Uh oh!
There was an error while loading. Please reload this page.
ci: unbreak main test suite - #1750
Conversation
- workflow-executor: the truncation guard test asserted the whole
select-record prompt does not contain '-1', but the prompt embeds the
current date; any date with '-1' as a substring (e.g. 2026-07-10)
matched and failed the assertion. Target the actual none-option text
('return -1') instead.
- ai-proxy: denylist gpt-5.6-luna/sol/terra, which reject function tools
with reasoning_effort on v1/chat/completions (v1/responses only).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>| // Reject reasoning_effort with function tools on v1/chat/completions (v1/responses only) | ||
| 'gpt-5.6-luna', | ||
| 'gpt-5.6-sol', | ||
| 'gpt-5.6-terra', | ||
| ]; |
There was a problem hiding this comment.
🟡 Mediumsrc/supported-models.ts:49
Adding gpt-5.6-luna, gpt-5.6-sol, and gpt-5.6-terra to OPENAI_UNSUPPORTED_MODELS causes isModelSupportingTools() to return false for all three models, even though they support function calling via v1/chat/completions. This blocks users from configuring these models entirely — any config referencing them is rejected with AIModelNotSupportedError during router construction, including plain chat requests and tool calls that don't use reasoning_effort. If the issue is specifically that reasoning_effort is rejected when combined with function tools on v1/chat/completions, a blanket model block is too broad. Consider handling that restriction at the request level (e.g., erroring only when reasoning_effort and function tools are combined) rather than marking the models fully unsupported.
'us-40-51r-vm-ev3', // Not a chat model (v1/completions only)
- // Reject reasoning_effort with function tools on v1/chat/completions (v1/responses only)- 'gpt-5.6-luna',- 'gpt-5.6-sol',- 'gpt-5.6-terra',🚀 Reply "fix it for me" or copy this AI Prompt for your agent:
In file @packages/ai-proxy/src/supported-models.ts around lines 49-53:
Adding `gpt-5.6-luna`, `gpt-5.6-sol`, and `gpt-5.6-terra` to `OPENAI_UNSUPPORTED_MODELS` causes `isModelSupportingTools()` to return `false` for all three models, even though they support function calling via `v1/chat/completions`. This blocks users from configuring these models entirely — any config referencing them is rejected with `AIModelNotSupportedError` during router construction, including plain chat requests and tool calls that don't use `reasoning_effort`. If the issue is specifically that `reasoning_effort` is rejected when combined with function tools on `v1/chat/completions`, a blanket model block is too broad. Consider handling that restriction at the request level (e.g., erroring only when `reasoning_effort` and function tools are combined) rather than marking the models fully unsupported.
Coverage Impact ⬆️ Merging this pull request will increase total coverage on 🛟 Help
|
Uh oh!
There was an error while loading. Please reload this page.
…s API The deny-list from #1750 worked around the chat/completions 400; the Responses switch makes these models pass tool calling again. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Why
The
Lint, Build, Test and Docworkflow has been red onmainsince 2026-07-09 due to two independent, non-code-regression failures.Fixes
1. workflow-executor — date-dependent test
load-related-record-step-executor.test.tsasserted the whole select-record promptnot.toContain('-1')to verify the-1"none" option is withheld when the candidate list is truncated. The prompt embeds the current date (Current date and time: 2026-07-10...), and any date containing the substring-1(day 10-19, or months 10-12) matched, failing the assertion. Passed on 2026-07-08 (last green run), broke on 2026-07-10.Fix: assert against the actual none-option text (
return -1) rather than the raw-1substring across the full prompt.2. ai-proxy — unsupported OpenAI models
gpt-5.6-luna,gpt-5.6-sol,gpt-5.6-terrareject function tools combined withreasoning_effortonv1/chat/completions(they only supportv1/responses), failing thellm.integrationtool-support test. Added them toOPENAI_UNSUPPORTED_MODELS, per the maintained allow/deny-list convention, plus a unit test.Test
workflow-executor: 102/102 passai-proxysupported-models: 8/8 pass🤖 Generated with Claude Code
Note
Fix CI test suite by blocking gpt-5.6-luna, gpt-5.6-sol, and gpt-5.6-terra from tool support
'gpt-5.6-luna','gpt-5.6-sol', and'gpt-5.6-terra'toOPENAI_UNSUPPORTED_MODELSin supported-models.ts; these models rejectreasoning_effortwith function tools onv1/chat/completions.'return -1'rather than'-1'.📊 Macroscope summarized 8cfea08. 1 file reviewed, 0 issues evaluated, 0 issues filtered, 0 comments posted
🗂️ Filtered Issues
No issues evaluated.