When the jr review agent reviews code for a task, its observations should be stored as comments on the task so humans can see what the review covered.
Current Behavior
- autoReview in Jr.hs runs tests and makes accept/reject decision
- Review reasoning is not captured anywhere
Proposed Behavior
- After review completes, add a comment with:
- What tests were run and results
- Any issues found
- Why accepted/rejected
- LLM reasoning summary if available
Challenge
- Currently using amp as the review agent
- amp doesn't have a hook to extract its reasoning
- May need custom reviewer agent to capture this
Options
1. Parse amp output for reasoning (fragile)
2. Build custom reviewer agent with structured output (preferred long-term)
3. Add review notes to retry_context.retryNotes (short-term hack)
For now
- Add manual review notes when human reviews
- Capture what tests were run and pass/fail status
- Full agent review comments needs custom reviewer (future work)
Blocked by
- t-183: Add comments UI (need to display comments first)