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.
TypeScript 3d-modelingagent-frameworkagent-harnessai-agentscadclaude-codecodexcoding-agentsdeveloper-toolsesp32game-developmentmcp-servermulti-agentopen-hardwareopencodeopenharnesspcbroboticsterminaltmux
fix: modify to edit the previous recap instead of generate a new one for each agent turn
b17245b09ab10135eadee316b20f
diff --git a/cli/README.md b/cli/README.mdindex c325a90ac..6454f2b65 100644--- a/cli/README.md+++ b/cli/README.md@@ -252,6 +252,7 @@ reported as such, not described as exercised.| `ADAPTER_CLI_DIR` | `~/.harness/cli` | install dir holding the `cli.js`/`notify.mjs` the updater swaps || `LOG_FRAMES` | `false` | one log line per backend frame — type, audience and opaque ids, never a payload body. Every content-bearing frame is encrypted before it reaches the socket, so this is the only way to see what the daemon actually sent || `HARNESS_HOOK_DEADLINE_MS` | `4500` | wall-clock budget a hook gives itself before abandoning optional work. Raise it on a slow or heavily loaded machine, where the budget is spent on load rather than on the hook and the offline registry fallback silently does nothing. Clamped, never below the default |+| `SUMMARY_MODE` | `model` | how the device recap is written. `model`: a one-shot of the session's own engine, fed the previous recap, the user's ask and the answer. `local`: no model — the answer's first sentence is excerpted, instantly, but each recap stands alone || `ORI_SUMMARY_MODEL` | `deepseek/deepseek-v4-flash` | recap model for agents routed through an OpenRouter gateway (`ori claude`), which have no vendor credential to spend || `ORI_VOICE_ROUTE_MODEL` | `deepseek/deepseek-v4-flash` | same, for the voice router's classification || `ORI_CREDENTIALS_PATH` | `~/.ori/credentials.json` | where `ori login` stores its key; read only when neither the daemon env nor the agent's own process supplies one |@@ -269,10 +270,17 @@ periodic heartbeat).The per-turn **summary/recap** matches the hosted runtime: on turn end, *only while a device isconnected*, the adapter runs a disposable one-shot from the session's own CLI engine(`SUMMARY_MODEL`, `CODEX_SUMMARY_MODEL`, or `CURSOR_SUMMARY_MODEL`) → `recap ≤15 words\n\n-body ≤200`, shows a `Summarizing…` indicator while it runs, persists it per session+body ≤120 words`, shows a `Summarizing…` indicator while it runs, persists it per session(`${ADAPTER_DATA_DIR}/summaries.json`), and returns it on `project_recent` at device boot. A newerturn aborts a stale recap.+The one-shot is `recap = llm(instruct, previous recap, the user's ask, the answer)`. The previous+recap is the session's last stored summary, quoted as *continuity only* — it lets a turn whose+answer is "done, same change in the other file" recap as what was done instead of a fragment — and+the prompt forbids repeating it or reporting it as this turn's news. `SUMMARY_MODE=local` drops the+model: the answer's first prose sentence is excerpted in the same tick (no `Summarizing…`), which is+the right trade when the dial sits next to a window already showing the full text.+## Pair an Autonomous device directlyHarness discovers the OS's existing `_autonomous._tcp` advertisement. Select a discovered devicediff --git a/cli/src/cli.ts b/cli/src/cli.tsindex 7c90fdc4f..2d5ab1d2c 100644--- a/cli/src/cli.ts+++ b/cli/src/cli.ts@@ -109,11 +109,12 @@ import { RemoteRelayPool } from './lib/remoteRelay.js'import { TERMINAL_BINARY_VERSION } from './lib/terminalBinary.js'import { foldTranscript, lastTurnTextFromRawLines, lineToEvents, newTurnState, type LiveEvent, type TurnState } from './lib/normalize.js'import { AskQuestionController, pollsQuestions, QuestionWatcher } from './lib/askQuestion.js'-import { CommanderMirror } from './lib/commander.js'+import { CommanderMirror, type CommanderMirrorOpts } from './lib/commander.js'import {setSummaryPoolDeviceConnected,shutdownSummaryPool,deriveTurnSummary,+ summarizeTurnText,syncSummaryPoolSessions,} from './lib/summarize.js'import type { CableAgent } from './cable/cableSession.js'@@ -1876,6 +1877,27 @@ async function runForeground(session: AuthSession): Promise<void> {})+ // SUMMARY_MODE picks the recap writer.+ // model (default) — recap = llm(instruct, previous recap, the user's ask, the answer): a disposable+ // one-shot of the session's own engine. The previous recap is what lets "same fix, other file"+ // recap as what was done rather than as a fragment. Costs the one-shot's latency on every turn.+ // local — NO MODEL IN THE LOOP. The dial is cabled to the Mac whose window already shows this text+ // in full, so the recap is a glance and the detail is one turn of the head away; the one-shot cost+ // ~9s of the user's turn to say something they were already looking at. Instant, but every recap+ // stands alone.+ const summarizer: Pick<CommanderMirrorOpts, 'summarize' | 'summarizeIsLocal'> = env.SUMMARY_MODE === 'local'+ ? { summarize: async (text) => deriveTurnSummary(text), summarizeIsLocal: true }+ : {+ summarize: async (text, signal, userMessage, sessionId, previousRecap) => {+ const session = sessionId ? registry.bySession(sessionId) : undefined+ // Gateway agents recap through OpenRouter directly (no vendor credential to spend). The probe is+ // cached per live process, so this resolves without touching the process table again.+ const gateway = session?.gateway === 'ori' && session.processIdentity+ ? await probeGatewayRuntime(session.processIdentity)+ : undefined+ return summarizeTurnText(text, signal, userMessage, session?.engine ?? 'claude', gateway, previousRecap)+ },+ }const mirror = new CommanderMirror({send: (frame) => backend.sendCommander(frame),sendWeb: (frame) => backend.send(frame), // turn_summary_pending / turn_summary → web indicator@@ -1883,11 +1905,7 @@ async function runForeground(session: AuthSession): Promise<void> {// Live cards stream to whatever is actually rendering. The dial has one screen and it is always the// one in front of the user, so a cable session counts as active by construction.active: () => backend.hasActiveCommander() || cableWatchingLocal(),- // NO MODEL IN THE LOOP. The dial is cabled to the Mac whose window already shows this text in- // full, so the recap is a glance and the detail is one turn of the head away. A one-shot recap- // cost ~9s of the user's turn to say something they were already looking at.- summarize: async (text) => deriveTurnSummary(text),- summarizeIsLocal: true,+ ...summarizer,nameFor: (sessionId) => { const s = registry.bySession(sessionId); return s ? projectDisplayName(s) : undefined },agentIdFor: (sessionId) => registry.bySession(sessionId)?.agentId,readLastTurn: async (sessionId) => {diff --git a/cli/src/config/env.ts b/cli/src/config/env.tsindex 15c1af34a..011e97b7b 100644--- a/cli/src/config/env.ts+++ b/cli/src/config/env.ts@@ -326,6 +326,11 @@ const envSchema = z.object({AMP_SUMMARY_MODE: z.string().default('medium'),// Shared recap reasoning level for Claude and Codex. Cursor effort is part of its model identifier.SUMMARY_EFFORT: z.enum(['low', 'medium', 'high']).default('low'),+ // How the device recap is written. `model`: a disposable one-shot of the session's own engine, fed the+ // PREVIOUS recap, the user's ask and the answer — so a turn that says "same fix, other file" recaps as+ // what it did, not as a fragment; costs the one-shot's latency per turn. `local`: no model in the loop,+ // the answer's first sentence is excerpted — instant, but every recap stands alone.+ SUMMARY_MODE: z.enum(['model', 'local']).default('model'),// Model for the voice router one-shot classifier (Overview voice → pick the agent). Small/fast by default.VOICE_ROUTE_MODEL: z.string().default('haiku'),// Test override: run the recap even with no device connected (mirrors node isRecapForced()).diff --git a/cli/src/lib/commander.spec.ts b/cli/src/lib/commander.spec.tsindex 3bbdf9f79..21246e613 100644--- a/cli/src/lib/commander.spec.ts+++ b/cli/src/lib/commander.spec.ts@@ -263,6 +263,40 @@ describe('CommanderMirror recap events', () => {expect(deviceFrames.filter((f) => f.payload.kind === 'summary')).toHaveLength(1)})+ it('hands the summariser the previous turn\'s recap, and nothing on the first turn', async () => {+ // recap = llm(instruct, previous_recap, ask, answer). The previous recap is read from the store+ // BEFORE this turn's lands there — otherwise "previous" is the turn being summarised.+ const previous: Array<string | undefined> = []+ let turn = 0+ const mirror = new CommanderMirror({+ send: () => {},+ sendWeb: () => {},+ hasDevice: () => true,+ summarize: async (_text, _signal, _ask, _sessionId, previousRecap) => {+ previous.push(previousRecap)+ turn+++ return `Recap ${turn}\n\nBody ${turn}`+ },+ dataDir,+ })++ const oneTurn = (ask: string, answer: string) => mirror.ingest([+ { type: 'turn_started', payload: { userMessage: ask } },+ { type: 'text_delta', payload: { content: answer } },+ { type: 'turn_ended', payload: {} },+ ] as LiveEvent[], 'session-prev')++ oneTurn('fix the retry path', 'Fixed the retry path in client.ts.')+ await vi.runAllTimersAsync()+ await Promise.resolve()+ oneTurn('same fix in the other file', 'Done, applied the same change to server.ts.')+ await vi.runAllTimersAsync()+ await Promise.resolve()++ expect(previous).toEqual([undefined, 'Recap 1\n\nBody 1'])+ expect(mirror.recent('session-prev', 3).map((r) => r.recap)).toEqual(['Recap 2', 'Recap 1'])+ })+it('renders normalized Codex Task and child tools through the same device cards as Claude', () => {const deviceFrames: CommanderFrame[] = []const mirror = new CommanderMirror({diff --git a/cli/src/lib/commander.ts b/cli/src/lib/commander.tsindex b45590f3d..cd16bfa0f 100644--- a/cli/src/lib/commander.ts+++ b/cli/src/lib/commander.ts@@ -8,7 +8,8 @@* tool_start Task/Agent → + {kind:'agents', agents:[{text,color}]} (live sub-agent list)* subagent_finished → + {kind:'agents', …} with that row ticked off* turn_ended (device on) → {kind:'processing', text:'Summarizing…'} then, once the LLM one-shot- * returns, {kind:'summary', text:body, recap} (persisted per session)+ * returns, {kind:'summary', text:body, recap} (persisted per session,+ * and handed to the NEXT turn's summariser as its previous recap)* turn_ended (no device) → nothing (device-gated; the summary map is left untouched)* turn_ended (empty text) → {kind:'done'} (clear busy)*@@ -38,7 +39,16 @@ export interface CommanderMirrorOpts {* Gates the live turn-card STREAM; the turn-done `summary` card ignores it so a background machine still* badges. Omitted → defaults to hasDevice (single-machine firmware: attached == active, streams as before). */active?: () => boolean- summarize: (text: string, signal?: AbortSignal, userMessage?: string, sessionId?: string) => Promise<string | null>+ /** `previousRecap` is the stored `recap\n\nbody` of this session's last summarised turn, when there is+ * one — the continuity a recap of THIS turn alone cannot carry ("same fix, other file" is a fragment+ * without it). Undefined on a session's first turn. */+ summarize: (+ text: string,+ signal?: AbortSignal,+ userMessage?: string,+ sessionId?: string,+ previousRecap?: string,+ ) => Promise<string | null>/** True when `summarize` is a local derivation rather than a model call — see startSummary. */summarizeIsLocal?: boolean/** Agent display name for a sessionId — rides the summary's outer frame so a BACKGROUND machine's device@@ -613,7 +623,9 @@ export class CommanderMirror {}const ask = userMessage.replace(/\s+/g, ' ').trim().slice(0, 60)- this.trace(sessionId, `${sid} summarizing · source=${source} · textLen=${text.length} · device=${device}${this.opts.recapForce ? ' · recapForce' : ''}${ask ? ` · ask="${ask}"` : ''}`)+ // Read BEFORE the new summary is stored below, or the "previous" recap is this turn's own.+ const previousRecap = this.summaries.get(sessionId)+ this.trace(sessionId, `${sid} summarizing · source=${source} · textLen=${text.length} · device=${device}${this.opts.recapForce ? ' · recapForce' : ''}${ask ? ` · ask="${ask}"` : ''} · prev=${previousRecap ? 'yes' : 'none'}`)// Device: busy "Summarizing…" card. Web: the "Summarizing for device…" indicator (mirrors the// node's handleBrainSummaryEvent — the web ignores the summary text, only toggles the flag).//@@ -626,7 +638,7 @@ export class CommanderMirror {}this.opts- .summarize(text, ac.signal, userMessage, sessionId)+ .summarize(text, ac.signal, userMessage, sessionId, previousRecap).then((summary) => {const ms = Date.now() - t0if (ac.signal.aborted) { this.trace(sessionId, `${sid} superseded after ${ms}ms (newer turn) — dropping result`); return }diff --git a/cli/src/lib/summarize.spec.ts b/cli/src/lib/summarize.spec.tsindex 9df327911..3dddce550 100644--- a/cli/src/lib/summarize.spec.ts+++ b/cli/src/lib/summarize.spec.ts@@ -103,6 +103,48 @@ describe('gateway recap', () => {})})+describe('previous recap', () => {+ it('quotes the previous turn\'s recap as continuity context, fenced off from the content', async () => {+ mocks.runClaude.mockResolvedValue({ text: 'Applied the same change to server.ts.\n\nDone.', sessionId: null })++ await summarizeTurnText(+ 'Done, applied the same change to server.ts.',+ undefined,+ 'same fix in the other file',+ 'claude',+ undefined,+ 'Fixed the retry path in client.ts.\n\nThe retry now backs off\nand caps at five attempts.',+ )++ const prompt = mocks.runClaude.mock.calls[0][0].prompt as string+ // Flattened — the stored body's line breaks must not read as structure to preserve.+ expect(prompt).toContain('«Fixed the retry path in client.ts. The retry now backs off and caps at five attempts.»')+ expect(prompt).toContain('PREVIOUS turn')+ // Context, not content: the block sits BEFORE the turn's own message, and the message is unchanged.+ expect(prompt.indexOf('PREVIOUS turn')).toBeLessThan(prompt.indexOf('---\nDone, applied the same change'))+ })++ it('says nothing about a previous turn when there is none', async () => {+ mocks.runClaude.mockResolvedValue({ text: 'Fixed the retry path.\n\nDone.', sessionId: null })++ await summarizeTurnText('Fixed the retry path in client.ts.', undefined, 'fix the retry path', 'claude')++ const prompt = mocks.runClaude.mock.calls[0][0].prompt as string+ expect(prompt).not.toContain('PREVIOUS turn')+ })++ it('does not let the previous recap\'s language trigger the drift retry', async () => {+ // The language check compares the OUTPUT against the ask + answer only. A previous recap in another+ // script is context the model was told to ignore for language, so it must not count as a mismatch.+ mocks.runClaude.mockResolvedValue({ text: 'Deploy finished.\n\nThe service is healthy.', sessionId: null })++ await summarizeTurnText('Deploy finished; service healthy.', undefined, 'deploy status?', 'claude', undefined,+ 'Đã sửa đường retry.\n\nRetry giờ lùi dần và dừng ở năm lần.')++ expect(mocks.runClaude).toHaveBeenCalledTimes(1)+ })+})+describe('Cursor recap', () => {it('uses a Cursor one-shot with the configured Cursor model', async () => {mocks.runCursor.mockResolvedValue({diff --git a/cli/src/lib/summarize.ts b/cli/src/lib/summarize.tsindex ffe09abd3..e53842b14 100644--- a/cli/src/lib/summarize.ts+++ b/cli/src/lib/summarize.ts@@ -40,6 +40,9 @@ import {const SUMMARY_MAX_WORDS = 120const RECAP_MAX_WORDS = 15 // ~12-15 words; ONE complete sentence (20 overflowed the device tile).const MAX_INPUT_CHARS = 50_000+// The previous turn's `recap\n\nbody` is ≤ 135 words by construction; this is a guard against a+// stored summary from an older build, not a budget anyone is expected to hit.+const PREVIOUS_RECAP_MAX_CHARS = 1_000const SUMMARY_SCRATCH = join(env.ADAPTER_DATA_DIR, 'summary-scratch')function ensureSummaryScratch(): string {@@ -304,17 +307,38 @@ function firstProseLine(stripped: string): string {/** Longer than this and a line is saying something, not naming a section. */const LABEL_MAX_CHARS = 24+/**+ * `recap = llm(instruct, previousRecap, userMessage, text)`.+ *+ * `previousRecap` is the stored summary of this agent's LAST turn. It is context, not content: it lets+ * a turn whose answer is "Done — same change in the other two files" recap as what was done, instead of+ * a fragment that names nothing. The prompt fences it off so it can only resolve references, never be+ * reported as this turn's news. Absent on a session's first turn.+ */export async function summarizeTurnText(text: string,signal?: AbortSignal,userMessage?: string,engine: AgentEngine = 'claude',gateway?: GatewayRuntime,+ previousRecap?: string,): Promise<string | null> {let last = (text || '').trim()if (!last) return nullif (last.length > MAX_INPUT_CHARS) last = last.slice(-MAX_INPUT_CHARS)+ // Continuity only. Flattened — the stored body carries line breaks the model would otherwise+ // read as structure to preserve — and capped so an odd stored value cannot crowd out the turn.+ const prev = (previousRecap || '').replace(/\s+/g, ' ').trim().slice(0, PREVIOUS_RECAP_MAX_CHARS)+ const prevBlock = prev+ ? `For continuity ONLY, the recap of this assistant's PREVIOUS turn was: «${prev}». Use it to ` ++ `resolve references in this turn ("it", "that one", "the same fix", "as before", "the other file") ` ++ `and to keep names and terms consistent — nothing more. Your recap describes THIS turn's message ` ++ `alone: do NOT repeat, restate or merge in the previous recap, and do NOT report anything from it ` ++ `as if it happened now. If this turn's message is complete on its own, ignore the previous recap ` ++ `entirely.\n\n`+ : ''+// If we know the user's request for this turn, tell the recap to LEAD with the direct answer to it// (asked a price → give the price), not a generic characterization of the topic — the answer text// still comes from the assistant's message. Whitespace-flattened + capped so a long paste can't@@ -338,7 +362,9 @@ export async function summarizeTurnText(`turn and the assistant message between the --- markers below. If both are in English, output English. ` +`If they use another language, output that language. If they mix languages, preserve that mix naturally. ` +`Never switch to a language that does not appear in the user's request or the assistant message. Ignore ` +- `previous conversation, account locale, environment locale, and the language of these instructions.\n\n` ++ `previous conversation (including the previous recap, if one is quoted below), account locale, ` ++ `environment locale, and the language of these instructions.\n\n` ++ prevBlock +askBlock +`Between the --- markers below is a message the assistant already sent to the user. Re-voice its ` +`CONTENT back to the user, in the FIRST PERSON as that same assistant (its own "I"). You are ONLY ` +@@ -423,8 +449,8 @@ export async function summarizeTurnText(const effort = engine === 'cursor' ? 'model-defined' : env.SUMMARY_EFFORTconsole.log(gatewayKey- ? `[recap] openrouter ${engine} · model=${model} · inputChars=${last.length}${ask ? ' · withAsk' : ''}`- : `[recap] one-shot ${engine} · model=${model} · effort=${effort} · inputChars=${last.length}${ask ? ' · withAsk' : ''}`,+ ? `[recap] openrouter ${engine} · model=${model} · inputChars=${last.length}${ask ? ' · withAsk' : ''}${prev ? ' · withPrev' : ''}`+ : `[recap] one-shot ${engine} · model=${model} · effort=${effort} · inputChars=${last.length}${ask ? ' · withAsk' : ''}${prev ? ' · withPrev' : ''}`,)const runEngine = engine === 'claude'? runClaudeOneShot