PR Feedback
Reviewers (humans and/or Copilot) have left feedback on your pull request. Your job is to review it, decide what is correct and worth implementing, and act on it.
Phase 1: Fetch feedback
Run the pr-comments fetch script to get all PR comments, reviews, and review threads.
!.claude/skills/pr-comments/scripts/fetch-pr-comments.sh $ARGUMENTS 2>&1
Phase 2: Triage
Go through every piece of feedback (review comments, inline threads, conversation comments) and categorise each item into one of these buckets:
| Bucket | Criteria | Action |
|---|---|---|
| Implement | Correct, improves the code, and in scope | Will fix |
| Acknowledge | Valid observation but out of scope or a deliberate choice | Note why — may reply to reviewer |
| Decline | Incorrect, based on a misunderstanding, or would make things worse | Note why — may reply to reviewer |
Present the triage as a table so the user can review your reasoning before you start work.
Triage guidelines
- Bug fixes and correctness issues — almost always Implement.
- Security concerns — always Implement unless clearly a false positive.
- Style nits — Implement if they match project conventions (check CLAUDE.md / gyrinx-conventions), otherwise Acknowledge.
- Suggestions to add features or scope — usually Acknowledge (out of scope for this PR).
- Copilot auto-review comments — evaluate on merit like any other review. Copilot can be wrong; don't implement something just because a bot said it.
- Conflicting feedback — flag it and ask the user.
Phase 3: Plan
For all items in the Implement bucket:
- Create a todo list with one item per change
- Group related changes that touch the same file
- Note which feedback items are addressed by each change
- Consider whether any changes require new tests
Present the plan to the user and wait for approval before proceeding.
Phase 4: Implement
After user approval:
- Work through the todo list, marking items as you go
- For each change, read the relevant source file first
- Follow project conventions (run
./scripts/fmt.shwhen done) - Run
pytest -n autoto verify nothing is broken
Phase 5: Summary
After implementation:
- Summarise what was changed and why
- List any feedback items you deliberately did not implement and the reason
- Suggest replying to reviewers if appropriate (but do not post replies without asking)
Converted and distributed by TomeVault — claim your Tome and manage your conversions.