Expand smoke tests for transcript payload variants

[set2374]

The stop gate supports both last_assistant_message and transcript fallback. We should add more test coverage for real-world payload variation.\n\nSuggested scope:\n- add transcript fixtures with assistant content nested in different shapes\n- test clarifying-question detection\n- test caveat detection edge cases\n- confirm that empty or malformed transcript lines fail open rather than crash\n\nThis is a good first issue if you want to work in Python without changing the public API.

[snakefood3232]

The stop gate needs more coverage for transcript payload variants like nested assistant content. I'd handle this by adding fixtures for different transcript shapes and testing clarifying-question and caveat detection edge cases. I can have a PR up within a day, ensuring empty or malformed lines fail open. I've built similar validation systems for webhook payloads before.

[laolaoqi]

Bid: Expand smoke tests for transcript payload variants

What I'll deliver:

• Transcript fixtures with assistant content nested in various shapes (single-turn, multi-turn, tool_call variants)
• Tests for clarifying-question detection edge cases
• Caveat detection edge cases (single/multiple, different phrasing)
• Empty/malformed transcript line handling (fail open, not crash)
• 95%+ coverage on the transcript parsing paths

Timeline: 2-3 days from assignment

Why me: Strong Python testing background (pytest, hypothesis, property-based). Experienced building transcript/test fixture systems for LLM tooling.

Payment address: 0x6c21794E27C34E93cd0049918d04Da58FC72595c