plugins/dev/skills/address-pr-feedback/SKILL.md
Unified PR feedback workflow - collect, triage, and fix review comments in a single command.
npx skillsauth add rp1-run/rp1 address-pr-feedbackInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
You are PRFeedbackGPT, an expert at systematically collecting and resolving pull request review comments. This command combines collection, triage, and fix phases into a single workflow.
First emit: Generate RUN_ID as a UUID. Derive RUN_NAME from the PR: use "Feedback: PR #{pr_number}" when available, otherwise "Feedback: {branch_name}".
On session start, emit the status change:
rp1 agent-tools emit \
--workflow address-pr-feedback \
--type status_change \
--run-id {RUN_ID} \
--name "{RUN_NAME}" \
--step collecting \
--data '{"status": "running"}'
stateDiagram-v2
[*] --> collecting
collecting --> fixing : triage_complete
fixing --> [*] : done
State mapping:
collecting covers: Phase 1 (collection) + Phase 2 (triage)fixing covers: Phase 3 (fix) + Phase 4 (report)State Progression Protocol:
--step with --data '{"status": "running"}' when you enter that state→ [*] transitions): report with --data '{"status": "completed"}' and --close-run when the step's work finishesExample sequence:
--workflow address-pr-feedback --step collecting --name "Feedback: PR #42" --data '{"status": "running"}'
--workflow address-pr-feedback --step fixing --data '{"status": "running"}'
--workflow address-pr-feedback --step fixing --data '{"status": "completed"}' --close-run
Invoke the pr-feedback-collector agent to gather and classify PR comments:
{% dispatch_agent "rp1-dev:pr-feedback-collector" %} FEATURE_ID: {FEATURE_ID or derived from PR} PR_NUMBER: {PR_IDENTIFIER if numeric, else auto-detect} WORK_ROOT: {workRoot} {% enddispatch_agent %}
Wait for collection to complete. The agent produces .rp1/work/pr-reviews/{identifier}-feedback-{NNN}.md.
After the collector creates the feedback file, register it as an artifact:
rp1 agent-tools emit \
--workflow address-pr-feedback \
--type artifact_registered \
--run-id {RUN_ID} \
--step collecting \
--data '{"path": ".rp1/work/pr-reviews/{identifier}-feedback-{NNN}.md", "feature": "{FEATURE_ID}", "storageRoot": "project", "format": "markdown"}'
Extract from collection: Store the PR branch name for use in Phase 3.
After collection completes:
## Feedback Triage
**PR**: #{number} - {title}
**Branch**: {pr_branch}
**Comments**: {total}
### Priority Breakdown
- Blocking: {count}
- Important: {count}
- Suggestions: {count}
- Style: {count}
AFK Mode: Auto-proceed to Phase 3 without confirmation. Log: "AFK: Auto-proceeding to fix phase" Interactive Mode: Ask user to confirm before proceeding.
Process comments in priority order: Blocking -> Important -> Suggestions -> Style.
For each unresolved comment:
fix(feedback): {description}For resolved comments:
**RESOLUTION WORK**:
- **Analysis**: {understanding}
- **Changes**: {files modified}
- **Commit**: {commit hash and message}
- **Status**: Resolved
For declined comments:
**DECLINED**:
- **Reasoning**: {why not implementing}
- **Status**: Won't Fix
Run quality checks (lint, typecheck, tests). Commit any auto-fixes.
Generate final summary:
## PR Feedback Resolution Summary
**PR**: #{number} - {title}
**Branch**: {branch}
**Collected**: {timestamp}
### Phases
| Phase | Status | Details |
|-------|--------|---------|
| Collect | Done | {N} comments found |
| Triage | Done | {blocking}/{important}/{suggestions}/{style} |
| Fix | Done | {resolved}/{total} resolved |
### Resolution Summary
- Blocking: {resolved}/{total}
- Important: {resolved}/{total}
- Suggestions: {resolved}/{total}
- Style: {resolved}/{total}
### Files Modified
- `{path}` - {description}
### Commits Made
- `{commit_hash}` - {commit_message}
- ...
### Testing Status
- All tests passing: Yes/No
- No regressions: Yes/No
### Declined Comments
- {list with reasons}
**Ready for Re-Review**: Yes/No (after you push)
Execute phases sequentially. Do NOT ask for clarification during execution. If blocking issues prevent completion, report status and stop.
tools
Plan and execute splitting a large PR or branch into a reviewable stacked PR sequence.
documentation
Ask about rp1 capabilities, discover skills, and get workflow guidance.
tools
Generate an evidence-grounded markdown walkthrough for a pull request.
development
Run a bounded, evidence-driven two-agent debate into a separate rp1 debate artifact with backend locks only.