Skip to content

feat(issue-auto-implement): pass PR review/comment feedback to implement step#264

Merged
leggetter merged 1 commit intomainfrom
feat/auto-implement-review-feedback-and-logging
Mar 17, 2026
Merged

feat(issue-auto-implement): pass PR review/comment feedback to implement step#264
leggetter merged 1 commit intomainfrom
feat/auto-implement-review-feedback-and-logging

Conversation

@leggetter
Copy link
Collaborator

Summary

When a reviewer comments on an auto-implement PR (e.g. "Could we add some unit and acceptance test coverage?"), the bot was replying "Verification passed" with no code changes because the implement step never received the review/comment text — it only saw the original issue.

Changes

  • Assess: When the trigger is pull_request_review, pull_request_review_comment, or issue_comment on a PR, add review_feedback to the JSON output (the exact review/comment body).
  • Action: Pass review_feedback from assess to the implement step as REVIEW_FEEDBACK.
  • Implement: When REVIEW_FEEDBACK is set, add a clear "Review feedback to address (you must implement this)" section to the Claude Code CLI prompt so it actually implements the reviewer's ask (e.g. add tests).
  • run-local-assess: Pass REVIEW_FEEDBACK when running implement locally (from result.review_feedback or COMMENT_BODY/REVIEW_BODY).
  • Logging: Always log Claude's raw assess response to stderr (and a one-line summary). ASSESS_DEBUG=1 additionally logs the full prompt.
  • Tests: Unit tests for review_feedback when event is pull_request_review or issue_comment on a PR.
  • Docs: README and AGENTS.md updated; added DIAGNOSIS-PR-261.md documenting the root cause (PR Add --local flag to login and ci commands #261 follow-up).

Testing

cd .github/actions/issue-auto-implement/assess && npm test — 18 tests pass.

Made with Cursor

…ent step

- Assess: add review_feedback to output when trigger is PR review or comment on PR
- Action: pass review_feedback as REVIEW_FEEDBACK to implement step
- Implement: include 'Review feedback to address' in prompt when REVIEW_FEEDBACK set
- run-local-assess: pass REVIEW_FEEDBACK when running implement
- Always log Claude assess response to stderr; ASSESS_DEBUG=1 logs full prompt
- Add unit tests for review_feedback (pull_request_review, issue_comment on PR)
- Add DIAGNOSIS-PR-261.md documenting why bot replied 'Verification passed' instead of adding tests

Fixes the case where a reviewer asks for tests (or other follow-up) on a PR and
the bot only saw the original issue, so it replied with no code changes.

Made-with: Cursor
@leggetter leggetter merged commit c4ab70b into main Mar 17, 2026
9 checks passed
@leggetter leggetter deleted the feat/auto-implement-review-feedback-and-logging branch March 17, 2026 11:09
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant