[giladsamuels-cognition]

Summary

Until now, the multi-turn tool-call formatting in response_api.py (add_responded_tool_calls and each builder's handle_tool_call) was only tested by @pytest.mark.pricy tests, and CI skips those with -m 'not pricy'. This PR adds non-pricy tests to tests/llm/test_response_api.py. They reuse the existing create_mock_* helpers, and no production code changes.

What the tests check:

• Full multi-turn loop through each mocked model (OpenAIResponseModel, ClaudeResponseModel, OpenAIChatCompletionModel): call 1 returns a tool call → response_text(...) → add_responded_tool_calls → call 2. The test then asserts the exact input= / messages= payload sent to create on call 2:
  • Responses API: raw output items, then {"type": "function_call_output", "call_id", "output"}
  • Anthropic: {"role": "assistant", "content": raw.content}, then {"role": "user", "content": [{"type": "tool_result", "tool_use_id", "content"}, ...]} (two parallel calls, order preserved)
  • Chat Completions: raw assistant message, then one {"name", "role": "tool", "tool_call_id", "content"} per call (two parallel calls)
• Responses computer_call branch produces computer_call_output with an input_image output.
• Guard cases:
  • An image in a function-call or tool response raises AssertionError (all three builders).
  • Text in a computer_call response raises AssertionError.
  • add_responded_tool_calls raises when only some tool calls have a response set.
  • A tool message whose responded_tool_calls is unset raises ValueError.

To confirm the tests catch real regressions, I temporarily broke the source in four places and 7 tests failed:

• renamed the call_id key
• changed the tool_use_id value
• changed the tool_call_id value
• disabled the all_responses_set assert

Result with the source restored: pytest -m 'not pricy' tests/llm/ gives 44 passed, 3 skipped.

Link to Devin session: https://app.devin.ai/sessions/b30c9ada7dad48f8921cb253abbf1def
Open in Devin Desktop: https://app.devin.ai/desktop/session/b30c9ada7dad48f8921cb253abbf1def?variant=devin
Requested by: @giladsamuels-cognition

---

Devin Review

Status	Commit
⚪ Not started	—

Run Devin Review

---

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.