsProcess.js')).runRecapsService],
+ ['agentCleanup', async () => (await import('./services/agentCleanupProcess.js')).runAgentCleanupService],
['store', async () => (await import('./services/storeProcess.js')).runStoreService],
['messages', async () => (await import('./services/messagesProcess.js')).runMessagesService],
['sharing', async () => (await import('./services/sharingProcess.js')).runSharingService],
diff --git a/cli/src/services/AGENTS.md b/cli/src/services/AGENTS.md
index a9f24f861..d409125c7 100644
--- a/cli/src/services/AGENTS.md
+++ b/cli/src/services/AGENTS.md
@@ -54,7 +54,7 @@ that process. A process per risk, not per feature (`SERVICE_HOSTS` in `src/harne
| `engine-claude`, `engine-codex` | history, live parsers, runtime profiles, screens, submission readings, model control, question navigation and Codex's app-server connection (`src/engines/worker/`) | engine behavior outside core; snapshots and revocable model- and question-control grants | on demand |
| `search` | search | native `node:sqlite`, the index's memory | always |
| `viewers` | viewers, store | the viewer servers, the remote viewer streams and rendered surfaces, minutes-long installs | always |
-| `edge` | workspaces, usage, monitor, projects, handoff, recaps, windowNames, shell | light pure-JS services: isolated from the core, not from each other | always |
+| `edge` | workspaces, usage, monitor, projects, handoff, recaps, windowNames, shell, connectors, agentCleanup | light pure-JS services: isolated from the core, not from each other | always |
| `gateway` | gateway | network, crypto and pure-JS WebRTC: the attack surface | on demand |
| `models` | models | grid's installs, downloads and commands | on demand |
| `devices` | devices, wifi | the dials' serial ports, the fleet's lane, the voice router's worker, the Wi-Fi device | on demand |
@@ -172,6 +172,9 @@ master too old to run it starts it beside itself, still in its own process (`src
what the core reads back in line it reads from what they last said of each session.
`e2e/recapsProcess.e2e.ts` proves turns end on time with them killed, hung or slow, and
`e2e/recapsCompat.e2e.ts` that the dial and the windows hear the same cards and recaps as before.
+ The agent cleanup (`src/services/agentCleanupProcess.ts`, `src/core/agentCleanupLink.ts`) is told the same way
+ (a harness stopped, the person typing into one another agent started), and acts on the core only through what the
+ core lets it: `closeStarted` closes a harness another agent started and refuses any other.
## On demand
diff --git a/cli/src/services/agentCleanup.spec.ts b/cli/src/services/agentCleanup.spec.ts
new file mode 100644
index 000000000..1b51e3b91
--- /dev/null
+++ b/cli/src/services/agentCleanup.spec.ts
@@ -0,0 +1,321 @@
+import { mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs'
+import { tmpdir } from 'node:os'
+import { join } from 'node:path'
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
+import { emptyPorts } from '../core/api.js'
+import type { AgentCloseResult } from '../lib/closeAgentService.js'
+import type { RegisteredSession } from '../lib/registry.js'
+import { fakeCore } from '../testing/fakeCore.js'
+import { ASK_AGAIN_MS, CASCADE_RETRY_MS, CASCADE_TRIES, createAgentCleanup, IDLE_CLOSE_MS, JOINED_MAX, joinedPath, LOOK_EVERY_MS, startAgentCleanup, type AgentCleanupDeps } from './agentCleanup.js'
+
+const NOW = Date.parse('2026-10-12T12:00:00Z')
+const MIN = 60_000
+
+const row = (agentId: string, over: Partial<RegisteredSession> = {}) =>
+ ({ agentId, sessionId: `s-${agentId}`, defaultName: agentId, registeredAt: NOW - 2 * IDLE_CLOSE_MS, ...over }) as RegisteredSession
+/** A harness another agent started: `lead` here, unless said. */
+const kid = (agentId: string, over: Partial<RegisteredSession> = {}) => row(agentId, { startedBy: 'agent', parentAgentId: 'lead', ...over })
+
+let dir: string
+beforeEach(() => { dir = mkdtempSync(join(tmpdir(), 'agent-cleanup-')) })
+afterEach(() => { rmSync(dir, { recursive: true, force: true }); vi.useRealTimers(); vi.restoreAllMocks() })
+
+/** The cleanup on a fake core: [sessions] are the live agents, [quietFor] how long ago each last moved. */
+function setup(sessions: RegisteredSession[], quietFor: Record<string, number | null> = {}, over: AgentCleanupDeps & { refuse?: Record<string, AgentCloseResult> } = {}) {
+ let clock = NOW
+ const closeStarted = vi.fn(async (agentId: string, _mode: 'idle' | 'now'): Promise<AgentCloseResult> => {
+ const refused = over.refuse?.[agentId]
+ if (refused) return refused
+ sessions.splice(sessions.findIndex((s) => s.agentId === agentId), 1)
+ return { closed: true }
+ })
+ const lastActivity = vi.fn(async (agentId: string) => {
+ const quiet = quietFor[agentId]
+ return quiet === null ? null : clock - (quiet ?? 0)
+ })
+ const core = fakeCore({
+ dataDir: dir,
+ agents: {
+ live: vi.fn(() => [...sessions]),
+ resolve: vi.fn((id: string) => sessions.find((s) => s.agentId === id || s.sessionId === id)),
+ closeStarted, lastActivity,
+ },
+ })
+ const log = vi.fn()
+ let tick: (() => void) | null = null
+ const waited: number[] = []
+ const cleanup = createAgentCleanup(core, { now: () => clock, log, every: (_ms, run) => { tick = run; return () => { tick = null } },
+ wait: async (ms) => { waited.push(ms) }, ...over })
+ return { cleanup, core, closeStarted, lastActivity, log, waited, advance: (ms: number) => { clock += ms }, tick: () => tick }
+}
+
+/** Lets the closes a notice started finish. */
+const settle = () => new Promise((done) => setTimeout(done, 0))
+
+describe('closing the harnesses a closed harness started', () => {
+ it('closes them now, working or not, with one line each, and only those another agent started there', async () => {
+ // Lead itself is gone: it was closed.
+ const sessions = [kid('a'), kid('b'), kid('other', { parentAgentId: 'someone' }), row('mine', { parentAgentId: 'lead' })]
+ const { cleanup, closeStarted, log } = setup(sessions)
+ cleanup.port.closed('lead')
+ await settle()
+ expect(closeStarted.mock.calls).toEqual([['a', 'now'], ['b', 'now']])
+ expect(log.mock.calls.map(([line]) => line)).toEqual([
+ '[agent-cleanup] closed a (a): the harness that started it was closed. Its conversation is kept.',
+ '[agent-cleanup] closed b (b): the harness that started it was closed. Its conversation is kept.',
+ ])
+ expect(sessions.map((s) => s.agentId)).toEqual(['other', 'mine'])
+ })
+
+ it('closes nothing once its starter is back, and once for two notices of one stop', async () => {
+ // Its starter resumed while the notice was held for a process that was down: its harnesses are its again.
+ const back = setup([row('lead'), kid('a')])
+ back.cleanup.port.closed('lead')
+ await settle()
+ expect(back.closeStarted).not.toHaveBeenCalled()
+ // Two stops of one harness tell this twice.
+ const twice = setup([kid('a')])
+ twice.cleanup.port.closed('lead')
+ twice.cleanup.port.closed('lead')
+ await settle()
+ expect(twice.closeStarted.mock.calls).toEqual([['a', 'now']])
+ })
+
+ it('leaves one the person typed into: it is theirs now', async () => {
+ const { cleanup, closeStarted } = setup([kid('joined'), kid('left')])
+ cleanup.port.typed('s-joined')
+ cleanup.port.closed('lead')
+ await settle()
+ expect(closeStarted.mock.calls).toEqual([['left', 'now']])
+ })
+
+ it('tries one that would not close again, a moment later, and says it once the tries are spent', async () => {
+ const refuse: Record<string, AgentCloseResult> = { stuck: { error: 'CLOSE_FAILED', detail: 'Could not save its screen.' }, odd: { error: 'AGENT_CHANGED' },
+ blip: { error: 'IDENTITY_UNAVAILABLE' } }
+ const { cleanup, closeStarted, log, waited } = setup([kid('stuck', { defaultName: undefined, title: 'Fix the login test' }), kid('odd', { defaultName: undefined }), kid('blip')], {},
+ { refuse, wait: async (ms) => { waited.push(ms); delete refuse.blip } })
+ cleanup.port.closed('lead')
+ await settle()
+ // Its identity between two writes for a moment: closed on the second try.
+ expect(closeStarted.mock.calls.filter(([id]) => id === 'blip')).toHaveLength(2)
+ expect(closeStarted.mock.calls.filter(([id]) => id === 'stuck')).toHaveLength(CASCADE_TRIES)
+ expect(waited).toEqual(Array(CASCADE_TRIES - 1).fill(CASCADE_RETRY_MS))
+ expect(log.mock.calls.map(([line]) => line)).toEqual([
+ '[agent-cleanup] closed blip (blip): the harness that started it was closed. Its conversation is kept.',
+ '[agent-cleanup] could not close Fix the login test (stuck) (the harness that started it was closed): Could not save its screen.',
+ '[agent-cleanup] could not close a harness (odd) (the harness that started it was closed): AGENT_CHANGED',
+ ])
+ // Stopped meanwhile: not tried again.
+ const halted = setup([kid('stuck')], {}, { refuse: { stuck: { error: 'CLOSE_FAILED' } } })
+ halted.cleanup.port.stop()
+ await halted.cleanup.closeStartedBy('lead')
+ expect(halted.closeStarted).toHaveBeenCalledTimes(1)
+ const failing = setup([], {}, { refresh: async () => { throw new Error('the core is gone') } })
+ failing.cleanup.port.closed('lead-agent-id')
+ await settle()
+ expect(failing.log).toHaveBeenCalledWith('[agent-cleanup] could not close what lead-age started: the core is gone')
+ // The real wait: a timer that holds nothing open.
+ vi.useFakeTimers()
+ const timed = setup([kid('stuck')], {}, { refuse: { stuck: { error: 'CLOSE_FAILED' } }, wait: undefined })
+ const tries = timed.cleanup.closeStartedBy('lead')
+ await vi.advanceTimersByTimeAsync((CASCADE_TRIES - 1) * CASCADE_RETRY_MS)
+ await tries
+ expect(timed.closeStarted).toHaveBeenCalledTimes(CASCADE_TRIES)
+ })
+})
+
+describe('closing a harness another agent started once it has sat idle', () => {
+ it(`closes one idle ${IDLE_CLOSE_MS / MIN} minutes after its last answer, once its screen is idle, with one line`, async () => {
+ const sessions = [kid('done'), kid('fresh'), row('mine'), kid('unknown')]
+ const { cleanup, closeStarted, lastActivity, log, advance } = setup(sessions, { done: IDLE_CLOSE_MS, fresh: IDLE_CLOSE_MS - MIN, mine: 10 * IDLE_CLOSE_MS, unknown: null })
+ // Just started, it has seen every harness just now: nothing is closed, nor asked about, for half an hour.
+ await cleanup.look()
+ expect(closeStarted).not.toHaveBeenCalled()
+ expect(lastActivity).not.toHaveBeenCalled()
+ advance(IDLE_CLOSE_MS)
+ await cleanup.look()
+ // The person's own harness is never asked about, and one whose last move is not known is left alone.
+ expect(closeStarted.mock.calls).toEqual([['done', 'idle']])
+ expect(log.mock.calls).toEqual([['[agent-cleanup] closed done (done): another agent started it, and it has sat idle 30 minutes. Its conversation is kept.']])
+ })
+
+ it('counts a harness starting again (resumed, restored) as moving: the person resumed it to read it', async () => {
+ // A resume writes nothing a conversation's time is read from: the one resumed was closed again within a minute.
+ const resumed = kid('kid', { processIdentity: { pid: 1, startMarker: 'a', executable: 'claude' } })
+ const sessions = [resumed]
+ const { cleanup, closeStarted, advance } = setup(sessions, { kid: 3 * IDLE_CLOSE_MS })
+ await cleanup.look()
+ advance(IDLE_CLOSE_MS - MIN)
+ sessions[0] = { ...resumed, processIdentity: { pid: 2, startMarker: 'b', executable: 'claude' } }
+ advance(MIN)
+ await cleanup.look()
+ expect(closeStarted).not.toHaveBeenCalled()
+ advance(IDLE_CLOSE_MS)
+ await cleanup.look()
+ expect(closeStarted.mock.calls).toEqual([['kid', 'idle']])
+ })
+
+ it('never closes one the person typed or pasted into, even across a restart', async () => {
+ const sessions = [kid('joined')]
+ const first = setup(sessions, { joined: 10 * IDLE_CLOSE_MS })
+ first.cleanup.port.typed('joined')
+ first.cleanup.port.typed('joined')
+ await first.cleanup.look()
+ first.advance(IDLE_CLOSE_MS)
+ await first.cleanup.look()
+ expect(first.closeStarted).not.toHaveBeenCalled()
+ expect(JSON.parse(readFileSync(joinedPath(dir), 'utf8'))).toEqual({ version: 1, joined: { joined: NOW } })
+ const restarted = setup(sessions, { joined: 10 * IDLE_CLOSE_MS })
+ await restarted.cleanup.look()
+ restarted.advance(IDLE_CLOSE_MS)
+ await restarted.cleanup.look()
+ expect(restarted.closeStarted).not.toHaveBeenCalled()
+ })
+
+ it('keeps nothing for the person\'s own harness, and an id the core named that it does not list', async () => {
+ const { cleanup } = setup([row('mine')])
+ cleanup.port.typed('mine')
+ expect(() => readFileSync(joinedPath(dir), 'utf8')).toThrow()
+ cleanup.port.typed('started-elsewhere')
+ expect(JSON.parse(readFileSync(joinedPath(dir), 'utf8')).joined).toEqual({ 'started-elsewhere': NOW })
+ })
+
+ it('asks one that would not close again only after a while, quietly, and forgets it once it is gone', async () => {
+ const sessions = [kid('busy')]
+ const refuse: Record<string, AgentCloseResult> = { busy: { error: 'SESSION_NOT_IDLE', activity: 'working' } }
+ // (A refusal for its state is quiet; one for anything else is said once: see below.)
+ const { cleanup, closeStarted, log, advance } = setup(sessions, { busy: IDLE_CLOSE_MS }, { refuse })
+ await cleanup.look()
+ advance(IDLE_CLOSE_MS)
+ await cleanup.look()
+ advance(ASK_AGAIN_MS - MIN)
+ await cleanup.look()
+ expect(closeStarted).toHaveBeenCalledTimes(1)
+ advance(MIN)
+ delete refuse.busy
+ await cleanup.look()
+ expect(closeStarted).toHaveBeenCalledTimes(2)
+ expect(log.mock.calls.map(([line]) => line)).toEqual([expect.stringMatching(/^\[agent-cleanup\] closed busy \(busy\): .* idle 3\d minutes/)])
+ // Refused, then gone: nothing is kept for it, and back it is a new run, seen from then.
+ const gone = setup([kid('x')], { x: IDLE_CLOSE_MS }, { refuse: { x: { error: 'SESSION_NOT_IDLE' } } })
+ await gone.cleanup.look()
+ gone.advance(IDLE_CLOSE_MS)
+ await gone.cleanup.look()
+ ;(gone.core.agents.live as ReturnType<typeof vi.fn>).mockReturnValue([])
+ await gone.cleanup.look()
+ ;(gone.core.agents.live as ReturnType<typeof vi.fn>).mockReturnValue([kid('x')])
+ await gone.cleanup.look()
+ expect(gone.closeStarted).toHaveBeenCalledTimes(1)
+ gone.advance(IDLE_CLOSE_MS)
+ await gone.cleanup.look()
+ expect(gone.closeStarted).toHaveBeenCalledTimes(2)
+ })
+
+ it('says once a close that failed for more than its state, and again only once it has come back', async () => {
+ const sessions = [kid('broken')]
+ const { cleanup, closeStarted, log, advance, core } = setup(sessions, { broken: IDLE_CLOSE_MS }, { refuse: { broken: { error: 'CLOSE_FAILED', detail: 'Could not save its screen.' } } })
+ await cleanup.look()
+ advance(IDLE_CLOSE_MS)
+ await cleanup.look()
+ advance(ASK_AGAIN_MS)
+ await cleanup.look()
+ expect(closeStarted).toHaveBeenCalledTimes(2)
+ expect(log.mock.calls.map(([line]) => line)).toEqual([expect.stringMatching(/^\[agent-cleanup\] could not close broken \(broken\) \(another agent started it, .*\): Could not save its screen\.$/)])
+ ;(core.agents.live as ReturnType<typeof vi.fn>).mockReturnValueOnce([])
+ await cleanup.look()
+ await cleanup.look()
+ advance(IDLE_CLOSE_MS)
+ await cleanup.look()
+ expect(log).toHaveBeenCalledTimes(2)
+ })
+
+ it('looks once at a time, and not at all once stopped', async () => {
+ let release!: () => void
+ const held = new Promise<void>((done) => { release = done })
+ const sessions = [kid('a'), kid('b')]
+ const { cleanup, closeStarted, lastActivity, tick, advance } = setup(sessions, { a: IDLE_CLOSE_MS, b: IDLE_CLOSE_MS })
+ await cleanup.look()
+ advance(IDLE_CLOSE_MS)
+ lastActivity.mockImplementationOnce(async () => { await held; return NOW })
+ const first = cleanup.look()
+ await cleanup.look()
+ expect(lastActivity).toHaveBeenCalledTimes(1)
+ // Stopped halfway: the rest is left.
+ cleanup.port.stop()
+ expect(tick()).toBeNull()
+ release()
+ await first
+ expect(closeStarted.mock.calls).toEqual([['a', 'idle']])
+ await cleanup.look()
+ expect(closeStarted).toHaveBeenCalledTimes(1)
+ })
+
+ it('looks every minute, and says a look that failed', async () => {
+ const { tick, log } = setup([], {}, { refresh: async () => { throw 'no core' } })
+ tick()!()
+ await settle()
+ expect(log).toHaveBeenCalledWith('[agent-cleanup] a look failed: no core')
+ vi.useFakeTimers()
+ const core = fakeCore({ dataDir: dir, agents: { live: vi.fn(() => []) } })
+ const timed = createAgentCleanup(core, { log: vi.fn() })
+ await vi.advanceTimersByTimeAsync(LOOK_EVERY_MS)
+ expect(core.agents.live).toHaveBeenCalledTimes(1)
+ timed.port.stop()
+ await vi.advanceTimersByTimeAsync(5 * LOOK_EVERY_MS)
+ expect(core.agents.live).toHaveBeenCalledTimes(1)
+ })
+})
+
+describe('what the cleanup keeps and how it runs', () => {
+ it('remembers the newest thousand the person joined, and reads a damaged file as none', async () => {
+ const full = Object.fromEntries([['old', 1], ['bad', 'x'], ...Array.from({ length: JOINED_MAX - 1 }, (_, i) => [`k${i}`, 2 + i])])
+ writeFileSync(joinedPath(dir), JSON.stringify({ version: 1, joined: full }))
+ const { cleanup, closeStarted } = setup([kid('old', { parentAgentId: 'p' }), kid('bad', { parentAgentId: 'p' })])
+ cleanup.port.closed('p')
+ await settle()
+ expect(closeStarted.mock.calls).toEqual([['bad', 'now']])
+ cleanup.port.typed('newest')
+ const kept = JSON.parse(readFileSync(joinedPath(dir), 'utf8')).joined
+ expect(Object.keys(kept)).toHaveLength(JOINED_MAX)
+ expect(kept.old).toBeUndefined()
+ expect(kept.newest).toBe(NOW)
+ for (const damaged of ['{ not json', '{}']) {
+ writeFileSync(joinedPath(dir), damaged)
+ const fresh = setup([kid('k0', { parentAgentId: 'p' })])
+ fresh.cleanup.port.closed('p')
+ await settle()
+ expect(fresh.closeStarted).toHaveBeenCalledWith('k0', 'now')
+ }
+ })
+
+ it('says it could not keep a join, and goes on', () => {
+ const blocked = join(dir, 'blocked')
+ writeFileSync(blocked, '')
+ const core = fakeCore({ dataDir: join(blocked, 'data') })
+ const log = vi.fn()
+ createAgentCleanup(core, { log, every: () => () => {} }).port.typed('kid')
+ expect(log).toHaveBeenCalledWith(expect.stringMatching(/^\[agent-cleanup\] could not keep which harnesses the person joined: /))
+ })
+
+ it('closes nothing, and knows no time, on a core that offers neither', async () => {
+ const sessions = [kid('a')]
+ const core = fakeCore({ dataDir: dir, agents: { live: vi.fn(() => sessions) } })
+ const log = vi.spyOn(console, 'log').mockImplementation(() => {})
+ let clock = NOW
+ const cleanup = createAgentCleanup(core, { every: () => () => {}, wait: async () => {}, now: () => clock })
+ await cleanup.look()
+ clock += IDLE_CLOSE_MS
+ await cleanup.look()
+ expect(core.agents.live).toHaveBeenCalledTimes(2)
+ await cleanup.closeStartedBy('lead')
+ expect(log).toHaveBeenCalledWith('[agent-cleanup] could not close a (a) (the harness that started it was closed): UNSUPPORTED')
+ })
+
+ it('starts as the agent cleanup\'s port', () => {
+ vi.useFakeTimers()
+ const ports = emptyPorts()
+ startAgentCleanup(fakeCore({ dataDir: dir }), ports)
+ expect(ports.agentCleanup).toMatchObject({ closed: expect.any(Function), typed: expect.any(Function), stop: expect.any(Function) })
+ ports.agentCleanup!.stop()
+ })
+})
diff --git a/cli/src/services/agentCleanup.ts b/cli/src/services/agentCleanup.ts
new file mode 100644
index 000000000..77d021df9
--- /dev/null
+++ b/cli/src/services/agentCleanup.ts
@@ -0,0 +1,217 @@
+/**
+ * Cleaning up after agents (docs/design/2026-10-11-agent-client.md, "Cleaning up"). Every harness an agent starts
+ * spends the person's tokens while it works and sits in their windows while it waits, and its starter may forget it:
+ * Harness closes the ones nobody needs any more (owner, October 12: "we need to properly clean up after too").
+ *
+ * - **With its starter.** A harness closed, by the person (a window, the phone, `harness close`) or by the agent that
+ * started it, closes the harnesses it started on this machine at once, working or not (`now`): nobody reads what an
+ * orphan does, and one that keeps working spends tokens for no one. The core tells this after every stop
+ * (core/agents/lifecycle.ts `stopped`).
+ * - **When idle.** A harness another agent started that has sat `IDLE_CLOSE_MS` after its last answer is closed, once
+ * its screen says it is idle (`idle`): never one that works, waits for the person or holds a draft. Looked at every
+ * `LOOK_EVERY_MS`, by when its conversation last moved (`CoreApi.agents.lastActivity`), so a look costs no screen
+ * read until one has been quiet that long. Its process starting again counts as moving: a resume writes nothing a
+ * conversation's time is read from, and a harness the person resumed to read its answer was closed again within a
+ * minute (the review of #1234).
+ * - **Never the person's.** Neither touches a harness the person typed or pasted into from a window or the phone: it is
+ * theirs from then on, and kept so in `<data>/agent-cleanup.json` across restarts. A harness the person started is
+ * never one this may close: the core refuses (`CoreApi.agents.closeStarted`).
+ *
+ * A close keeps the conversation, as `harness close` does: the person can resume it, and `harness show` still reads its
+ * answer. One line in the log for each close.
+ */
+import { mkdirSync, readFileSync, renameSync, writeFileSync } from 'node:fs'
+import { dirname, join } from 'node:path'
+import type { AgentCleanupPort, CoreApi, CorePorts } from '../core/api.js'
+import type { AgentCloseResult } from '../lib/closeAgentService.js'
+import type { RegisteredSession } from '../lib/registry.js'
+
+/** How long a harness another agent started may sit idle after its last answer before it is closed. */
+export const IDLE_CLOSE_MS = 30 * 60_000
+/** How often the harnesses agents started are looked at. */
+export const LOOK_EVERY_MS = 60_000
+/** How long one that would not close (working after all, waiting for the person) is left before it is asked again. */
+export const ASK_AGAIN_MS = 5 * 60_000
+/** How often a starter's harnesses are tried, and how far apart: a close can fail for a moment (its conversation's
+ * identity between two writes, `IDENTITY_UNAVAILABLE`), and one left open may go on working for no one. */
+export const CASCADE_TRIES = 3
+export const CASCADE_RETRY_MS = 5_000
+/** The harnesses the person joined that are remembered, the newest. */
+export const JOINED_MAX = 1_000
+
+/** Where the harnesses the person joined are kept, in the daemon's data folder. */
+export const joinedPath = (dataDir: string): string => join(dataDir, 'agent-cleanup.json')
+
+export interface AgentCleanupDeps {
+ now?: () => number
+ log?: (line: string) => void
+ /** Before each look and each close of a starter's harnesses: a cleanup in its own process asks the core for the live
+ * agents then (services/agentCleanupProcess.ts). */
+ refresh?: () => Promise<void>
+ /** Runs [look] every [ms]; returns how to stop. */
+ every?: (ms: number, look: () => void) => () => void
+ /** Waits [ms] before a starter's harnesses are tried again. */
+ wait?: (ms: number) => Promise<void>
+}
+
+function readJoined(file: string): Map<string, number> {
+ try {
+ const saved = (JSON.parse(readFileSync(file, 'utf8')) as { joined?: Record<string, unknown> }).joined ?? {}
+ return new Map(Object.entries(saved).filter((entry): entry is [string, number] => typeof entry[1] === 'number'))
+ } catch {
+ return new Map()
+ }
+}
+
+/** One run of a harness's process: a resume, a restore or a restart is another. */
+const incarnation = (s: RegisteredSession): string => `${s.agentId}|${s.processIdentity?.pid ?? ''}|${s.processIdentity?.startMarker ?? ''}`
+
+const label = (s: RegisteredSession): string => `${s.defaultName || s.title || 'a harness'} (${s.agentId.slice(0, 8)})`
+const reason = (error: unknown): string => (error instanceof Error ? error.message : String(error))
+const STARTER_CLOSED = 'the harness that started it was closed'
+
+export function createAgentCleanup(core: CoreApi, deps: AgentCleanupDeps = {}) {
+ const now = deps.now ?? Date.now
+ const log = deps.log ?? ((line: string) => console.log(line))
+ const refresh = deps.refresh ?? (async () => {})
+ const wait = deps.wait ?? ((ms: number) => new Promise<void>((done) => { setTimeout(done, ms).unref() }))
+ const file = joinedPath(core.dataDir)
+ const joined = readJoined(file)
+ /** When one that would not close may be asked again. */
+ const askAgainAt = new Map<string, number>()
+ /** The ones whose close failed for more than their state, said once each. */
+ const said = new Set<string>()
+ /** When each run of a harness was first seen here: one seen since less than IDLE_CLOSE_MS is not closed, whatever
+ * its conversation says. A cleanup that just started has seen every run just now, so it closes nothing for its
+ * first half hour: not the harnesses the person typed into before it kept them (`joined`). */
+ const seen = new Map<string, number>()
+ /** The starters whose harnesses are being closed: two stops of one harness tell this twice. */
+ const cascading = new Set<string>()
+ let looking = false
+ let stopped = false
+
+ const save = (): void => {
+ try {
+ mkdirSync(dirname(file), { recursive: true })
+ // Written whole and renamed: a crash mid-write leaves the last list, never half of one.
+ writeFileSync(`${file}.tmp`, JSON.stringify({ version: 1, joined: Object.fromEntries(joined) }), { mode: 0o600 })
+ renameSync(`${file}.tmp`, file)
+ } catch (error) {
+ log(`[agent-cleanup] could not keep which harnesses the person joined: ${reason(error)}`)
+ }
+ }
+
+ /** Closes [s] as the core lets this; a close done is said. */
+ const close = async (s: RegisteredSession, mode: 'idle' | 'now', why: string): Promise<AgentCloseResult> => {
+ const result: AgentCloseResult = await (core.agents.closeStarted?.(s.agentId, mode) ?? Promise.resolve({ error: 'UNSUPPORTED' }))
+ if (result.closed) log(`[agent-cleanup] closed ${label(s)}: ${why}. Its conversation is kept.`)
+ return result
+ }
+ const failed = (s: RegisteredSession, why: string, result: AgentCloseResult): void =>
+ log(`[agent-cleanup] could not close ${label(s)} (${why}): ${result.detail ?? result.error}`)
+
+ /** The harnesses [parentId] started here, closed now: told after it stopped. Tried again a moment later when one
+ * would not close, and said once the tries are spent. */
+ const closeStartedBy = async (parentId: string): Promise<void> => {
+ if (cascading.has(parentId)) return
+ cascading.add(parentId)
+ try { await closeOrphans(parentId) } finally { cascading.delete(parentId) }
+ }
+ const closeOrphans = async (parentId: string): Promise<void> => {
+ for (let attempt = 1; ; attempt++) {
+ await refresh()
+ const live = core.agents.live()
+ // Its starter is back (resumed while this notice was held, or between tries): its harnesses are its again.
+ if (live.some((s) => s.agentId === parentId)) return
+ const orphans = live.filter((s) => s.parentAgentId === parentId && s.startedBy === 'agent' && !joined.has(s.agentId))
+ const left: Array<[RegisteredSession, AgentCloseResult]> = []
+ // One at a time: each close saves a screen and stops a process, and a starter may have several.
+ for (const s of orphans) {
+ const result = await close(s, 'now', STARTER_CLOSED)
+ if (!result.closed) left.push([s, result])
+ }
+ if (!left.length || stopped) return
+ if (attempt >= CASCADE_TRIES) {
+ for (const [s, result] of left) failed(s, STARTER_CLOSED, result)
+ return
+ }
+ await wait(CASCADE_RETRY_MS)
+ }
+ }
+
+ /** One look at the harnesses agents started: each quiet for IDLE_CLOSE_MS is closed if it is idle. */
+ const look = async (): Promise<void> => {
+ if (looking || stopped) return
+ looking = true
+ try {
+ await refresh()
+ const live = core.agents.live()
+ const gone = (id: string) => !live.some((s) => s.agentId === id)
+ for (const id of askAgainAt.keys()) if (gone(id)) askAgainAt.delete(id)
+ for (const id of said) if (gone(id)) said.delete(id)
+ const runs = new Set(live.map(incarnation))
+ for (const run of seen.keys()) if (!runs.has(run)) seen.delete(run)
+ for (const s of live) if (!seen.has(incarnation(s))) seen.set(incarnation(s), now())
+ for (const s of live) {
+ if (stopped) return
+ if (s.startedBy !== 'agent' || joined.has(s.agentId) || (askAgainAt.get(s.agentId) ?? 0) > now()) continue
+ // Not before this run of it has been here that long; then by when its conversation last moved, its last answer
+ // or later. Not known (an engine that keeps no dated transcript), it is left alone.
+ const since = seen.get(incarnation(s))!
+ if (now() - since < IDLE_CLOSE_MS) continue
+ const last = await (core.agents.lastActivity?.(s.agentId) ?? Promise.resolve(null))
+ if (last === null) continue
+ const quiet = now() - Math.max(last, since)
+ if (quiet < IDLE_CLOSE_MS) continue
+ const why = `another agent started it, and it has sat idle ${Math.round(quiet / 60_000)} minutes`
+ const result = await close(s, 'idle', why)
+ if (result.closed) continue
+ askAgainAt.set(s.agentId, now() + ASK_AGAIN_MS)
+ // Working after all, or waiting for the person: asked again later, nothing to say. Anything else is said once.
+ if (result.error !== 'SESSION_NOT_IDLE' && !said.has(s.agentId)) {
+ said.add(s.agentId)
+ failed(s, why, result)
+ }
+ }
+ } finally {
+ looking = false
+ }
+ }
+
+ const every = deps.every ?? ((ms: number, run: () => void) => {
+ const timer = setInterval(run, ms)
+ timer.unref()
+ return () => clearInterval(timer)
+ })
+ const unschedule = every(LOOK_EVERY_MS, () => {
+ void look().catch((error: unknown) => log(`[agent-cleanup] a look failed: ${reason(error)}`))
+ })
+
+ const port: AgentCleanupPort = {
+ closed: (agentId) => {
+ void closeStartedBy(agentId).catch((error: unknown) => log(`[agent-cleanup] could not close what ${agentId.slice(0, 8)} started: ${reason(error)}`))
+ },
+ typed: (id) => {
+ const s = core.agents.resolve(id)
+ // The person's own harness has nothing to keep; one not known here is the core's word (it sends only these).
+ if (s && s.startedBy !== 'agent') return
+ const agentId = s?.agentId ?? id
+ if (joined.has(agentId)) return
+ joined.set(agentId, now())
+ for (const oldest of joined.keys()) {
+ if (joined.size <= JOINED_MAX) break
+ joined.delete(oldest)
+ }
+ save()
+ },
+ stop: () => {
+ stopped = true
+ unschedule()
+ },
+ }
+ return { port, look, closeStartedBy }
+}
+
+export function startAgentCleanup(core: CoreApi, ports: CorePorts): void {
+ ports.agentCleanup = createAgentCleanup(core).port
+}
diff --git a/cli/src/services/agentCleanupProcess.spec.ts b/cli/src/services/agentCleanupProcess.spec.ts
new file mode 100644
index 000000000..3682b0b50
--- /dev/null
+++ b/cli/src/services/agentCleanupProcess.spec.ts
@@ -0,0 +1,104 @@
+import { mkdtempSync, rmSync } from 'node:fs'
+import { tmpdir } from 'node:os'
+import { join } from 'node:path'
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
+import type { CoreApi } from '../core/api.js'
+import type { RegisteredSession } from '../lib/registry.js'
+import { agentCleanupCoreApi, runAgentCleanupService } from './agentCleanupProcess.js'
+import { runServiceProcess, type CoreConnection, type ServiceProcessOptions } from './process.js'
+
+// The real default reaches a real socket: never in a test.
+vi.mock('./process.js', () => ({ runServiceProcess: vi.fn(() => ({ stop: vi.fn() })) }))
+
+const kid = (agentId: string) => ({ agentId, sessionId: `s-${agentId}`, startedBy: 'agent', parentAgentId: 'lead' }) as RegisteredSession
+
+let dir: string
+beforeEach(() => { dir = mkdtempSync(join(tmpdir(), 'agent-cleanup-process-')) })
+afterEach(() => { rmSync(dir, { recursive: true, force: true }); vi.clearAllMocks() })
+
+describe('the agent cleanup in its own process', () => {
+ const setup = () => {
+ let options: ServiceProcessOptions | null = null
+ let api: CoreApi | null = null
+ const port = { closed: vi.fn(), typed: vi.fn(), stop: vi.fn() }
+ let refresh: (() => Promise<void>) | undefined
+ const stop = vi.fn()
+ const service = runAgentCleanupService({
+ dataDir: dir, socketPath: '/data/daemon-1.sock', machineId: 'm', token: 't',
+ run: (given) => { options = given; return { stop } },
+ create: (core, deps) => { api = core; refresh = deps?.refresh; return { port, look: vi.fn(), closeStartedBy: vi.fn() } },
+ })
+ return { options: options!, api: api!, port, refresh: refresh!, service, stop }
+ }
+
+ it('hears from the core which harness was closed and which the person typed into, and nothing else', () => {
+ const { options, port } = setup()
+ expect(options).toMatchObject({ name: 'agentCleanup', socketPath: '/data/daemon-1.sock', machineId: 'm', token: 't', requests: {} })
+ options.onEvent!({ kind: 'closed', agentId: 'lead' })
+ options.onEvent!({ kind: 'typed', agentId: 'kid' })
+ options.onEvent!({ kind: 'typed', agentId: '' })
+ options.onEvent!({ kind: 'closed' })
+ options.onEvent!({ kind: 'other', agentId: 'x' })
+ expect(port.closed.mock.calls).toEqual([['lead']])
+ expect(port.typed.mock.calls).toEqual([['kid']])
+ })
+
+ it('reads the live agents from the core before it acts, and none when the core cannot say', async () => {
+ const { options, api, refresh } = setup()
+ // No core yet: none.
+ await refresh()
+ expect(api.agents.live()).toEqual([])
+ let answer: Record<string, unknown> = { agents: [kid('k1')] }
+ const query = vi.fn(async (_query: string, _payload?: Record<string, unknown>) => answer)
+ options.onConnected!({ query } satisfies CoreConnection)
+ await refresh()
+ expect(query).toHaveBeenLastCalledWith('live', {})
+ expect(api.agents.live().map((s) => s.agentId)).toEqual(['k1'])
+ answer = { error: 'UNKNOWN_QUERY' }
+ await refresh()
+ expect(api.agents.live()).toEqual([])
+ options.onDisconnected!()
+ await refresh()
+ expect(api.agents.live()).toEqual([])
+ })
+
+ it('asks the core when a harness last moved and to close it, and takes no answer as none and not closed', async () => {
+ const query = vi.fn(async (name: string, payload: Record<string, unknown> = {}): Promise<Record<string, unknown>> =>
+ (name === 'lastActivity' ? { at: payload.agentId === 'k1' ? 1_000 : null } : { closed: true }))
+ const ask = (name: string, payload: Record<string, unknown>) => query(name, payload)
+ const api = agentCleanupCoreApi(dir, () => [kid('k1')], ask)
+ expect(await api.agents.lastActivity!('k1')).toBe(1_000)
+ expect(await api.agents.lastActivity!('k2')).toBeNull()
+ expect(await api.agents.closeStarted!('k1', 'now')).toEqual({ closed: true })
+ expect(query).toHaveBeenLastCalledWith('closeStarted', { agentId: 'k1', mode: 'now' })
+ expect(api.agents.byAgent('k1')?.agentId).toBe('k1')
+ const gone = agentCleanupCoreApi(dir, () => [], async () => { throw new Error('the core went away') })
+ expect(await gone.agents.lastActivity!('k1')).toBeNull()
+ expect(await gone.agents.closeStarted!('k1', 'idle')).toEqual({ error: 'SERVICE_UNAVAILABLE' })
+ })
+
+ it('stops its looks with the process', () => {
+ const { service, port, stop } = setup()
+ service.stop()
+ expect(port.stop).toHaveBeenCalled()
+ expect(stop).toHaveBeenCalled()
+ })
+
+ it('runs as a real service by default, on the cleanup', async () => {
+ vi.useFakeTimers()
+ try {
+ const service = runAgentCleanupService({ dataDir: dir, socketPath: '/data/daemon-1.sock', machineId: 'm', token: 't' })
+ const options = vi.mocked(runServiceProcess).mock.calls.at(-1)![0]
+ expect(options.name).toBe('agentCleanup')
+ // The real cleanup on the process's core API: a harness closed, what it started is asked of the core and closed.
+ const query = vi.fn(async (name: string, _payload?: Record<string, unknown>) => (name === 'live' ? { agents: [kid('k1')] } : { closed: true }))
+ options.onConnected!({ query })
+ const log = vi.spyOn(console, 'log').mockImplementation(() => {})
+ options.onEvent!({ kind: 'closed', agentId: 'lead' })
+ await vi.advanceTimersByTimeAsync(0)
+ expect(query.mock.calls).toEqual([['live', {}], ['closeStarted', { agentId: 'k1', mode: 'now' }]])
+ expect(log).toHaveBeenCalledWith(expect.stringMatching(/^\[agent-cleanup\] closed a harness \(k1\): the harness that started it was closed/))
+ service.stop()
+ } finally { vi.useRealTimers() }
+ })
+})
diff --git a/cli/src/services/agentCleanupProcess.ts b/cli/src/services/agentCleanupProcess.ts
new file mode 100644
index 000000000..059805281
--- /dev/null
+++ b/cli/src/services/agentCleanupProcess.ts
@@ -0,0 +1,68 @@
+/**
+ * The agent cleanup in its own process (`harness __service agentCleanup`, in the edge host): the same cleanup as in
+ * the core's process (services/agentCleanup.ts), on the live agents the core names as each look and each close of a
+ * starter's harnesses begins (`service_query live`). The core tells it what happens (`service_event closed`, `typed`)
+ * and never waits for it. It asks the core when a harness last moved (`lastActivity`) and to close one
+ * (`closeStarted`), which the core does for a harness another agent started and nothing else (core/agentCleanupLink.ts).
+ */
+import type { CoreApi } from '../core/api.js'
+import type { AgentCloseResult } from '../lib/closeAgentService.js'
+import type { RegisteredSession } from '../lib/registry.js'
+import { createAgentCleanup } from './agentCleanup.js'
+import { runServiceProcess, type CoreConnection, type ServiceProcess } from './process.js'
+import { agentsIn, processCoreApi } from './processCoreApi.js'
+
+export interface AgentCleanupServiceOptions {
+ dataDir: string
+ socketPath: string
+ machineId: string
+ token: string
+ /** Swapped in tests for one that does not touch the real socket or process. */
+ run?: typeof runServiceProcess
+ /** Swapped in tests. */
+ create?: typeof createAgentCleanup
+}
+
+/** The core API the cleanup runs on here: the light services' (processCoreApi.ts), with the two ways it acts. */
+export function agentCleanupCoreApi(dataDir: string, live: () => RegisteredSession[], ask: (query: string, payload: Record<string, unknown>) => Promise<Record<string, unknown>>): CoreApi {
+ const api = processCoreApi(dataDir, 'agentCleanup', { live })
+ return {
+ ...api,
+ agents: {
+ ...api.agents,
+ // Not answered (the core gone, a link that dropped) is not closed: asked again at the next look.
+ closeStarted: async (agentId, mode) => await ask('closeStarted', { agentId, mode }).catch(() => ({ error: 'SERVICE_UNAVAILABLE' })) as AgentCloseResult,
+ lastActivity: async (agentId) => {
+ const answer = await ask('lastActivity', { agentId }).catch(() => null)
+ return typeof answer?.at === 'number' ? answer.at : null
+ },
+ },
+ }
+}
+
+export function runAgentCleanupService(options: AgentCleanupServiceOptions): ServiceProcess {
+ let agents: RegisteredSession[] = []
+ let core: CoreConnection | null = null
+ const ask = (query: string, payload: Record<string, unknown>): Promise<Record<string, unknown>> =>
+ core ? core.query(query, payload) : Promise.reject(new Error('no core to ask'))
+ const cleanup = (options.create ?? createAgentCleanup)(agentCleanupCoreApi(options.dataDir, () => agents, ask), {
+ // A core that cannot say leaves none: nothing is closed on a list that may be stale.
+ refresh: async () => { agents = agentsIn(await ask('live', {}).catch(() => null)) ?? [] },
+ })
+ const service = (options.run ?? runServiceProcess)({
+ name: 'agentCleanup',
+ socketPath: options.socketPath,
+ machineId: options.machineId,
+ token: options.token,
+ // The apps ask it nothing; the core tells it what happens.
+ requests: {},
+ onEvent: (payload) => {
+ if (typeof payload.agentId !== 'string' || !payload.agentId) return
+ if (payload.kind === 'closed') cleanup.port.closed(payload.agentId)
+ else if (payload.kind === 'typed') cleanup.port.typed(payload.agentId)
+ },
+ onConnected: (connection) => { core = connection },
+ onDisconnected: () => { core = null },
+ })
+ return { stop: () => { cleanup.port.stop(); return service.stop() } }
+}
diff --git a/cli/src/services/inline.spec.ts b/cli/src/services/inline.spec.ts
index 10a8f3a4c..b01aa894a 100644
--- a/cli/src/services/inline.spec.ts
+++ b/cli/src/services/inline.spec.ts
@@ -8,6 +8,7 @@ import { engineTranscriptFor } from '../engines/transcripts.js'
import { liveFor } from '../engines/live.js'
import { runtimeFor } from '../engines/runtime.js'
import { startGateway } from '../gateway/start.js'
+import { startAgentCleanup } from './agentCleanup.js'
import { startMemory } from './memory.js'
import { startConnectors } from './connectors.js'
import { startDevices } from './devices.js'
@@ -32,6 +33,6 @@ import { startWorkspaces } from './workspaces.js'
describe('the services the core runs in its own process only when they do not run in theirs', () => {
it('are their own starts, unchanged: the same services either way', () => {
- expect({ ...inline }).toEqual({ engineTranscriptFor, liveFor, runtimeFor, screenFor, submissionFor, nativeControlFor, modelControlFor, questionControlFor, startConnectors, startDevices, startHandoff, startGateway, startMemory, startModels, startMonitor, startOrchestrator, startProjects, startRecaps, startRouter, startSearch, startSharing, startShell, startStoreInCore, startMessagesInCore, startUsageInCore, startViewers, startWifi, startWindowNames, startWorkspaces })
+ expect({ ...inline }).toEqual({ engineTranscriptFor, liveFor, runtimeFor, screenFor, submissionFor, nativeControlFor, modelControlFor, questionControlFor, startAgentCleanup, startConnectors, startDevices, startHandoff, startGateway, startMemory, startModels, startMonitor, startOrchestrator, startProjects, startRecaps, startRouter, startSearch, startSharing, startShell, startStoreInCore, startMessagesInCore, startUsageInCore, startViewers, startWifi, startWindowNames, startWorkspaces })
})
})
diff --git a/cli/src/services/inline.ts b/cli/src/services/inline.ts
index 36329b644..b0a987636 100644
--- a/cli/src/services/inline.ts
+++ b/cli/src/services/inline.ts
@@ -21,6 +21,7 @@ export { nativeControlFor } from '../engines/nativeControls.js'
export { modelControlFor } from '../engines/modelControls.js'
export { runtimeFor } from '../engines/runtime.js'
export { startGateway } from '../gateway/start.js'
+export { startAgentCleanup } from './agentCleanup.js'
export { startMemory } from './memory.js'
export { startConnectors } from './connectors.js'
export { startDevices } from './devices.js'
diff --git a/docs/design/2026-10-11-agent-client.md b/docs/design/2026-10-11-agent-client.md
index 436df9d6b..ea7cdff55 100644
--- a/docs/design/2026-10-11-agent-client.md
+++ b/docs/design/2026-10-11-agent-client.md
@@ -71,9 +71,15 @@ Every Claude Code and Codex harness Harness starts is told, as it starts (owner,
```
You are running in Harness, and the person can watch you. You can hand work to other agents (Claude Code, Codex)
that run in their own harnesses: `harness new`, `harness wait`, `harness show`, `harness close` (see `harness
---help`). Only use or close harnesses you started.
+--help`). Start one only for work that can run in parallel, needs another model, or takes long; do quick work
+yourself, and close each one once you have its answer. Only use or close harnesses you started.
```
+The sentence on when to start one, and to close it, was added on October 12 (owner: "we need to be very careful about
+when to start a new agent so we don't waste user token and we need to properly clean up after too"), once agents ran
+the verbs with no approval prompts. The skill, the MCP server's instructions and harness.autonomous.ai/agents say the
+same.
+
Claude Code gets it with `--append-system-prompt`, joined into one with a grid's or a store harness's own (an engine may
keep only the last of a repeated flag), and records it with the rest of the system prompt on the first request
(`--system-prompt-snapshot`). Codex gets it in `developer_instructions`, after the person's own value and before the
@@ -232,12 +238,16 @@ the orchestrator's harnesses, and a machine whose daemon predates `startedBy`, w
## Limits
-Spawning is cheap, so runaway spawning is the risk (the Jev loops cost $60 at one caller). Every limit refuses
-with a clear error; nothing is dropped silently.
+Spawning is cheap, so runaway spawning is the risk (the Jev loops cost $60 at one caller), and every harness started
+spends the person's tokens. Every limit refuses with a clear error; nothing is dropped silently.
-- Live harnesses started by one harness: 8.
+- Live harnesses started by one harness: 4 (8 until October 12).
- Live harnesses started by agents, per machine: 40.
-- Depth: an agent-started harness may start harnesses; one started two levels down may not.
+- Depth: one level. A harness another agent started may not start harnesses itself (two levels until October 12):
+ `A harness another agent started may not start harnesses itself. Do the work yourself, or say in your answer what
+ else is needed.` Told by its line in the record, or by the core's row when the record has none: every frame says
+ `startedBy: agent` for a harness another agent started, from another machine or from outside Harness too (the
+ record is kept where the starter ran, and only `parentAgentId` was in the frame before; the review of #1234).
- `new` per harness: 20 in 10 minutes.
- An agent never answers a permission dialog, and never closes a harness it did not start.
- The person can turn agents starting agents off on a computer (Settings ▸ Agents, `harness settings
@@ -289,6 +299,49 @@ the verbs without asking.
could hold a `harness` of the agent's own, which the bare name would then reach outside the sandbox: a rule names
a command, not a file.
+## Cleaning up (October 12)
+
+Owner, October 12: "we need to properly clean up after too". What an agent starts and forgets either goes on working
+for no one or sits in the person's windows. Harness closes those itself, in a service (`services/agentCleanup.ts`, in
+the edge host; `core/agentCleanupLink.ts` is the core's side). A close is `harness close`'s: the screen saved, the
+process stopped, the conversation kept, so the person can resume it and `harness show` still reads its answer. One log
+line for each (`[agent-cleanup] closed kid (3f2a9c1e): …`).
+
+- **With its starter.** Closing a harness closes the harnesses it started on this machine (registry `parentAgentId`),
+ at once and working or not: an orphan that keeps working spends tokens for a starter that will never read it. Every
+ stop goes through one place in the core (`core/agents/lifecycle.ts` `stopAgent`): a window's or the phone's close
+ (`agent_close`), Stop Harness (`agent_delete`), `harness close` and a purge, so the apps, the CLI and an agent
+ closing its own are all covered. The core tells the cleanup once the stop is done (`AgentCleanupPort.closed`). Not
+ once the starter is live again (resumed while the notice was held for a process that was down), and once for two
+ notices of one stop. A close that fails is tried twice more, five seconds apart.
+- **When idle.** A harness another agent started (`startedBy: agent`, so an agent outside Harness's too) that has sat
+ 30 minutes (`IDLE_CLOSE_MS`) after its last answer is closed, once its screen says it is idle: never one that works,
+ waits for the person or holds a draft (`agent_close` `idle`). Looked at once a minute, by when its conversation last
+ moved (the latest dated work in its transcript), so a look reads no screen until one has been quiet that long; one
+ that would not close is asked again five minutes later. A run of its process counts as moving: a resume or a
+ restore writes nothing a conversation's time is read from, and a child the person resumed to read was closed again
+ within a minute (the review of #1234). So the cleanup closes nothing in its first half hour after it starts, and an
+ engine that keeps no dated transcript (only Claude Code and Codex declare one) is never closed for being idle.
+- **Never the person's.** Neither touches a harness the person typed or pasted into from a window or the phone (a
+ `message` frame, or terminal input that is more than what the terminal sends by itself: focus, its attributes, the
+ mouse, `lib/typedInput.ts`). It is theirs from then on, kept in `<data>/agent-cleanup.json` across restarts. A harness
+ the person started is never one the cleanup may close: the core decides (`CoreApi.agents.closeStarted` refuses any
+ other).
+
+A close of a harness already closed, the same conversation, is answered closed (`closeAgentService.ts` `closedAlready`):
+a window closing a tab closes its panes one at a time, and the starter's close had already taken the next pane's
+harness with it; that pane's close was answered `AGENT_CHANGED` and shown as "Could not confirm that it stopped".
+
+Agents are told too: `harness show` on an idle harness of theirs ends with `Close it once you have what you need:
+harness close kid`, and `harness ps` ends with `You have 2 harnesses of your own open, 1 idle. Close each once you have
+its answer: harness close <name>.` (the MCP tools name `harness_close` instead). `wait` on one Harness closed says where
+its answer still is: `kid: stopped · it was stopped; its answer is kept: harness show kid`.
+
+Not covered: harnesses an agent started on another machine when its starter closes (each machine's cleanup sees only
+its own); a cascade cut short by a restart (it is kept in memory; the idle close catches what then goes quiet); a
+harness the person drives from a tmux of their own, which no window carries; and a harness the person typed into
+before this version, which nothing recorded (closable, conversation kept, once idle half an hour after the upgrade).
+
## How it works
`cli/src/agentClient/`, a CLI module outside the core, composed only of requests the daemon already answers.
@@ -359,7 +412,10 @@ may message another only where the person turned Agent Messages on (the tab's ch
- End to end on the real daemon with the fake engines (`e2e/agentClient.e2e.ts`, 10 s): from an agent's pane,
`ps`, `new` with a task, `wait`, `show` (the header and the echoed answer), `send` both ways with no tab on
(refused since October 11), `close` refused for its own harness and for one its child did not start, then allowed
- for its child. The fake
+ for its child. Since October 12 also: a harness another agent started refused `new`, `show` and `ps` ending with
+ their hints, a window's close (`agent_close`) and the person's `harness close` each closing what the closed harness
+ started, and a harness the person sent a message to left open, with the cleanup in the edge host's process. The
+ 30-minute close is unit-tested on a fake clock (`services/agentCleanup.spec.ts`). The fake
engine now takes a first message on the command line, as both CLIs do; it used to drop it.
- Live, on the owner's Mac from this session's own pane against the running daemon: `ps` (22 harnesses, 0.8 s),
`new claude --name probe-child -- "Reply with the single word ready"` (0.4 s), `wait` (idle in 7 s), `show`
diff --git a/plugins/harness/skills/harness/SKILL.md b/plugins/harness/skills/harness/SKILL.md
index 08d0ef650..0f0330ebc 100644
--- a/plugins/harness/skills/harness/SKILL.md
+++ b/plugins/harness/skills/harness/SKILL.md
@@ -36,15 +36,16 @@ Harness is not running.
## When to start a harness
-Start one for work that:
+Every harness you start spends the person's tokens. Start one only for work that:
-1. **can run beside yours**: tests in one folder while you work in another, several tasks at once;
-2. **needs another agent's view**: a review, a second opinion from Codex;
+1. **can run in parallel with yours**: tests in one folder while you work in another, several tasks at once;
+2. **needs another model's view**: a review, a second opinion from Codex;
3. **takes long**: it keeps running after your turn or your context ends;
4. **belongs in its own folder**, or on another of the person's machines (`harness machines` lists them).
-Do small, quick work yourself, and a quick parallel search or edit inside your own task with your own
-sub-agents: cheaper and faster. Everything you start appears in Harness, where the person can watch and step in.
+Do quick work yourself, and a quick parallel search or edit inside your own task with your own sub-agents:
+cheaper and faster. Close each harness once you have its answer. Everything you start appears in Harness, where
+the person can watch and step in.
## New harness or an existing one
@@ -81,9 +82,10 @@ harness close <name> [--now] # when you are done with one
only the person answers those, so tell them), 4 when it stopped or is gone, 124 on timeout. `--timeout 0`
waits as long as it takes.
- Read the result with `show` instead of asking for a summary.
-- `close` what you started once you have what you need. It keeps the conversation; the person can resume it.
+- `close` what you started once you have its answer. It keeps the conversation; the person can resume it.
It refuses while the harness is working, unless `--now`. One closed while still starting may have no
- conversation to keep yet, and `close` says so.
+ conversation to keep yet, and `close` says so. `show` reminds you when one of yours sits idle, and `ps`
+ says how many of yours are open.
- Add `--json` to any verb for JSON. Every error says what to run next.
## Inside a harness, or outside
@@ -119,8 +121,12 @@ If Harness's MCP server (`harness mcp`) is connected, its tools are the same ver
## Limits and rules
-- At most 8 live harnesses of yours, 20 starts in 10 minutes, and a harness you start may start others, but
- the ones it starts may not. Every limit refuses with a message; do not work around it.
+- At most 4 live harnesses of yours, and 20 starts in 10 minutes. A harness another agent started may not start
+ harnesses itself: if you were started by one, do the work yourself, or say in your answer what else is needed.
+ Every limit refuses with a message; do not work around it.
+- Harness cleans up after you: closing a harness closes the harnesses it started, and one another agent started
+ is closed once it has sat idle 30 minutes after its last answer, unless the person typed into it. Its
+ conversation is kept, and `show` still reads its answer.
- A harness you start runs in its agent's auto mode, with the agent's own checks on. `--mode full` (checks and
sandbox off) is the person's to choose: from an agent it is refused.
- Never answer a harness's question or permission prompt yourself, and never type into it some other way.
diff --git a/website/public/agents/index.html b/website/public/agents/index.html
index 79cdd8d05..d0db9a39f 100644
--- a/website/public/agents/index.html
+++ b/website/public/agents/index.html
@@ -70,10 +70,10 @@
<li><strong>The person's view.</strong> Everything you start appears in Harness, where the person can watch it,
answer its questions or stop it. Your own sub-agents are invisible to them.</li>
</ul>
- <p>Start a harness for work that can run beside yours, needs another agent's view (a review, a second opinion
- from Codex), takes long, or belongs in its own folder. Do small, quick work yourself, and a quick parallel
- search or edit inside one task with your own sub-agents (Claude Code's Agent tool, Codex sub-agents): they are
- cheaper and faster.</p>
+ <p>Every harness spends the person's tokens. Start one only for work that can run in parallel with yours, needs
+ another model's view (a review, a second opinion from Codex), takes long, or belongs in its own folder. Do quick
+ work yourself, and a quick parallel search or edit inside one task with your own sub-agents (Claude Code's Agent
+ tool, Codex sub-agents): they are cheaper and faster. Close each harness once you have its answer.</p>
<h2 id="install">Install</h2>
<pre><code>curl -fsSL https://harness.autonomous.ai/cli/install.sh | bash
@@ -271,9 +271,13 @@ claude mcp add harness -- npx -y @autonomous-ai/harness-mcp</code></pre>
person.</li>
<li>Never ask another harness to publish, deploy, delete or push. Those are the person's to ask for.</li>
<li>Close what you started once you have its answer.</li>
- <li>An agent may have at most 8 live harnesses it started, and start 20 in 10 minutes; one it started may start
- more, but those may not. Each machine allows 40 live harnesses started by agents. Every limit refuses with a
- message; do not work around it.</li>
+ <li>An agent may have at most 4 live harnesses it started, and start 20 in 10 minutes. A harness another agent
+ started may not start harnesses itself: it does the work itself, or says in its answer what else is needed.
+ Each machine allows 40 live harnesses started by agents. Every limit refuses with a message; do not work around
+ it.</li>
+ <li>Harness cleans up after agents: closing a harness closes the harnesses it started, and one another agent
+ started is closed once it has sat idle 30 minutes after its last answer, unless the person typed into it. Its
+ conversation is kept, and <code>harness show</code> still reads its answer.</li>
<li>A harness you start runs in its agent's auto mode, with the agent's own checks on. <code>--mode full</code>
(checks and sandbox off) is the person's to choose: from an agent it is refused.</li>
</ul>
diff --git a/website/public/llms.txt b/website/public/llms.txt
index c5af22154..06f87ace6 100644
--- a/website/public/llms.txt
+++ b/website/public/llms.txt
@@ -26,7 +26,7 @@ Once Harness runs on a computer, Claude Code and Codex there run these commands
`harness mcp` serves the same as MCP tools over stdio (command `harness`, arguments `["mcp"]`): `harness_list`, `harness_start`, `harness_wait`, `harness_read`, `harness_close`, and `harness_send` inside a harness (teammates in a tab with Agent Messages on only). For clients that start servers with npx, `npx -y @autonomous-ai/harness-mcp` runs `harness mcp`, or, where Harness is not installed, offers one tool, `harness_setup`, that says how to install it.
-Start a harness for work that can run beside yours, needs another agent's view (a review, a second opinion from Codex), takes long, or belongs in its own folder; do small, quick work yourself. For scratch or new work, give a project name (`harness new codex review -- …`) rather than a folder of your own: Claude Code and Codex ask the person to trust a folder Harness did not make. List first, name every harness, then wait, read and close it. Give work only to harnesses you started, and close only those; starting one does not let you message it, so put everything in the task; never answer a harness's question or permission prompt, and never ask one to publish, deploy, delete or push. An agent outside Harness (Claude Code or Codex in a terminal app or editor) is recognized by its session: it gets the same limits, may close only what its session started, and cannot be messaged.
+Every harness spends the person's tokens. Start one only for work that can run in parallel with yours, needs another model's view (a review, a second opinion from Codex), takes long, or belongs in its own folder; do quick work yourself. For scratch or new work, give a project name (`harness new codex review -- …`) rather than a folder of your own: Claude Code and Codex ask the person to trust a folder Harness did not make. List first, name every harness, then wait, read it, and close it once you have its answer. An agent may have at most 4 live harnesses it started, and a harness another agent started may not start harnesses itself. Harness cleans up after agents: closing a harness closes the harnesses it started, and one another agent started is closed once it has sat idle 30 minutes after its last answer, unless the person typed into it; its conversation is kept. Give work only to harnesses you started, and close only those; starting one does not let you message it, so put everything in the task; never answer a harness's question or permission prompt, and never ask one to publish, deploy, delete or push. An agent outside Harness (Claude Code or Codex in a terminal app or editor) is recognized by its session: it gets the same limits, may close only what its session started, and cannot be messaged.
## For agents