fix: treat zero limit.output as unknown to enable fallback to bundled…

[ahuangsnail]

Summary

• Fix maxOutputTokens must be >= 1 error when using custom OpenAI-compatible providers (e.g., infini-ai) that don't return model limits
• Treat zero/negative limit.output values as unknown to enable fallback to bundled snapshot with correct limits
• Aligns readRuntimeModelLimitOutput behavior with the nullish coalescing fallback chain in get-model-capabilities.ts

Changes

• Modified readRuntimeModelLimitOutput() in src/shared/model-capabilities/runtime-model-readers.ts:
  • Changed return from readNumber(limit.output) to check for positive values only
  • Returns undefined for 0 or negative values instead of passing them through
  • This allows the ?? operator in getModelCapabilities() to fall through to bundled snapshot entries

Screenshots

Testing

bun run typecheck
bun test
To verify the fix manually:
1. Configure a custom OpenAI-compatible provider that doesn't return model limits
2. Use any agent (e.g., sisyphus) with a model from that provider
3. Confirm no maxOutputTokens must be >= 1 error occurs
4. Verify maxOutputTokens falls back to bundled snapshot values (e.g., 32768 for kimi-k2.5)
Related Issues
<!-- Add issue number if one exists, e.g., Closes #123 -->

<!-- This is an auto-generated description by cubic. -->
---
## Summary by cubic
Fixes maxOutputTokens errors with some OpenAI-compatible providers by treating zero or negative `limit.output` as unknown. `readRuntimeModelLimitOutput()` now returns `undefined` for non‑positive values so `getModelCapabilities()` falls back to bundled snapshot limits.

<sup>Written for commit 6c3308aab3904b1e91ea29fba3487e77865c689b. Summary will update on new commits.</sup>

<!-- End of auto-generated description by cubic. -->

[github-actions]

All contributors have signed the CLA. Thank you! ✅
_{Posted by the CLA Assistant Lite bot.}

[cubic-dev-ai]

No issues found across 1 file

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Auto-approved: Fixes maxOutputTokens errors by treating non-positive limits as unknown, allowing fallback to bundled snapshots. Logic is sound and carries no regression risk.}

[codgician]

Tested locally and fixes #3247

[ahuangsnail]

I have read the CLA Document and I hereby sign the CLA

[ahuangsnail]

recheck