Diagnosed failing AI package testsPublic

./test.sh fails because packages/ai has 2 failing tests. Everything else passed.

Failing tests

• packages/ai/test/openai-completions-tool-choice.test.ts:203

  • enables tool_stream for supported z.ai models with tools
  • Expected params.tool_stream === true
  • Got undefined

• packages/ai/test/openai-completions-tool-choice.test.ts:240

  • stores z.ai tool_stream support in model compat metadata
  • Expected getModel("zai", "glm-5")?.compat?.zaiToolStream === true
  • Got undefined

What this points to

• The z.ai model metadata is missing compat.zaiToolStream for supported models.
• The provider code already checks that flag in packages/ai/src/providers/openai-completions.ts:398.
• The generated model data does not include it for the affected models:
  • packages/ai/src/models.generated.ts:13911 (glm-4.7)
  • packages/ai/src/models.generated.ts:13929 (glm-4.7-flash)
  • packages/ai/src/models.generated.ts:13965 (glm-5)
• Example current compat block:
  • compat: {"supportsDeveloperRole":false,"thinkingFormat":"zai"}
  • Missing zaiToolStream:true

Likely root cause

• packages/ai/scripts/generate-models.ts:402 has logic to add zaiToolStream: true, but packages/ai/src/models.generated.ts looks stale or the generation path is not applying it for these models.

High level result

• Workspace summary: 1 failed | 24 passed | 19 skipped in packages/ai
• Monorepo test.sh exits with code 1 because of that workspace failure

If you want, I can trace why the generated models missed zaiToolStream and fix it.