Greptile Summary

This PR scopes parallel_tool_calls payload injection to OpenAI-compatible APIs (openai-completions, openai-responses) to fix a regression (#37048) where models rejecting the field returned HTTP 400s. It also adds support for the camelCase alias parallelToolCalls, validates that only boolean values are accepted (warning and ignoring otherwise), and includes thorough regression tests.

Key strengths:

• resolveParallelToolCalls correctly handles both parallel_tool_calls and parallelToolCalls with ??-based precedence (snake_case wins) and proper invalid-value guarding.
• createParallelToolCallsWrapper correctly defers the API-compatibility check to call time, which is the right pattern.
• Test coverage is solid: covers injection for both openai-completions and openai-responses, no-op for anthropic-messages, invalid-value warning, and unconfigured passthrough.

Issue to address:

• The debug log on line 1182 fires when the wrapper is created, but the actual injection is deferred and guarded. For non-OpenAI APIs, the log will emit "applying parallel_tool_calls=..." even though injection will be silently skipped, which could confuse debugging. Consider moving the log inside the wrapper after the API check to ensure it only fires when injection actually occurs.

Confidence Score: 4/5

• Safe to merge; the fix is well-scoped, logic is correct, and backed by solid regression tests.
• The implementation correctly scopes payload injection to OpenAI-compatible APIs only, properly handles both parameter name variants with appropriate validation, and includes comprehensive test coverage for the key scenarios (supported APIs, unsupported API no-op, invalid values). The one identified issue—a misleading debug log that fires even when injection will be skipped—is a clarity concern with no runtime impact and can be fixed with a straightforward refactoring.
• The debug log at line 1182 in src/agents/pi-embedded-runner/extra-params.ts should be moved inside the wrapper function to only fire when injection actually occurs, preventing confusion during debugging sessions.

_{Last reviewed commit: ce91a5b}

Closing this PR because the author has more than 10 active PRs in this repo. Please reduce the active PR queue and reopen or resubmit once it is back under the limit. You can close your own PRs to get back under the limit.