[HamzaKhanBUIC]

Summary

Fixes #63 (AUDIT-06: High/Medium — Gemini thought parts are appended to the answer).

Problem

In agentkit/llm/gemini.py, the candidate parts loop appended part.text to text_parts without checking the thought flag on types.Part. This caused internal chain-of-thought reasoning to be concatenated into LLMResponse.text, leaking into assistant turns, persisted history, and the API final_answer.

Changes

1. Exclude Thought Parts from Response Text:
   • In agentkit/llm/gemini.py, updated candidate content parsing to check if getattr(part, text, None) and not getattr(part, thought, False): text_parts.append(part.text).
   • Ensured that thought parts carrying a function_call still yield the tool call and preserve its thought_signature.
2. Tests and Docs:
   • Added unit test verifying thought parts are stripped from response.text.
   • Added integration test verifying the API endpoint POST /runs emits an answer with zero thought text.
   • Added unit test verifying thought parts with function calls properly yield the tool call and thought signature.
   • Updated docs/PROGRESS.md.

Verification

• pytest tests/unit/test_audit_regressions.py tests/unit/test_gemini_adapter.py: 41/41 passed.
• ruff check: clean (0 errors).
• ruff format check: clean (110 files formatted).
• mypy agentkit: clean (no issues found in 42 source files).
• Verification gate: Semgrep differential security check passed (0 warnings).

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.