Fix GPT-5 tool calling format incompatibility

[devin-ai-integration]

Fix GPT-5 tool calling format incompatibility

Summary

Fixes issue #3889 where GPT-5's tool calling format causes CrewAI to fail with "Tool input must be a valid dictionary" errors.

The Problem:

• GPT-5 wraps tool arguments in an array: [{"arg": "value"}, []]
• GPT-4 uses a flat dict: {"arg": "value"}
• CrewAI's tool input validation only accepted dict format, causing GPT-5 calls to fail

The Solution:
Added a _coerce_args_dict() helper method in ToolUsage class that normalizes both formats:

• Returns dict unchanged (GPT-4 format)
• Unwraps array when first element is dict and trailing elements are empty (GPT-5 format)
• Returns None for invalid formats

The normalization is applied across all 4 parsing attempts (JSON, Python literal, JSON5, repaired JSON) to ensure consistent behavior.

Test Coverage:
Added 6 new tests covering:

• GPT-5 array-wrapped format with trailing empty array
• GPT-5 format with single dict element
• GPT-4 flat dict format (backward compatibility)
• Complex nested arguments with both formats
• Invalid list formats (properly rejected)
• Multiple trailing empty elements

All 71 tests in the tools module pass.

Review & Testing Checklist for Human

• Review the normalization logic in _coerce_args_dict() (lines 529-550 in tool_usage.py) - specifically the all(not x for x in trailing_elements) check. Does this correctly identify GPT-5's format without being too permissive?
• Test with actual GPT-5 model - The unit tests mock the input, but we should verify this works end-to-end with a real GPT-5 API call to ensure the format matches what GPT-5 actually outputs
• Review uv.lock changes - The PR includes a regenerated lock file with dependency updates (necessary to fix a corrupted lock file). Verify these updates are acceptable or if we should revert to the original lock file

Notes

• Session: https://app.devin.ai/sessions/773c3bb698e841868ac24ea7a71d31bd
• Requested by: João ([email protected])
• The fix is backward compatible - GPT-4 format continues to work unchanged
• No changes to public APIs or tool definitions
• Lint checks (ruff) pass

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring

[Mukhsin0508]

Did you just do this after I requested the new feature? if so, it was too fast!