[juliusmarminge]

Problem

When delegate_task is given a model slug the provider does not advertise, it refuses with Model codemie/gpt-6.1-terra is not advertised by provider codex.. The message names no valid slug, so an agent that built the slug from prose has to call orchestrator_capabilities before it can retry. To reproduce, call delegate_task with target: {providerInstanceId: "codex", model: "codemie/gpt-6.1-terra"}.

Change

The model_unavailable refusal in OrchestratorMcpService target resolution now ends with a hint built from provider.models, the same set delegate_task accepts:

• When advertised slugs share the requested slug's last / segment (case-insensitive), it adds Did you mean <slug>?. For example, codex/gpt-6.1-sol suggests gpt-6.1-sol.
• Otherwise it lists up to LISTED_MODEL_LIMIT (8) advertised slugs, current models first. If any are left out, it adds and N more; orchestrator_capabilities with search finds them.

The error code and the start of the message are unchanged. create_threads uses the same target resolver, so its refusal gets the same hint. No contract, option or persisted state changes.

Scope and approval

Fixes #31890

Triaged via the automated issue flow.

Verification

• vp test run apps/server/src/mcp/OrchestratorMcpService.test.ts: 61/61 pass.

  • an unadvertised model refusal names the advertised slugs to retry with fails on main: expected 'Model is not advertised by provider …' to include 'gpt-6.1-terra', with the requested slug removed before the check. It passes with the fix.
  • an unadvertised model with no close match lists a capped set of current slugs covers the capped fallback list. Legacy models come last and the rest are counted as "N more".

• vp run check:changed: format, lint, knip, typecheck and tests all pass.

• Runtime transcript. Each side ran its own server (t3 serve) on a fresh, empty home, with a scratch git project added. t3 thread start --provider claudeAgent --wait drove a real agent turn that called the delegate_task MCP tool with three targets and copied each error verbatim:

  Target	Before (main 2756da4)	After (b985946)
  {providerInstanceId: "codex", model: "codemie/gpt-6.1-terra"}	Model codemie/gpt-6.1-terra is not advertised by provider codex.	Model codemie/gpt-6.1-terra is not advertised by provider codex. Advertised models: gpt-6.1-sol, gpt-6-astra, gpt-6-luna, gpt-6-sol, gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5.
  {driverKind: "codex", model: "codex/gpt-6.1-sol"}	Model codex/gpt-6.1-sol is not advertised by provider codex.	Model codex/gpt-6.1-sol is not advertised by provider codex. Did you mean gpt-6.1-sol?
  {providerInstanceId: "claudeAgent", model: "opus-5.5"}	Model opus-5.5 is not advertised by provider claudeAgent.	Model opus-5.5 is not advertised by provider claudeAgent. Advertised models: claude-opus-5-5, claude-sonnet-5-5, claude-fable-5-1, claude-haiku-5-5, claude-fable-5, claude-opus-5, claude-opus-4-8, claude-opus-4-7 and 5 more; orchestrator_capabilities with search finds them.

  All three calls returned code: "model_unavailable" on both sides. This catalog has no gpt-6.1-terra, so the first row falls back to the list.

• Server-only change with no visible surface, so there are no captures.

• Not checked: the hint does not apply the user's hidden-model and model-order preferences. It lists from the same raw catalog that delegate_task validates against.

AI-assisted; verified with the tests and tools listed above.

🤖 Generated with Claude Code

[macroscopeapp]

Approvability

Verdict: Approved at 7281cc2

Macroscope's review found this PR approvable — Model-unavailable errors now suggest advertised model slugs to help callers retry. Existing validation and model-selection behavior remain unchanged, limiting the change to diagnostic guidance.

^{You can add or adjust custom eligibility rules. Learn more.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	21.5 KiB	21.5 KiB	0 B (0.0%)	24.4 KiB	✅
Codex	Live turn messages	1	1	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.9 KiB	21.9 KiB	0 B (0.0%)	24.4 KiB	✅
Claude	Live turn messages	1	1	0 (0.0%)	8	✅

Other scenarios

All 64 scenario measurements match main.

All 64 scenario measurements

Provider	Scenario	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	short turn resume	WebSocket decoded	10.5 KiB	10.5 KiB	0 B (0.0%)	12.7 KiB	✅
Codex	short turn resume	WebSocket messages	2	2	0 (0.0%)	2	✅
Codex	long turn running, cold open	HTTP decoded	281.7 KiB	281.7 KiB	0 B (0.0%)	302.7 KiB	✅
Codex	long turn running, cold open	rows	75	75	0 (0.0%)	75	✅
Codex	long turn running, history page	HTTP decoded	92.4 KiB	92.4 KiB	0 B (0.0%)	107.4 KiB	✅
Codex	long turn running, history page	rows	75	75	0 (0.0%)	75	✅
Codex	long turn finished, cold open	HTTP decoded	381.5 KiB	381.5 KiB	0 B (0.0%)	410.2 KiB	✅
Codex	long turn finished, cold open	rows	151	151	0 (0.0%)	151	✅
Codex	long turn finished, socket cold open	WebSocket decoded	381.6 KiB	381.6 KiB	0 B (0.0%)	410.2 KiB	✅
Codex	long turn finished, socket cold open	WebSocket messages	2	2	0 (0.0%)	2	✅
Codex	long turn finished, socket cold open	rows	151	151	0 (0.0%)	151	✅
Codex	long turn finished, work page	HTTP decoded	186.2 KiB	186.2 KiB	0 B (0.0%)	200.2 KiB	✅
Codex	long turn finished, work page	rows	150	150	0 (0.0%)	150	✅
Codex	idle thread resume	WebSocket decoded	65 B	65 B	0 B (0.0%)	200 B	✅
Codex	idle thread resume	WebSocket messages	1	1	0 (0.0%)	1	✅
Codex	fleet shell snapshot	HTTP decoded	42.4 KiB	42.4 KiB	0 B (0.0%)	53.7 KiB	✅
Codex	fleet shell snapshot	rows	27	27	0 (0.0%)	27	✅
Codex	fleet shell subscribe	WebSocket decoded	65 B	65 B	0 B (0.0%)	200 B	✅
Codex	fleet shell subscribe	WebSocket messages	1	1	0 (0.0%)	1	✅
Codex	fleet shell resume check	HTTP decoded	18 B	18 B	0 B (0.0%)	100 B	✅
Codex	fleet shell resume check	rows	0	0	0	0	✅
Codex	stale shell reload	HTTP decoded	42.4 KiB	42.4 KiB	0 B (0.0%)	53.7 KiB	✅
Codex	stale shell reload	rows	27	27	0 (0.0%)	27	✅
Codex	stale shell resume	WebSocket decoded	140 B	140 B	0 B (0.0%)	400 B	✅
Codex	stale shell resume	WebSocket messages	2	2	0 (0.0%)	2	✅
Codex	stale todo list cold open	HTTP decoded	72.4 KiB	72.4 KiB	0 B (0.0%)	83.0 KiB	✅
Codex	stale todo list cold open	rows	75	75	0 (0.0%)	75	✅
Codex	stale todo list cold open	nodes	2	2	0 (0.0%)	4	✅
Codex	stale todo list cold open	subagents	2	2	0 (0.0%)	4	✅
Codex	stale todo list cold open	plans	1	1	0 (0.0%)	2	✅
Codex	stale todo list cold open	providerTurns	1	1	0 (0.0%)	2	✅
Codex	stale todo list cold open	checkpoints	1	1	0 (0.0%)	2	✅
Claude	short turn resume	WebSocket decoded	10.6 KiB	10.6 KiB	0 B (0.0%)	12.7 KiB	✅
Claude	short turn resume	WebSocket messages	2	2	0 (0.0%)	2	✅
Claude	long turn running, cold open	HTTP decoded	283.3 KiB	283.3 KiB	0 B (0.0%)	302.7 KiB	✅
Claude	long turn running, cold open	rows	75	75	0 (0.0%)	75	✅
Claude	long turn running, history page	HTTP decoded	93.4 KiB	93.4 KiB	0 B (0.0%)	107.4 KiB	✅
Claude	long turn running, history page	rows	75	75	0 (0.0%)	75	✅
Claude	long turn finished, cold open	HTTP decoded	384.2 KiB	384.2 KiB	0 B (0.0%)	410.2 KiB	✅
Claude	long turn finished, cold open	rows	151	151	0 (0.0%)	151	✅
Claude	long turn finished, socket cold open	WebSocket decoded	384.3 KiB	384.3 KiB	0 B (0.0%)	410.2 KiB	✅
Claude	long turn finished, socket cold open	WebSocket messages	2	2	0 (0.0%)	2	✅
Claude	long turn finished, socket cold open	rows	151	151	0 (0.0%)	151	✅
Claude	long turn finished, work page	HTTP decoded	188.1 KiB	188.1 KiB	0 B (0.0%)	200.2 KiB	✅
Claude	long turn finished, work page	rows	150	150	0 (0.0%)	150	✅
Claude	idle thread resume	WebSocket decoded	66 B	66 B	0 B (0.0%)	200 B	✅
Claude	idle thread resume	WebSocket messages	1	1	0 (0.0%)	1	✅
Claude	fleet shell snapshot	HTTP decoded	43.0 KiB	43.0 KiB	0 B (0.0%)	53.7 KiB	✅
Claude	fleet shell snapshot	rows	27	27	0 (0.0%)	27	✅
Claude	fleet shell subscribe	WebSocket decoded	66 B	66 B	0 B (0.0%)	200 B	✅
Claude	fleet shell subscribe	WebSocket messages	1	1	0 (0.0%)	1	✅
Claude	fleet shell resume check	HTTP decoded	18 B	18 B	0 B (0.0%)	100 B	✅
Claude	fleet shell resume check	rows	0	0	0	0	✅
Claude	stale shell reload	HTTP decoded	43.0 KiB	43.0 KiB	0 B (0.0%)	53.7 KiB	✅
Claude	stale shell reload	rows	27	27	0 (0.0%)	27	✅
Claude	stale shell resume	WebSocket decoded	140 B	140 B	0 B (0.0%)	400 B	✅
Claude	stale shell resume	WebSocket messages	2	2	0 (0.0%)	2	✅
Claude	stale todo list cold open	HTTP decoded	73.4 KiB	73.4 KiB	0 B (0.0%)	83.0 KiB	✅
Claude	stale todo list cold open	rows	75	75	0 (0.0%)	75	✅
Claude	stale todo list cold open	nodes	2	2	0 (0.0%)	4	✅
Claude	stale todo list cold open	subagents	2	2	0 (0.0%)	4	✅
Claude	stale todo list cold open	plans	1	1	0 (0.0%)	2	✅
Claude	stale todo list cold open	providerTurns	1	1	0 (0.0%)	2	✅
Claude	stale todo list cold open	checkpoints	1	1	0 (0.0%)	2	✅

Baseline: e1458f2 · PR result: 7281cc2 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

When delegation requests an unadvertised model, the refusal now includes matching advertised slugs or a capped list of advertised models with guidance to search capabilities. Tests cover matching and fallback suggestions.

Changes: Delegation model hints

Layer / File(s)	Summary
Generate and attach model hints apps/server/src/mcp/OrchestratorMcpService.ts, apps/server/src/mcp/OrchestratorMcpService.test.ts, apps/server/src/mcp/toolkits/project/handlers.test.ts	The service suggests up to eight models whose final slug segment matches the request, ignoring case. If no models match, it prioritizes non-legacy models, lists up to eight, reports the omitted count, and points to capabilities search. Tests cover both suggestion paths and check for the updated refusal message.

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix · Severity of issue fixed: Low

Suggested reviewers: t3dotgg

Merge Risk: Minimal · up to 7281c

Delegation refusals for unadvertised models now suggest valid model slugs to retry with. The change is low-risk and affects only message text, so it is ready to merge.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title names the main change: delegate_task model refusals now name advertised slugs so the caller can retry. It uses a conventional commit prefix and fits the changeset, which adds an advertisedMo…
Description check		The description follows the template structure: Problem, Change, Scope, and Verification sections are all present. The Problem and Change sections are clear and specific. Scope links issue #31890 and …
Linked Issues check		[#31890] The only directly linked issue is active, so its coding requirement applies: a model_unavailable refusal for an unadvertised slug should list advertised models or close matches, so the agen…
Out of Scope Changes check		The whole-PR service diff contains only advertisedModelsHint and its single call site in requireAdvertisedModel. Both support [#31890]. The shared resolver also covers create_threads, as the PR …

Full details: Title check

Explanation

The title names the main change: delegate_task model refusals now name advertised slugs so the caller can retry. It uses a conventional commit prefix and fits the changeset, which adds an advertisedModelsHint to the model_unavailable refusal in OrchestratorMcpService. The phrasing is slightly awkward ("name slugs to retry with"), but it is specific and related to the change.

Full details: Description check

Explanation

The description follows the template structure: Problem, Change, Scope, and Verification sections are all present. The Problem and Change sections are clear and specific. Scope links issue #31890 and explains the automated triage. Verification gives focused test results, a before/after runtime table, and notes what was not checked, which meets the template's requirement beyond 'tests pass'. The template's opening comment about conventional commit titles is satisfied by the title. The template asks for the model and harness that did the work when an agent was used; the description states it is AI-assisted and names Claude Code, so this is covered. The description is mostly complete.

Full details: Linked Issues check

Explanation

[#31890] The only directly linked issue is active, so its coding requirement applies: a model_unavailable refusal for an unadvertised slug should list advertised models or close matches, so the agent can retry without calling orchestrator_capabilities. The whole-PR diff (merge base e1458f2 to head 7281cc2) shows requireAdvertisedModel in apps/server/src/mcp/OrchestratorMcpService.ts now appends advertisedModelsHint(provider.models, model) to the unchanged model_unavailable message. The helper suggests Did you mean <slugs>? when advertised slugs share the requested slug's last / segment, compared case-insensitively. Otherwise it lists up to LISTED_MODEL_LIMIT slugs, legacy models last, and reports the remainder as and N more; orchestrator_capabilities with search finds them. The error code and opening text are unchanged. The PR summary states that a test file covers both the close-match suggestion and the capped fallback list. I did not open the test file, so the test result rests on the PR summary and author report. The runtime transcripts in the PR description show the intended message on the reproduction case.

Full details: Out of Scope Changes check

Explanation

The whole-PR service diff contains only advertisedModelsHint and its single call site in requireAdvertisedModel. Both support [#31890]. The shared resolver also covers create_threads, as the PR description states. The tests in OrchestratorMcpService.test.ts and the adjusted assertion in toolkits/project/handlers.test.ts (per the raw summary) verify this behavior. The adjusted assertion relaxes an exact-match check to a substring check on the unchanged not advertised text, which follows from the new hint suffix. No unrelated change is evident in the reviewed diff. One edge case is not covered by the linked issue: when a provider advertises zero models, the fallback text becomes Advertised models: .. This is a wording defect, not an out-of-scope change.

Finishing Touches

Generate unit tests (beta)

• Commit to this branch
• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Recent review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: fix/issue-31890
• File: .coderabbit.config.ts
• Commit: 7281cc2
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Team
• Run ID: 83f2095f-df3e-451e-b53f-97579e9fda32

Commits

Reviewing files that changed from the base of the PR and between f38dc7c and 7281cc2.

Files selected for processing: 3

• apps/server/src/mcp/OrchestratorMcpService.test.ts
• apps/server/src/mcp/OrchestratorMcpService.ts
• apps/server/src/mcp/toolkits/project/handlers.test.ts

Files skipped from review as they are similar to previous changes: 1

• apps/server/src/mcp/OrchestratorMcpService.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 0 remain after this review.

[coderabbitai]

Additional comments: 1

apps/server/src/mcp/OrchestratorMcpService.test.ts: 1

1836-1938: Trivial · Maintainability & Code Quality · Quick win

Unverified finding
Verification ran but could not confirm this finding. It is shown for review, not as a verified issue.

Two tests duplicate large setup blocks.

Both tests build almost identical layer stacks. They also repeat the same delegateTask call shape. The testing rules say: "Extend before you add." Fold the setup into a small local helper that takes the model list and the requested model. This also avoids a third copy when another case is added.

The new tests do exercise real behavior. The second test asserts the full message and covers the cap and the omitted count. No change is required for coverage.

Source: Path instructions

Review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: fix/issue-31890
• File: .coderabbit.config.ts
• Commit: 292947e
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Team
• Run ID: 0276c54b-0c1a-4736-ac2d-a4bf457a57f8

Commits

Reviewing files that changed from the base of the PR and between 202d5b5 and 292947e.

Files selected for processing: 2

• apps/server/src/mcp/OrchestratorMcpService.test.ts
• apps/server/src/mcp/OrchestratorMcpService.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 8 remain after this review.

Dismissing prior approval to re-evaluate f38dc7c

[juliusmarminge]

Evidence

Ran both commits as real Linux dev servers with fresh isolated databases and identical synthetic ACP catalogs (sample-1 through sample-10). During an active thread, called orchestrator_capabilities, then delegate_task with extra/sample-3, then unrelated-invalid.

Videos replay captured RPC output with extended event holds for readability; they are not UI screen recordings. Both use a 1440×900 dark presentation and identify their environment and commit.

Before	After
9cd7a04797	f38dc7ce73
Recording · Transcript · Raw RPC	Recording · Transcript · Raw RPC
Both refusals identify the rejected selection but give no valid alternative.	Prefixed selection suggests sample-3; unrelated selection lists eight advertised slugs, reports two more, and names capabilities search.

Reproduction scripts: server and seed driver, ACP catalog agent, recording renderer.

AI-assisted; verified with the tests and tools listed above.

Not shown:

• Case-insensitive suffix matching, multiple matching suffixes, and legacy-model ordering were not separately exercised; evidence covers the two requested scenarios.
• Capabilities search and a successful delegation retry were not exercised; captures establish the refusal content and bounded retry guidance.

Dismissing prior approval to re-evaluate 7281cc2

[juliusmarminge]

Evidence

Ran the same scenario against fresh real servers and web clients, using a replay agent with ten advertised models. The agent called delegate_task with extra/multi-turn, then unrelated-invalid-model, and displayed the server's verbatim responses. Captures use Chromium, 1280×800, light theme, after message synchronization completed.

Before	After
e1458f25e2	7281cc23b9
Missing-hint assertion fails	All retry-guidance assertions pass
Before recording excerpt	After recording excerpt

Reproduction scenario and replay driver. scenarios:prove exited 0: fails on base, passes on head.

Not shown:

• The shared refusal helper's create_threads and t3_thread_launch entry points were not exercised; this proof covers the reported delegate_task reproduction.