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.

Route offers real Codex models, not 'default default' (#3045)

harness-for-autonomous[bot]a793554721e2
diff --git a/backend/integration/superTabs.test.ts b/backend/integration/superTabs.test.ts
index cdb4022a2..500e1f1fd 100644
--- a/backend/integration/superTabs.test.ts
+++ b/backend/integration/superTabs.test.ts
@@ -2459,7 +2459,9 @@ describe('Super Tabs on a real Mongo replica set', () => {
const data = read.json().data
// An older Desktop shows the read's warnings atop its routing dialog, and reads no save's `note`: they say where routing went.
expect(data).toMatchObject({ policy: DEFAULT_ROUTING_POLICY, warnings: [IGNORED_NOTE], overridden: [], override: null, account: DEFAULT_ROUTING_POLICY })
- expect(data.catalog.models).toHaveLength(7)
+ // Six rows since task #2972: Codex has Sol and Astra by name, and no `default default` row.
+ expect(data.catalog.models).toHaveLength(6)
+ expect(data.catalog.models.filter((m: { model: string; effort: string }) => m.model === 'default' || m.effort === 'default')).toEqual([])
expect(data.catalog.grid).toEqual({ strength: 1, models: [{ model: 'kimi k3', strength: 2 }, { model: 'deepseek v4.1', strength: 2 }] })
expect(data.catalog.prices.map((p: { model: string }) => p.model)).toContain('claude-opus-5-5')
expect(data.catalog.priors.models.length).toBeGreaterThan(0)
diff --git a/backend/src/superTabs/catalog.test.ts b/backend/src/superTabs/catalog.test.ts
index 06ceb5cb0..a0009fad8 100644
--- a/backend/src/superTabs/catalog.test.ts
+++ b/backend/src/superTabs/catalog.test.ts
@@ -11,14 +11,18 @@ describe('the model catalog', () => {
expect(MODEL_CATALOG.models).toEqual([
{ engine: 'claude', model: 'opus', effort: 'high', strength: 3, cost: 3, offered: true },
{ engine: 'claude', model: 'opus', effort: 'xhigh', strength: 3, cost: 4, offered: true },
- { engine: 'codex', model: 'default', effort: 'default', strength: 3, cost: 2, offered: true },
- { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true },
+ { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true, engineDefault: true },
{ engine: 'codex', model: 'gpt-6-astra', effort: 'xhigh', strength: 3, cost: 5, offered: true },
{ engine: 'claude', model: 'sonnet', effort: 'high', strength: 2, cost: 2, offered: true },
{ engine: 'claude', model: 'haiku', effort: 'auto', strength: 1, cost: 1, offered: false, note: NOT_OFFERED },
])
// Ken, 2026-11-02: small work runs on Sonnet or Sol, nothing below high effort; Haiku stays off.
- expect(MODEL_CATALOG.models.filter(m => m.offered).every(m => m.strength >= 2 && ['high', 'xhigh', 'default'].includes(m.effort))).toBe(true)
+ expect(MODEL_CATALOG.models.filter(m => m.offered).every(m => m.strength >= 2 && ['high', 'xhigh'].includes(m.effort))).toBe(true)
+ // Task #2972: the route offers the models the catalog rates, never an engine's own `default`, which ran whatever the account's CLI chose.
+ expect(MODEL_CATALOG.models.filter(m => m.model === 'default' || m.effort === 'default')).toEqual([])
+ expect(MODEL_CATALOG.models.filter(m => m.engineDefault).map(m => `${m.engine} ${m.model} ${m.effort}`)).toEqual(['codex gpt-6.1-sol high'])
+ // gpt-6-luna has no rating or benchmark, so it is not a row: the route offers it only when a person names it, unrated.
+ expect(MODEL_CATALOG.models.some(m => m.model.includes('luna'))).toBe(false)
expect(MODEL_CATALOG.grid).toEqual({ strength: 1, models: [{ model: 'kimi k3', strength: 2 }, { model: 'deepseek v4.1', strength: 2 }] })
})
diff --git a/backend/src/superTabs/catalog.ts b/backend/src/superTabs/catalog.ts
index eba09f62e..8842254b0 100644
--- a/backend/src/superTabs/catalog.ts
+++ b/backend/src/superTabs/catalog.ts
@@ -11,9 +11,10 @@ import type { PriorsTable } from './benchmarkPriors/priors.js'
import { PRICES, type ModelPrice } from './costs/prices.js'
/** One model at one effort. `offered`: the route offers it when nobody names a model; a row not offered is still rated, for
- * a person who names it. `cost` ranks rows of equal strength (lower first); it is not a price. `default` keeps the engine's
- * own configured model or effort. */
-export interface CatalogModel { engine: 'claude' | 'codex'; model: string; effort: string; strength: 1 | 2 | 3; cost: number; offered: boolean; note?: string }
+ * a person who names it. `cost` ranks rows of equal strength (lower first); it is not a price. `engineDefault` marks the row a
+ * start runs when it asks the engine for `default`: the catalog names the model, never the account's own CLI setting (task
+ * #2972: `default default` ran gpt-6-luna on one owner's Mac, which no row rates). */
+export interface CatalogModel { engine: 'claude' | 'codex'; model: string; effort: string; strength: 1 | 2 | 3; cost: number; offered: boolean; engineDefault?: true; note?: string }
/** How strong a model served on the Swarm's grid counts: the first named row whose every word is in its id or label, else
* `strength` (the CLI's route, superTabs/route.ts gridStrength). */
export interface CatalogGrid { strength: number; models: ReadonlyArray<{ model: string; strength: number }> }
@@ -26,10 +27,11 @@ export const MODEL_CATALOG: { models: readonly CatalogModel[]; grid: CatalogGrid
// Opus 5.5 at xhigh, as the coordinator runs (BUILT_IN_ROLE_DEFAULTS.lead): the same price per token as at high, more
// thinking per task, so dearer; at equal strength the route keeps choosing high unless the instructions ask for more.
{ engine: 'claude', model: 'opus', effort: 'xhigh', strength: 3, cost: 4, offered: true },
- { engine: 'codex', model: 'default', effort: 'default', strength: 3, cost: 2, offered: true },
- // Codex's own default model, named: GPT-6.1 Sol scores 58.18 on Terminal-Bench 4.0 in Codex (benchmarkPriors/data.ts) and
- // costs Sonnet's API rate (costs/prices.ts).
- { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true },
+ // Codex's default model, named: GPT-6.1 Sol scores 58.18 on Terminal-Bench 4.0 in Codex (benchmarkPriors/data.ts) and
+ // costs Sonnet's API rate (costs/prices.ts). There is no `default default` row: 95% of Codex candidates were that row and
+ // 67 of 71 Codex picks (Ken, 2026-11-04), and it ran whatever the account's CLI chose. A start that asks Codex for `default`
+ // runs this row (commands.ts agent.start); gpt-6-luna, which an account's CLI chose, has no rating and so no row.
+ { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true, engineDefault: true },
// Big builds run on Opus high or Astra xhigh (the owners' standing decision). Its API rate is five times Sol's and 2.5
// times Opus's, so it costs 5, after Opus at xhigh: at equal strength the route keeps choosing Opus high first.
{ engine: 'codex', model: 'gpt-6-astra', effort: 'xhigh', strength: 3, cost: 5, offered: true },
@@ -45,6 +47,12 @@ export const MODEL_CATALOG: { models: readonly CatalogModel[]; grid: CatalogGrid
/** What the tab's routing-policy read serves under `catalog`. */
export interface CatalogRead { models: readonly CatalogModel[]; grid: CatalogGrid; priors: PriorsTable; prices: readonly ModelPrice[] }
+/** The model and effort a start runs when it asks [engine] for `default`, or null when the catalog names none (Claude). */
+export function namedDefault(engine: string): { model: string; effort: string } | null {
+ const row = MODEL_CATALOG.models.find(m => m.engine === engine && m.engineDefault)
+ return row ? { model: row.model, effort: row.effort } : null
+}
+
export function catalogRead(): CatalogRead {
return { models: MODEL_CATALOG.models, grid: MODEL_CATALOG.grid, priors: BENCHMARK_PRIORS, prices: PRICES }
}
diff --git a/backend/src/superTabs/commands.test.ts b/backend/src/superTabs/commands.test.ts
index a4971460b..c3957e1af 100644
--- a/backend/src/superTabs/commands.test.ts
+++ b/backend/src/superTabs/commands.test.ts
@@ -849,6 +849,48 @@ it('starts only in the chosen conversation, captures authority, and never assign
expect(() => f.run({ type: 'agent.start', request, ...via }, actor, { contribution, ...decisionIn(group) })).toThrow('paused')
})
+// Task #2972 (Ken, 2026-11-04): 67 of 71 Codex picks were `default default`, which ran whatever the account's CLI defaulted to
+// (gpt-6-luna on one Mac owner's account). A start that asks Codex for `default` runs the catalog's named default instead.
+describe('a Codex start that asks for default', () => {
+ const stored = (patch: Record<string, unknown>, withVia = false) => {
+ const f = fixture(), lead = f.add('lead', 'owner', 'lead', true), group = f.group()
+ const actor: Actor = { kind: 'agent', agentId: lead, machineOwnerId: 'host-owner' }
+ const contribution = { machineOwnerId: 'target-owner', contributorId: 'owner', hostGrant: 'grant-1' }
+ const request = { conversationId: group, machineId: 'host', engine: 'codex', name: 'Builder', cwd: '/repo', worktree: true,
+ requirements: { memoryBytes: 100, diskBytes: 200, maxCpuPercent: 80, tools: ['git'], devices: [], quotaRequired: false }, ...patch }
+ f.run({ type: 'agent.start', request, ...(withVia ? { routeDecisionId: 'd1' } : via) } as Command, actor, { contribution, ...decisionIn(group,
+ { candidates: [candidate('c0', { machineId: 'host', engine: 'codex', model: 'default', effort: 'default' }), candidate('c1', { machineId: 'host', engine: 'codex', model: 'gpt-6.1-sol', effort: 'high' })] }) })
+ const { model, effort } = f.state.starts![0]!.request
+ return { model, effort, link: f.state.starts![0] }
+ }
+ it('runs the catalog\'s named default, Sol at high', () => {
+ expect(stored({ model: 'default', effort: 'default' })).toMatchObject({ model: 'gpt-6.1-sol', effort: 'high' })
+ expect(stored({ model: 'default' })).toMatchObject({ model: 'gpt-6.1-sol', effort: 'high' })
+ })
+ it('keeps an effort a person named', () => {
+ expect(stored({ model: 'default', effort: 'xhigh' })).toMatchObject({ model: 'gpt-6.1-sol', effort: 'xhigh' })
+ })
+ it('leaves a named model, a grid model, an unnamed model and Claude as they were', () => {
+ expect(stored({ model: 'gpt-6-astra', effort: 'xhigh' })).toMatchObject({ model: 'gpt-6-astra', effort: 'xhigh' })
+ expect(stored({ model: 'gpt-6-luna', effort: 'high' })).toMatchObject({ model: 'gpt-6-luna', effort: 'high' })
+ expect(stored({ model: 'default', effort: 'default', gridModel: 'kimi', gridName: 'grid' })).toMatchObject({ model: 'default', effort: 'default' })
+ expect(stored({})).toEqual(expect.objectContaining({ model: undefined, effort: undefined }))
+ expect(stored({ engine: 'claude', model: 'default', effort: 'default' })).toMatchObject({ model: 'default', effort: 'default' })
+ })
+ it('still links the start to the route candidate that was asked for, `default default` from an older route or Sol at high', () => {
+ for (const asked of [{ model: 'default', effort: 'default' }, { model: 'gpt-6.1-sol', effort: 'high' }]) {
+ const f = fixture(), lead = f.add('lead', 'owner', 'lead', true), group = f.group()
+ const actor: Actor = { kind: 'agent', agentId: lead, machineOwnerId: 'host-owner' }
+ const contribution = { machineOwnerId: 'target-owner', contributorId: 'owner', hostGrant: 'grant-1' }
+ const request = { conversationId: group, machineId: 'host', engine: 'codex', name: 'Builder', cwd: '/repo', worktree: true, ...asked,
+ requirements: { memoryBytes: 100, diskBytes: 200, maxCpuPercent: 80, tools: ['git'], devices: [], quotaRequired: false } }
+ const routing = decisionIn(group, { candidates: [candidate('c0', { machineId: 'host', engine: 'codex', ...asked })], pick: undefined, via: 'instructions', p: null })
+ const result = f.run({ type: 'agent.start', request, routeDecisionId: 'd1', overrideReason: 'Routing rule.' } as Command, actor, { contribution, ...routing })
+ expect(result.messages.map(m => m.text).join('\n')).toMatch(/Routed to Title c0/)
+ }
+ })
+})
+
it.each(['opencode', 'kilo'])('refuses to dispatch a stored legacy %s start before native launch', engine => {
const f = fixture(), lead = f.add('lead', 'owner', 'lead', true), group = f.group()
const actor: Actor = { kind: 'agent', agentId: lead, machineOwnerId: 'host-owner' }
diff --git a/backend/src/superTabs/commands.ts b/backend/src/superTabs/commands.ts
index 91831605c..57bf82ddf 100644
--- a/backend/src/superTabs/commands.ts
+++ b/backend/src/superTabs/commands.ts
@@ -45,6 +45,7 @@ import { assignedNotice, tellBuddies } from './buddyNews.js'
import { bindPullRequestCards, bindTaskCard, pullRequestCard, recordCardMerged } from './cards.js'
import { avatarChoice, setAgentAvatar, setMemberAvatar, setStatus, setTimeZone, statusSchema, type AvatarUpload } from './profile.js'
import { cancelledTags, cancelReason, switchExpired, unfinishedWork } from './tidy.js'
+import { namedDefault } from './catalog.js'
import { BUILT_IN_ROLE_DEFAULTS, agentProjectId, audienceOf, groupRepositoryUrl, canReadAgent, canReadProject, deleteProject, markConfidential, onTeamAccount, requireSingleRegistration, groupProject, inviteLanding, ledProject, manageProject, MAX_ROLES, mutableProjects, placeRules, projectAgent, projectMembers, projectOf, projectsOf, readAgent, readConversation, readProject, requestedByLine, sameProject, sharedWithLeads, withRoleDefaults, type SwarmProject, PROJECT_STATUS_MAX } from './projects.js'
import { recordClosing, sessionClosed } from './closedAgents.js'
import { declinedWith, declinedWithout } from './sharing.js'
@@ -2386,7 +2387,13 @@ function applyCommandOnce(original: SuperTabState, actor: Actor, input: Command,
// An older client starts in a kind or a custom role, kept on the start for older readers and read by no command: the
// role name it means rides on the start to the agent (instructions.ts legacyRoleName).
const roleName = command.request.roleName ?? legacyRoleName(home, command.request)
- const named = roleName ? { ...command.request, roleName } : command.request
+ // A start that asks Codex for `default` runs the catalog's named default, never whatever the account's CLI defaults to
+ // (task #2972). A named effort is kept; none, or `default`, takes the row's.
+ const catalogDefault = command.request.model === 'default' && !command.request.gridModel ? namedDefault(command.request.engine) : null
+ const asked = catalogDefault
+ ? { ...command.request, model: catalogDefault.model, effort: !command.request.effort || command.request.effort === 'default' ? catalogDefault.effort : command.request.effort }
+ : command.request
+ const named = roleName ? { ...asked, roleName } : asked
// The Lead's account has machines the Swarm never prepared (no clone of the project), so a Lead may start only where the
// Swarm already lists a machine: a contributed one, or one an agent of the Swarm runs on. A person's start may name any
// machine of theirs, because the first Lead is created before anything is listed.
diff --git a/backend/src/superTabs/routingPolicy.test.ts b/backend/src/superTabs/routingPolicy.test.ts
index 36a481763..deec6d6ed 100644
--- a/backend/src/superTabs/routingPolicy.test.ts
+++ b/backend/src/superTabs/routingPolicy.test.ts
@@ -2,6 +2,10 @@ import { describe, expect, it } from 'vitest'
import { DEFAULT_ROUTING_POLICY, effectivePolicy, priorFor, routingOverrideSaveSchema, routingOverrideSchema, routingPolicySaveSchema, routingPolicySchema } from './routingPolicy.js'
import type { PriorsTable } from './benchmarkPriors/priors.js'
+/** The `default default` Codex row policies saved before task #2972 hold; the catalog no longer has it, and a saved policy's Codex
+ * switch is still read off it (effectivePolicy). */
+const SAVED_CODEX_DEFAULT = { engine: 'codex' as const, model: 'default', effort: 'default', quality: 3, cost: 2, enabled: true }
+
describe('routing policy', () => {
it('ships a valid default that needs no setup', () => {
expect(routingPolicySchema.parse(DEFAULT_ROUTING_POLICY)).toEqual(DEFAULT_ROUTING_POLICY)
@@ -81,7 +85,7 @@ describe('routing policy', () => {
const opusCosts = DEFAULT_ROUTING_POLICY.models.filter(m => m.model === 'opus').map(m => m.cost)
expect(astra.cost).toBeGreaterThan(Math.max(...opusCosts))
// A policy saved before the row reads it.
- const before = { ...DEFAULT_ROUTING_POLICY, models: DEFAULT_ROUTING_POLICY.models.filter(m => m.model !== 'gpt-6-astra') }
+ const before = { ...DEFAULT_ROUTING_POLICY, models: [...DEFAULT_ROUTING_POLICY.models.filter(m => m.model !== 'gpt-6-astra'), SAVED_CODEX_DEFAULT] }
expect(effectivePolicy(before, null).policy.models.filter(m => m.model === 'gpt-6-astra')).toEqual([astra])
// An owner's own Astra row wins over the built-in one, on or off, and is never doubled (review of #2271).
const owned = { ...before, models: [...before.models, { ...astra, effort: 'high', quality: 2, cost: 9, enabled: false }] }
@@ -90,7 +94,7 @@ describe('routing policy', () => {
describe('a built-in row added on read (review of #2271 at 1d9499646)', () => {
const sol = DEFAULT_ROUTING_POLICY.models.find(m => m.model === 'gpt-6.1-sol')!
- const codexDefault = DEFAULT_ROUTING_POLICY.models.find(m => m.engine === 'codex' && m.model === 'default')!
+ const codexDefault = SAVED_CODEX_DEFAULT
const claudeRows = DEFAULT_ROUTING_POLICY.models.filter(m => m.engine === 'claude')
const solOf = (models: Array<{ engine: string; model: string; enabled: boolean }>) => models.find(m => m.model === 'gpt-6.1-sol')
diff --git a/cli/src/superTabs/__snapshots__/route.answers.json b/cli/src/superTabs/__snapshots__/route.answers.json
index a50061c9f..6f11a46f7 100644
--- a/cli/src/superTabs/__snapshots__/route.answers.json
+++ b/cli/src/superTabs/__snapshots__/route.answers.json
@@ -457,8 +457,8 @@
"machineId": "b",
"machineName": "b",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
"quota": {
@@ -495,8 +495,8 @@
"account": "B"
},
"id": "c1",
- "title": "New agent · Codex default default · b",
- "label": "New agent · Codex default default · strength 3, cost 2 · short: needs about 50% of the 5-hour window (5% left, resets in 4h, otherwise unused); weekly 8% of 60% · b · new · quota 5-hour 5% left (resets in 4h), weekly 60% left (resets in 3 days)"
+ "title": "New agent · Codex gpt-6.1-sol high · b",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · short: needs about 50% of the 5-hour window (5% left, resets in 4h, otherwise unused); weekly 8% of 60% · b · new · quota 5-hour 5% left (resets in 4h), weekly 60% left (resets in 3 days)"
},
{
"kind": "new",
@@ -590,12 +590,12 @@
"kind": "new",
"machineId": "b",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
- "title": "New agent · Codex default default · b",
- "label": "New agent · Codex default default · strength 3, cost 2 · short: needs about 50% of the 5-hour window (5% left, resets in 4h, otherwise unused); weekly 8% of 60% · b · new · quota 5-hour 5% left (resets in 4h), weekly 60% left (resets in 3 days)",
+ "title": "New agent · Codex gpt-6.1-sol high · b",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · short: needs about 50% of the 5-hour window (5% left, resets in 4h, otherwise unused); weekly 8% of 60% · b · new · quota 5-hour 5% left (resets in 4h), weekly 60% left (resets in 3 days)",
"fit": {
"status": "short",
"coverage": 0.1,
@@ -742,8 +742,8 @@
"machineId": "c",
"machineName": "c",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
"quota": {
@@ -775,8 +775,8 @@
"account": "C"
},
"id": "c2",
- "title": "New agent · Codex default default · c",
- "label": "New agent · Codex default default · strength 3, cost 2 · fits: 5-hour window unknown; weekly 1% of 30% · c · new · quota weekly 30% left (resets in 6 days)"
+ "title": "New agent · Codex gpt-6.1-sol high · c",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · fits: 5-hour window unknown; weekly 1% of 30% · c · new · quota weekly 30% left (resets in 6 days)"
}
],
"excluded": [],
@@ -848,12 +848,12 @@
"kind": "new",
"machineId": "c",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
- "title": "New agent · Codex default default · c",
- "label": "New agent · Codex default default · strength 3, cost 2 · fits: 5-hour window unknown; weekly 1% of 30% · c · new · quota weekly 30% left (resets in 6 days)",
+ "title": "New agent · Codex gpt-6.1-sol high · c",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · fits: 5-hour window unknown; weekly 1% of 30% · c · new · quota weekly 30% left (resets in 6 days)",
"fit": {
"status": "fits",
"coverage": 30,
@@ -1055,8 +1055,8 @@
"machineId": "a",
"machineName": "a",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
"quota": {
@@ -1081,16 +1081,16 @@
"account": "a:codex"
},
"id": "c0",
- "title": "New agent · Codex default default · a",
- "label": "New agent · Codex default default · strength 3, cost 2 · fits: metered, no usage window to run out · a · new · quota metered"
+ "title": "New agent · Codex gpt-6.1-sol high · a",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · fits: metered, no usage window to run out · a · new · quota metered"
},
{
"kind": "new",
"machineId": "b",
"machineName": "b",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
"quota": {
@@ -1115,16 +1115,16 @@
"account": "b:codex"
},
"id": "c1",
- "title": "New agent · Codex default default · b",
- "label": "New agent · Codex default default · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · b · new · quota unknown (signed out of usage)"
+ "title": "New agent · Codex gpt-6.1-sol high · b",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · b · new · quota unknown (signed out of usage)"
},
{
"kind": "new",
"machineId": "c",
"machineName": "c",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
"quota": {
@@ -1148,16 +1148,16 @@
"account": "c:codex"
},
"id": "c2",
- "title": "New agent · Codex default default · c",
- "label": "New agent · Codex default default · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · c · new · quota 40% left, resets in 2h"
+ "title": "New agent · Codex gpt-6.1-sol high · c",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · c · new · quota 40% left, resets in 2h"
},
{
"kind": "new",
"machineId": "d",
"machineName": "d",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
"quota": {
@@ -1182,8 +1182,8 @@
"account": "d:codex"
},
"id": "c3",
- "title": "New agent · Codex default default · d",
- "label": "New agent · Codex default default · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · d · new · quota read on Team Mac, whose subscription this account spends: 40% left"
+ "title": "New agent · Codex gpt-6.1-sol high · d",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · d · new · quota read on Team Mac, whose subscription this account spends: 40% left"
}
],
"excluded": [],
@@ -1208,12 +1208,12 @@
"kind": "new",
"machineId": "a",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
- "title": "New agent · Codex default default · a",
- "label": "New agent · Codex default default · strength 3, cost 2 · fits: metered, no usage window to run out · a · new · quota metered",
+ "title": "New agent · Codex gpt-6.1-sol high · a",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · fits: metered, no usage window to run out · a · new · quota metered",
"fit": {
"status": "fits",
"coverage": null,
@@ -1231,12 +1231,12 @@
"kind": "new",
"machineId": "b",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
- "title": "New agent · Codex default default · b",
- "label": "New agent · Codex default default · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · b · new · quota unknown (signed out of usage)",
+ "title": "New agent · Codex gpt-6.1-sol high · b",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · b · new · quota unknown (signed out of usage)",
"fit": {
"status": "unknown",
"coverage": null,
@@ -1254,12 +1254,12 @@
"kind": "new",
"machineId": "c",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
- "title": "New agent · Codex default default · c",
- "label": "New agent · Codex default default · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · c · new · quota 40% left, resets in 2h",
+ "title": "New agent · Codex gpt-6.1-sol high · c",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · c · new · quota 40% left, resets in 2h",
"fit": {
"status": "unknown",
"coverage": null,
@@ -1277,12 +1277,12 @@
"kind": "new",
"machineId": "d",
"engine": "codex",
- "model": "default",
- "effort": "default",
+ "model": "gpt-6.1-sol",
+ "effort": "high",
"quality": 3,
"cost": 2,
- "title": "New agent · Codex default default · d",
- "label": "New agent · Codex default default · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · d · new · quota read on Team Mac, whose subscription this account spends: 40% left",
+ "title": "New agent · Codex gpt-6.1-sol high · d",
+ "label": "New agent · Codex gpt-6.1-sol high · strength 3, cost 2 · unknown: 5-hour window unknown; weekly unknown · d · new · quota read on Team Mac, whose subscription this account spends: 40% left",
"fit": {
"status": "unknown",
"coverage": null,
@@ -2938,7 +2938,7 @@
"machine": "a"
},
{
- "title": "New agent · Codex default default · b",
+ "title": "New agent · Codex gpt-6.1-sol high · b",
"reason": "Codex is signed out.",
"machine": "b"
}
diff --git a/cli/src/superTabs/fixtures/planState.ts b/cli/src/superTabs/fixtures/planState.ts
index 2cd3f18c4..6dbe4368b 100644
--- a/cli/src/superTabs/fixtures/planState.ts
+++ b/cli/src/superTabs/fixtures/planState.ts
@@ -5,14 +5,14 @@ export const NOW = Date.parse('2026-10-27T01:00:00Z')
/** The routing read of a backend from before the catalog: a policy alone (planCommand.spec.ts reads it as the catalog below). */
export const policy = { version: 1, floor: { low: 1, medium: 2, high: 3 }, quotaReserve: 10, notes: '', localModels: { quality: 1, enabled: true }, models: [
{ engine: 'claude', model: 'opus', effort: 'high', quality: 3, cost: 3, enabled: true },
- { engine: 'codex', model: 'default', effort: 'default', quality: 3, cost: 2, enabled: true },
+ { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', quality: 3, cost: 2, enabled: true },
{ engine: 'claude', model: 'sonnet', effort: 'medium', quality: 2, cost: 2, enabled: true },
{ engine: 'claude', model: 'haiku', effort: 'auto', quality: 1, cost: 1, enabled: false },
] }
/** The same rows as a catalog (route.ts catalogFrom). */
export const catalog: Catalog = { grid: { strength: 1, models: [] }, models: [
{ engine: 'claude', model: 'opus', effort: 'high', strength: 3, cost: 3, offered: true },
- { engine: 'codex', model: 'default', effort: 'default', strength: 3, cost: 2, offered: true },
+ { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true },
{ engine: 'claude', model: 'sonnet', effort: 'medium', strength: 2, cost: 2, offered: true },
{ engine: 'claude', model: 'haiku', effort: 'auto', strength: 1, cost: 1, offered: false },
] }
diff --git a/cli/src/superTabs/planCommand.spec.ts b/cli/src/superTabs/planCommand.spec.ts
index 599c72f86..732c1550b 100644
--- a/cli/src/superTabs/planCommand.spec.ts
+++ b/cli/src/superTabs/planCommand.spec.ts
@@ -21,7 +21,7 @@ describe('planWork', () => {
it('gives each open group its machine, engine, model and effort, start order and expected finish, against the route', async () => {
const run = vi.fn<SolverRun>(async () => answer([
- { task: 'g1', machine: 'm1', model: 'codex-default-default', start: 0, end: 60, order: 1 },
+ { task: 'g1', machine: 'm1', model: 'codex-gpt-6.1-sol-high', start: 0, end: 60, order: 1 },
{ task: 'g2', machine: 'm2', model: 'grid:deepseek/flash', start: 60, end: 180, order: 2 }]))
const s = state({ rules, machines: [machine('m1', {}, served), machine('m2')] })
const out = await planWork({ request: { groups: { g1: { difficulty: 'high', risk: 'low', size: 'medium' } } } }, scoped(s), { run, now: () => NOW, dataDir: '/unused' })
@@ -29,8 +29,8 @@ describe('planWork', () => {
expect(out).toEqual({
ok: true, status: 'OPTIMAL', makespanMinutes: 180, cost: 540, route: { makespanMinutes: 240, cost: 600 },
plan: [
- // Codex default leaves the effort to the engine; an owner's minimum effort no longer lifts it (project instructions, phase 4).
- { order: 1, groupId: 'g1', group: 'Group g1', machineId: 'm1', machine: 'm1.local', engine: 'codex', model: 'default', effort: 'default',
+ // Codex is planned on a model the catalog names, never `default default` (task #2972).
+ { order: 1, groupId: 'g1', group: 'Group g1', machineId: 'm1', machine: 'm1.local', engine: 'codex', model: 'gpt-6.1-sol', effort: 'high',
startsInMinutes: 0, finishesInMinutes: 60, finishAt: at(60) },
{ order: 2, groupId: 'g2', group: 'Group g2', machineId: 'm2', machine: 'm2.local', engine: 'claude', gridModel: 'deepseek/flash', gridName: 'swarm-g',
startsInMinutes: 60, finishesInMinutes: 180, finishAt: at(180) },
diff --git a/cli/src/superTabs/planInput.spec.ts b/cli/src/superTabs/planInput.spec.ts
index a0632e59f..f8c695b1c 100644
--- a/cli/src/superTabs/planInput.spec.ts
+++ b/cli/src/superTabs/planInput.spec.ts
@@ -162,14 +162,14 @@ describe('planSetup', () => {
expect((s.instance.tasks as Json[]).map(t => [t.id, t.floor])).toEqual([['b1', 3], ['r1', 2]])
for (const task of s.instance.tasks as Json[]) {
expect(task).toMatchObject({ independent_of: [], machines: ['m1', 'm2'],
- models: ['claude-opus-high', 'codex-default-default', 'claude-sonnet-medium', 'grid:deepseek/flash'] })
+ models: ['claude-opus-high', 'codex-gpt-6.1-sol-high', 'claude-sonnet-medium', 'grid:deepseek/flash'] })
}
// A review still waits for the work it reviews.
expect(taskOf(s, 'r1').deps).toEqual(['b1'])
expect((s.instance.machines as Json[]).find(m => m.id === 'm1')).toMatchObject({ engines: ['claude', 'codex'] })
expect((s.instance.accounts as Json[]).map(a => a.reserve)).toEqual([0, 0, 0])
// A row starts at its own effort: no minimum lifts it.
- expect(s.rows['codex-default-default']).toMatchObject({ effort: 'default' })
+ expect(s.rows['codex-gpt-6.1-sol-high']).toMatchObject({ effort: 'high' })
})
// Final review: every task had floor 1 and the solver minimizes finish time then cost, so the plan picked the cheapest
diff --git a/cli/src/superTabs/route.spec.ts b/cli/src/superTabs/route.spec.ts
index fcf8fb701..d6338b7fe 100644
--- a/cli/src/superTabs/route.spec.ts
+++ b/cli/src/superTabs/route.spec.ts
@@ -1,6 +1,6 @@
// cli/src/superTabs/route.spec.ts
import { afterEach, describe, expect, it, vi } from 'vitest'
-import { API_BUILDERS, CHOOSE_BY_INSTRUCTIONS, NONE_COVERS, catalogFrom, quotaBrief, catalogRow, choose, decide, gridStrength, LOCAL_NONE, OTHER_REPOSITORY, parseAssessment, quotaOf2, recordOf, repoMatch, repositoryKey, strengthOf,
+import { API_BUILDERS, CHOOSE_BY_INSTRUCTIONS, CODEX_DEFAULT_NOT_OFFERED, NONE_COVERS, catalogFrom, quotaBrief, catalogRow, choose, decide, gridStrength, LOCAL_NONE, OTHER_REPOSITORY, parseAssessment, quotaOf2, recordOf, repoMatch, repositoryKey, strengthOf,
type Assessment, type Catalog, type Commitment, type MemberFrame, type Probe, type RoutePlugin, type TaskFolder } from './route.js'
const object = (v: unknown) => v as Record<string, unknown>
@@ -9,6 +9,7 @@ const object = (v: unknown) => v as Record<string, unknown>
const policy: Catalog = { grid: { strength: 1, models: [] }, models: [
{ engine: 'claude', model: 'opus', effort: 'high', strength: 3, cost: 3, offered: true },
{ engine: 'codex', model: 'default', effort: 'default', strength: 3, cost: 2, offered: true },
+ { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true },
{ engine: 'claude', model: 'sonnet', effort: 'medium', strength: 2, cost: 2, offered: true },
{ engine: 'claude', model: 'haiku', effort: 'auto', strength: 1, cost: 1, offered: false },
] }
@@ -87,6 +88,33 @@ describe('quota and rates', () => {
})
})
+// Task #2972 (Ken, 2026-11-04): the route put `Codex default default` first 67 times of 71 Codex picks; it ran whatever the
+// account's CLI defaulted to. The catalog now names Sol as the engine's default (`engineDefault`) and has no `default` row.
+describe('Codex default in the catalog', () => {
+ const sol = { engine: 'codex' as const, model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true }
+ const named: Catalog = { grid: policy.grid, models: [{ engine: 'claude', model: 'opus', effort: 'high', strength: 3, cost: 3, offered: true }, { ...sol, engineDefault: true }] }
+ it('rates a member launched on Codex\'s own model as the catalog\'s named default, and a small variant not at all', () => {
+ expect(rate(named, 'codex', null, null)).toEqual({ quality: 3, cost: 2 })
+ expect(rate(named, 'codex', 'default', 'default')).toEqual({ quality: 3, cost: 2 })
+ expect(rate(named, 'codex', 'gpt-5.5', 'high', true)).toEqual({ quality: 3, cost: 2 })
+ expect(rate(named, 'codex', 'gpt-5.5-mini', 'high', true)).toBeNull()
+ expect(rate(named, 'codex', 'gpt-5.5', 'high')).toBeNull()
+ })
+ it('reads the named default from the served catalog, and never offers a `default` Codex row an older backend still serves', () => {
+ const rows = [{ engine: 'codex', model: 'default', effort: 'default', strength: 3, cost: 2, offered: true },
+ { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true, engineDefault: true },
+ { engine: 'codex', model: 'gpt-6-astra', effort: 'xhigh', strength: 3, cost: 5, offered: true }]
+ const read = catalogFrom({ catalog: { models: rows, grid: { strength: 1, models: [] } } })
+ expect(read.models.filter(m => m.offered).map(m => `${m.model} ${m.effort}`)).toEqual(['gpt-6.1-sol high', 'gpt-6-astra xhigh'])
+ expect(read.models.find(m => m.model === 'gpt-6.1-sol')).toMatchObject({ engineDefault: true })
+ // A member launched on the engine's own model is still rated, by that row.
+ expect(rate(read, 'codex', null, null, true)).toEqual({ quality: 3, cost: 2 })
+ // The same from an older backend that serves only its policy.
+ const old = catalogFrom({ policy: { models: rows.map(({ strength, offered, ...r }) => ({ ...r, quality: strength, enabled: offered })) } })
+ expect(old.models.filter(m => m.offered).map(m => `${m.model} ${m.effort}`)).toEqual(['gpt-6.1-sol high', 'gpt-6-astra xhigh'])
+ })
+})
+
describe('choose', () => {
it('returns NO_CANDIDATE with reasons when nothing can run', () => {
expect(choose([probe({ refusals: ['Machine is offline.'] })], policy, a())).toEqual({ error: 'NO_CANDIDATE',
@@ -796,7 +824,9 @@ describe('facts, not ranking', () => {
it('reads the catalog the backend serves, or one derived from an older backend\'s policy', () => {
const served = { policy: { version: 1, models: [] }, catalog: { models: catalog.models, grid: catalog.grid, priors: {}, prices: [] } }
- expect(catalogFrom(served)).toEqual(catalog)
+ // A Codex row at `default` is kept to rate a member launched on the engine's own model, but is never offered (task #2972).
+ expect(catalogFrom(served)).toEqual({ ...catalog, models: catalog.models.map(m => m.engine === 'codex' && m.model === 'default'
+ ? { ...m, offered: false, note: CODEX_DEFAULT_NOT_OFFERED } : m) })
const old = { policy: { version: 1, floor: { low: 1, medium: 2, high: 3 }, quotaReserve: 10, notes: '', models: [
{ engine: 'claude', model: 'opus', effort: 'high', quality: 3, cost: 3, enabled: true },
{ engine: 'claude', model: 'sonnet', effort: 'medium', quality: 2, cost: 2, enabled: false },
@@ -829,7 +859,8 @@ describe('facts, not ranking', () => {
* record must stay byte for byte the same. */
describe('route answers without plug-ins', () => {
const OPUS = { model: 'opus', effort: 'high', quality: 3, cost: 3 } as const
- const CODEX = { engine: 'codex', model: 'default', effort: 'default', quality: 3, cost: 2 } as const
+ // Codex as the route offers it since task #2972: the catalog's named model, never `default default`.
+ const CODEX = { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', quality: 3, cost: 2 } as const
const on = (machine: string, row: Partial<Probe>, quota: Probe['quota'] = windows([80, 3 * H], [90, 5 * D], machine)) =>
probe({ machineId: machine, machineName: machine, ...row, quota })
const frame: MemberFrame = { title: 'Fix the dial', project: 'openharness', cwd: '/r', root: '/r', remote: 'github.com/o/r', branch: 'fix/dial', updatedAt: NOW - 20 * M }
diff --git a/cli/src/superTabs/route.ts b/cli/src/superTabs/route.ts
index 324de44ba..b8ae689cd 100644
--- a/cli/src/superTabs/route.ts
+++ b/cli/src/superTabs/route.ts
@@ -25,7 +25,10 @@ export interface Needs { tools: string[]; devices: string[]; cwd?: string; heavy
export interface Assessment { difficulty: Level; risk: Level; size: Size; why: string; kind?: string; local?: boolean; model?: string; needs: Needs }
/** A model of the reviewed catalog (backend superTabs/catalog.ts): how strong it is, what it costs, and whether it is offered
* when nobody names it. */
-export interface CatalogModel { engine: 'claude' | 'codex'; model: string; effort: string; strength: number; cost: number; offered: boolean; note?: string }
+export interface CatalogModel { engine: 'claude' | 'codex'; model: string; effort: string; strength: number; cost: number; offered: boolean; engineDefault?: true; note?: string }
+/** Why a Codex row at `default` model or effort is never offered. A backend from before task #2972 still serves one: it ran whatever
+ * the account's CLI defaulted to (gpt-6-luna on one owner's Mac), 67 of 71 Codex picks in a day, and no one rated that. */
+export const CODEX_DEFAULT_NOT_OFFERED = 'Codex at its default model or effort is not offered: it runs whatever the account\'s CLI chooses. Start the catalog\'s named models.'
/** The catalog's models and how strong the models served on the Swarm's grid count: the first named row whose every word is
* in a model's name, else `strength`. */
export interface Catalog { models: CatalogModel[]; grid: { strength: number; models: Array<{ model: string; strength: number }> } }
@@ -212,8 +215,11 @@ export function catalogFrom(effective: Json): Catalog {
const models = rows.map(object).flatMap((m): CatalogModel[] => {
const strength = served ? m.strength : m.quality, offered = served ? m.offered : m.enabled
if (!ENGINES.includes(String(m.engine)) || typeof m.model !== 'string' || typeof m.effort !== 'string' || !finite(strength) || !finite(m.cost)) return []
- return [{ engine: m.engine as CatalogModel['engine'], model: m.model, effort: m.effort, strength, cost: m.cost, offered: offered === true,
- ...(served && typeof m.note === 'string' && m.note ? { note: m.note } : {}) }]
+ // A Codex row at `default` model or effort stays for rating a member launched on the engine's own model, but is never offered.
+ const unnamed = m.engine === 'codex' && (m.model === 'default' || m.effort === 'default')
+ return [{ engine: m.engine as CatalogModel['engine'], model: m.model, effort: m.effort, strength, cost: m.cost, offered: offered === true && !unnamed,
+ ...(served && m.engineDefault === true ? { engineDefault: true as const } : {}),
+ ...(unnamed ? { note: CODEX_DEFAULT_NOT_OFFERED } : served && typeof m.note === 'string' && m.note ? { note: m.note } : {}) }]
})
const grid = served ? object(catalog.grid) : object(policy.localModels)
const gridRows = (Array.isArray(grid.models) ? grid.models : []) as unknown[]
@@ -303,13 +309,15 @@ export function repositoryKey(url: string): string | null {
}
/** A member's model is what its engine reported after work: Claude's full id (`claude-opus-4-7`, `claude-sonnet-4-5[1m]`),
- * Codex's slug (`gpt-5.5`). The catalog names Claude families and a Codex `default`, so the id is reduced to its family
- * first. Its row: the exact one, else the model's at another effort, else the engine's `default` row when the agent was
+ * Codex's slug (`gpt-5.5`). The catalog names Claude families and Codex's default model, so the id is reduced to its family
+ * first. Its row: the exact one, else the model's at another effort, else the engine's named default when the agent was
* launched on the engine's own model ([engineDefault]) and is no small variant; otherwise none, so a small Codex variant is
* never taken for a strong one (review of #2137). */
export function catalogRow(c: Catalog, engine: string, model: string | null, effort: string | null, engineDefault = false): CatalogModel | null {
const name = engine === 'claude' && model ? claudeAliasForModel(model.replace(/\[1m\]$/i, '')) ?? model : model ?? 'default'
- const rowOf = (named: string) => c.models.find(m => m.engine === engine && m.model === named && m.effort === (effort ?? 'default'))
+ // The engine's own model is the catalog's named default (`engineDefault`), else a `default` row of an older backend.
+ const rowOf = (named: string) => named === 'default' ? c.models.find(m => m.engine === engine && m.engineDefault) ?? c.models.find(m => m.engine === engine && m.model === 'default')
+ : c.models.find(m => m.engine === engine && m.model === named && m.effort === (effort ?? 'default'))
?? c.models.find(m => m.engine === engine && m.model === named)
// A small model it reports is never promoted by its launch (review of ee4ee6c9a).
return rowOf(name) ?? (engineDefault && !SMALL.test(model ?? '') ? rowOf('default') : undefined) ?? null
diff --git a/cli/src/superTabs/routeCommand.spec.ts b/cli/src/superTabs/routeCommand.spec.ts
index c6fa308ba..f8323de9b 100644
--- a/cli/src/superTabs/routeCommand.spec.ts
+++ b/cli/src/superTabs/routeCommand.spec.ts
@@ -18,6 +18,7 @@ const policy = { version: 1, floor: { low: 1, medium: 2, high: 3 }, quotaReserve
{ engine: 'claude', model: 'opus', effort: 'high', quality: 3, cost: 3, enabled: true },
{ engine: 'claude', model: 'sonnet', effort: 'medium', quality: 3, cost: 2, enabled: true },
{ engine: 'codex', model: 'default', effort: 'default', quality: 3, cost: 2, enabled: false },
+ { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', quality: 3, cost: 2, enabled: false },
] }
/** The table with Sonnet at its real rating, 2, for the specs of strength and floors. */
const floorPolicy = { ...policy, models: policy.models.map(m => m.model === 'sonnet' ? { ...m, quality: 2 } : m) }
@@ -1945,6 +1946,30 @@ describe('routing by facts', () => {
expect(shown(old).map(r => r.title)).toContain('New agent · Claude sonnet medium · remote-1')
})
+ // Task #2972: 95% of Codex candidates were `Codex default default` (6,570 against 122 Sol and 116 Astra), and 67 of 71 Codex picks;
+ // it ran whatever the account's CLI defaulted to (gpt-6-luna on one owner's account), which the catalog does not rate.
+ it('offers a Codex machine only the models the catalog rates, never `default default`, and gpt-6-luna only when named', async () => {
+ const catalog = { models: [{ engine: 'claude', model: 'opus', effort: 'high', strength: 3, cost: 3, offered: true },
+ { engine: 'codex', model: 'default', effort: 'default', strength: 3, cost: 2, offered: true },
+ { engine: 'codex', model: 'gpt-6.1-sol', effort: 'high', strength: 3, cost: 2, offered: true, engineDefault: true },
+ { engine: 'codex', model: 'gpt-6-astra', effort: 'xhigh', strength: 3, cost: 5, offered: true }],
+ grid: { strength: 1, models: [] }, priors: {}, prices: [] }
+ const both = () => opener(id => daemon(id, { agents: id === 'm1' ? ['h1'] : [],
+ engines: [{ engine: 'claude', installed: true, signedIn: true }, { engine: 'codex', installed: true, signedIn: true }] }))
+ const context = (extra: Record<string, unknown> = {}) => scoped({ context: { ok: true, value: { agentId: 'lead', tab: tab(), policy: { policy, warnings: [], catalog }, machines: [{ machineId: 'm1', name: 'remote-1' }], ...extra } } })
+ const served = context()
+ await routeTask(input, served, where as never, both(), () => 'op')
+ const codex = shown(served).filter(r => r.title.startsWith('New agent · Codex'))
+ expect(codex.map(r => r.title)).toEqual(['New agent · Codex gpt-6-astra xhigh · remote-1', 'New agent · Codex gpt-6.1-sol high · remote-1'])
+ expect(codex.map(r => r.label).join('\n')).toMatch(/strength 3, cost 5[\s\S]*strength 3, cost 2/)
+ expect(shown(served).map(r => r.title).join('\n')).not.toMatch(/Codex default|luna/)
+ // A person who names the model gets it, unrated: gpt-6-luna is no row of the catalog.
+ const luna = context()
+ await routeTask({ ...input, assessment: { ...input.assessment, model: 'gpt-6-luna' } }, luna, where as never, both(), () => 'op')
+ expect(shown(luna).filter(r => r.title.startsWith('New agent')).map(r => r.title)).toEqual(['New agent · Codex gpt-6-luna default · remote-1'])
+ expect(shown(luna)[0]!.label).toContain('unrated')
+ })
+
// Final review: a named grid model whose host is down became an unrated Codex candidate; nothing serves it, and the route says
// why. An unreadable catalog became an empty one without a word: no new agent was offered and nobody heard why.
it('names no stand-in for a named model whose host is offline, and says why', async () => {
diff --git a/docs/plans/2026-11-04-route-real-codex-models-plan.md b/docs/plans/2026-11-04-route-real-codex-models-plan.md
new file mode 100644
index 000000000..1680b3d4f
--- /dev/null
+++ b/docs/plans/2026-11-04-route-real-codex-models-plan.md
@@ -0,0 +1,86 @@
+# Route offers real Codex models Implementation Plan
+
+> Implemented by the task's Engineer task by task, each failing test committed before its code (project instructions,
+> Engineer). One independent review on the pull request.
+
+**Goal:** The route never offers Codex "default default". It offers the models the catalog rates (GPT-6.1 Sol at high,
+GPT-6 Astra at xhigh), and a start that asks Codex for `default` runs the catalog's named default, Sol at high, not
+whatever the account's CLI defaults to (task #2972, Ken, 2026-11-04).
+
+**Architecture:** Measured in the 24 hours before #3010's release: 95% of Codex candidates were `codex default default`
+and 67 of 71 Codex picks were that row, so a PM that started it as offered got a model nobody rated; the two Codex runs
+were `gpt-6-luna`, the default of one Mac owner's account, which the catalog does not list.
+
+- Backend `catalog.ts`: drop the `codex/default/default` row. Mark the Sol row `engineDefault`, the catalog's own word for
+ "the model Codex runs when nobody names one", so the backend and the CLI read it from one place.
+- Backend `commands.ts` `agent.start`: a Codex start with `model: 'default'` (not a grid model) is stored as the
+ catalog's default model, and an effort of `default` or none as that row's effort. A named effort is kept. The route
+ decision link still matches on what the caller sent, so a start of an old `default default` candidate stays linked.
+- Backend `routingPolicy.ts`: an owner's saved policy gets the Sol and Astra rows when Codex is on; that was read off the
+ `default` row, which no longer exists, so it reads any enabled Codex row.
+- CLI `route.ts`: `catalogFrom` keeps `engineDefault`. A `codex` row at `default` model or effort read from an older
+ backend is not offered. `catalogRow` takes the `engineDefault` row for a member launched on the engine's own model, so
+ it keeps its Sol rating.
+- A Codex start that names no model at all is left as it is: Desktop's import sends none to keep an existing session's model, and
+ the route always names one. Only `model: 'default'` is resolved.
+- `gpt-6-luna` is not rated: there is no benchmark for it, so it is never offered unless a person names it, when it is
+ offered unrated as before.
+
+**Tech Stack:** TypeScript; vitest in `backend/` and `cli/`.
+
+**Spec:** task #2972's brief.
+
+## Global Constraints
+
+- Red first for each risk the task names; red/green evidence in a PR comment.
+- Order stays the instructions' (#2901): nothing here ranks.
+- Swarm code stays out of `cli/src/core`.
+- No new setting; no migration: the catalog is served on read and old CLIs read the same rows as a policy.
+
+## Design
+
+- growth: N/A because the catalog is a fixed list of rows in code; one row fewer.
+- concurrency: N/A because nothing is stored beyond the start's request, written under the existing command lock.
+- read cost: N/A because the catalog read is unchanged in size (one row fewer).
+- scale: N/A because nothing scales with the change.
+- compatibility: an older CLI reading the catalog or the policy sees no `default` row, so it offers the named models;
+ a newer CLI on an older backend does not offer a `default` row it still serves; a start that sends `default` from an
+ older client is resolved by the backend.
+- migration + rollback: none; reverting the commit restores the row, and a start made meanwhile keeps the explicit model
+ it was stored with.
+- audience: N/A because the catalog is public to members already.
+- identity: N/A because no login is read or written.
+- failure: a Team account whose owner's ChatGPT plan does not offer Sol is refused before launch with what it offers
+ (the existing `codexModelNotOffered` check), instead of launching on whatever the CLI defaults to.
+- platforms: all; no platform code.
+
+## Review Focus
+
+- A start with `model: 'default'` on Codex, with and without an effort, a named effort, a grid model, and Claude.
+- Saved routing policies keep their Sol and Astra rows with the `default` row gone.
+- Members launched on the engine's own model keep their rating.
+
+## Tasks
+
+### Task 1: The route offers only named Codex models
+
+**Files:** `backend/src/superTabs/catalog.ts`, `backend/src/superTabs/catalog.test.ts`,
+`backend/src/superTabs/routingPolicy.ts`, `cli/src/superTabs/route.ts`, `cli/src/superTabs/routeCommand.spec.ts`,
+`cli/src/superTabs/__snapshots__/route.answers.json`
+
+- [ ] Failing test: the catalog has no Codex `default` row, and Sol is its `engineDefault`.
+- [ ] Failing test: a route over a catalog served with a `default` row (older backend) offers Sol high and Astra xhigh
+ and never `default default`, on a Codex machine; `gpt-6-luna` is never offered, and is offered unrated when named.
+- [ ] Failing test: a member launched on the engine's own model is rated as the catalog's `engineDefault`.
+- [ ] Failing test: a saved policy still gains its Sol and Astra rows.
+- [ ] Code for the above; update the route snapshot.
+
+### Task 2: A Codex start for `default` runs the catalog's default
+
+**Files:** `backend/src/superTabs/commands.ts`, `backend/src/superTabs/commands.test.ts`
+
+- [ ] Failing test: `agent.start` Codex `default`/`default` is stored as Sol at high; `default` with a named effort
+ keeps the effort; `default` with no effort gets high; a named model, a grid model and Claude are untouched.
+- [ ] Failing test: starting a candidate by the route decision still links to it, for an old `default default` candidate
+ and for Sol high.
+- [ ] Code.