chore(scripts): model sync emits `tools: []` and skips ANTHROPIC_COMBINED_TOOLS_AND_SCHEMA_MODELS, so every new Anthropic model lands red (#1516)
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 68/100
- Issue type
- Bug
- Clarity
- Clearly specified
- Activity status
- Active
- Tech stack
- typescript
- Domain
- ai, testing-qa, tooling
Research direction
Start with scripts/model-sync/provider-supports.ts and scripts/sync-provider-models.ts, then run the failing Anthropic and OpenRouter tests named in the issue. Check the generated metadata against the hand-maintained Anthropic capability list and the catalog-based OpenRouter test. Done means the sync regenerates green metadata and the Sync Model Metadata workflow no longer fails.
Written by the indexing model from the issue text.
Description
Problem
The daily model sync (automated/sync-models, PR #1516) is red on Test and Coverage. The Coverage job is not reporting a coverage drop — @tanstack/ai-anthropic:test:coverage errors out, so the job exits 1 before any comparison runs.
Run: https://github.com/TanStack/ai/actions/runs/36563236332 (head 3cbc3a061)
Three failures, all from the newly synced claude-sonnet-5-5 plus one removed OpenRouter model. All three reproduce locally:
-
@tanstack/ai-anthropic:test:lib—tests/model-meta.test.ts:418, "covers every registered model that accepts output_config.format":AssertionError: expected [ 'claude-sonnet-5-5' ] to deeply equal []The sync added
CLAUDE_SONNET_5_5.idtoANTHROPIC_MODELSand to the three type maps, but not toANTHROPIC_COMBINED_TOOLS_AND_SCHEMA_MODELS(packages/ai-anthropic/src/model-meta.ts:778). -
@tanstack/ai-anthropic:test:types—tests/tools-per-model-type-safety.test.ts:252, "every registered model declares at least one provider tool":error TS2344: Type '"claude-opus-5-fast"' does not satisfy the constraint '"Expected: literal string: claude-opus-5-fast, Actual: literal string: claude-sonnet-5-5"'The generated entry has
supports: { tools: [] }, so the new model joinsclaude-opus-5-fastin the "declares no provider tool" set that the test pins to exactly one id. For comparison, the hand-writtenCLAUDE_SONNET_5(packages/ai-anthropic/src/model-meta.ts:464) listsweb_search,web_fetch,code_execution,computer_use,bash,text_editor,memory. The generated entry also omits theextended_thinking/adaptive_thinking/priority_tierflags its siblings carry — same root cause, not caught by any test. -
@tanstack/ai-openrouter:test:types—tests/openrouter-combined-structured-output.test.ts:187:'anthropic/claude-3-haiku'is no longer inOPENROUTER_CHAT_MODELS. This sync dropped it (along withdeepseek/deepseek-r1-distill-llama-70band four:freevariants), so the literal no longer type-checks.
Root cause
1 and 2 are generator bugs and recur on every sync that adds a native Anthropic model.
scripts/model-sync/provider-supports.ts hardcodes tools: [] for every provider (case 'anthropic' at line 118, and the same literal in the openai, gemini and grok branches) — a conservative placeholder that was never filled in. scripts/sync-provider-models.ts:443 emits it verbatim, and nothing in the sync knows about ANTHROPIC_COMBINED_TOOLS_AND_SCHEMA_MODELS, so a new model lands registered-but-unlisted.
Same class as #1163: the sync writes a generated model list while a hand-maintained companion list goes stale.
3 is unrelated test brittleness — the test pins a literal catalog id that OpenRouter can (and did) delete.
Fix
scripts/model-sync/provider-supports.ts: fill the real provider server tools for new models instead of emittingtools: [], per provider (Anthropic, OpenAI, Gemini, Grok each have their own vocabulary — Anthropicweb_search/code_execution/…, OpenAIweb_search/file_search/code_interpreter/mcp/…, Geminicode_execution/google_search/url_context/…). While in there, emit the per-generation Anthropic flags (extended_thinking/adaptive_thinking/priority_tier) the hand-written entries carry.scripts/sync-provider-models.ts: insert a new native Anthropic model id intoANTHROPIC_COMBINED_TOOLS_AND_SCHEMA_MODELSwhen it acceptsoutput_config.format.packages/ai-openrouter/tests/openrouter-combined-structured-output.test.ts:187: stop pinning a literal that a sync can delete — pick the model from the catalog by capability instead.- Then re-run Sync Model Metadata (
workflow_dispatch) so #1516 regenerates green. Fixing #1516's branch directly is pointless:.github/workflows/sync-models.yml:52rebuilds it frommainandgit push --forces it every day at 06:00 UTC.
- Dominant language
- TypeScript
- Stars
- 3.1k
- Forks
- 340
- Avg merge
- 2d 7h
- Merged PRs (30d)
- 175
Getting set up
- No Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from TanStack/ai
-
signin with openaiOpen
Difficulty 5/5 Over a week Newbie friendliness 45/100
Maintainers usually reply within 1 day
-
update elevenlabsPossibly taken @tombeckenham claimed this 1 day ago. Openhas-pr waiting-on: maintainer
Difficulty 4/5 3-5 days Newbie friendliness 25/100
TanStack/ai#1564 · 1 assignee ·
Maintainers usually reply within 1 day
-
onAfterToolCall failure records a second, contradictory result for a successful server toolPossibly taken @AlemTuzlak claimed this 2 days ago. Openhas-pr waiting-on: maintainer
TanStack/ai#1558 · 1 assignee ·
Maintainers usually reply within 1 day
-
SSE and NDJSON response streams drain unread sources without backpressurePossibly taken @tombeckenham claimed this 2 days ago. Openhas-pr waiting-on: maintainer
TanStack/ai#1556 · 1 assignee ·
Maintainers usually reply within 1 day
-
Solid useChat drops earlier turns after a reactive request option changesPossibly taken @AlemTuzlak claimed this 2 days ago. Openhas-pr waiting-on: maintainer
TanStack/ai#1552 · 1 assignee ·
Maintainers usually reply within 1 day
Similar issues
-
Difficulty 1/5 Under an hour Newbie friendliness 78/100
lichess-org/api#678 ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 84/100
PostHog/posthog.com#20628 ·
Maintainers usually reply within 1 day
-
bug status:Needs Triage
Difficulty 1/5 Under an hour Newbie friendliness 92/100
jupyterlab/jupyterlab#19964 ·
Maintainers usually reply within 1 day
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
agentscope-ai/QwenPaw#8064 · 1 comment ·
Maintainers usually reply within 1 day
-
area: notebooks-jupyter bug theme: new notebook frontend
Difficulty 2/5 1-3 hours Newbie friendliness 88/100
posit-dev/positron#16347 · 1 comment ·
Maintainers usually reply within 1 day