synesenom/ran / .claude/skills/review-pr
review-pr
VerifiedPR Review Command
Scope: This skill is for reviewing externally submitted GitHub PRs (pull requests opened by contributors on GitHub). It is entirely distinct from:
/review— reviews local branch changes before committing; used inside the/buildpipelinereview-security,review-correctness, etc. — internal quality-check subagents spawned by/reviewand/build; never invoked by users directlyDo not use this skill during task resolution. It is only for triaging incoming contributor PRs.
You are reviewing a GitHub PR and posting a binding decision: approve + merge if all quality checks pass, or request changes if any blocking issue is found.
Core Principle
Catch issues CI cannot. The output is a GitHub review action posted to the PR, not a chat report. Every invocation ends with either an APPROVE+merge or a REQUEST_CHANGES comment on GitHub.
Workflow
When the user invokes /review-pr <PR-URL-or-number>:
1. Parse Input
Extract owner, repo, and PR number:
- Full URL
https://github.com/<owner>/<repo>/pull/<N>→ parse all three fields - Bare number → default to
synesenom/ran
2. Fetch PR Context (parallel)
Use the GitHub MCP tools in a single parallel call:
mcp__github__pull_request_read(method="get")— title, state, draft flag, head SHA, bodymcp__github__pull_request_read(method="get_diff")— full unified diffmcp__github__pull_request_read(method="get_files")— changed file list with stats
Guard rails — stop before doing any review if:
state !== "open"→ "PR #N is already<state>. Nothing to do."draft === true→ "PR #N is a draft. Mark it ready for review first."- diff is empty → "PR #N has no diff. Nothing to review."
3. Check Out the PR Branch and Run the Suite
Fetch the PR head into a local ref, run lint + tests, then return to the original branch:
git fetch origin pull/<N>/head:pr-<N>-review
git checkout pr-<N>-review
npm run standard 2>&1
npm test 2>&1
git checkout -
git branch -D pr-<N>-review
Run lint and tests sequentially (tests depend on lint passing). Capture full output.
If either command exits non-zero, record a Block finding:
[tests] CI—npm run standardfailed:<first error line(s)>[tests] CI—npm testfailed:<failing test names and coverage breach lines>
Do not stop — continue to Pass 1 and Pass 2 so the full picture is reported in one review comment.
4. Save Diff
mkdir -p .claude/tmp
# write the diff to a temp file so review agents can read it
Save the diff text to .claude/tmp/review-diff-pr-<N>.patch.
5. Pass 1 — Convention Compliance (against diff)
External PRs have no local plan file, so Pass 1 checks project conventions from CLAUDE.md directly against the diff:
| Check | Fail condition |
|---|---|
| CHANGELOG | User-visible change (bug fix, new feature, removed code) present but no CHANGELOG.md hunk in the diff |
| Debug leftovers | console.log, TODO, FIXME, or large commented-out code blocks added |
| Comment style | Comments that describe what the code does rather than why (WHAT comments are flagged; WHY comments are fine) |
| Distribution constants | this.c = [...] positional array instead of this.c = { name: value, ... } named object |
| Subclass constants | Leaf-subclass uses this.c = { ... } when a parent already sets this.c (should be Object.assign) |
| Error conventions | undefined returned as an error sentinel (should be NaN, Infinity, or throw) |
Each failed check is a Warn finding. Record them alongside the agent findings from Pass 2.
6. Pass 2 — Code Quality (Parallel Agents)
CRITICAL: Launch exactly 8 review agents in a single parallel call. Verify all 8 returned before proceeding.
Tell each agent to read .claude/tmp/review-diff-pr-<N>.patch and provide enough diff context so they can assess without needing extra file access. Each agent returns findings rated Block/Warn.
- review-security agent
- review-performance agent
- review-structure agent
- review-conventions agent
- review-impact agent
- review-tests agent
- review-docs agent
- review-correctness agent
Wait for all 8. Deduplicate overlapping findings. If you downgrade a finding's severity, note the original severity and your rationale.
7. Verdict
PASS — zero Block findings across both passes. Warn items are informational and do not block.
FAIL — one or more Block findings.
8. Act on Verdict
PASS → Approve and merge
Post an APPROVE review:
mcp__github__pull_request_review_write(
method="create",
event="APPROVE",
commitID=<head SHA>,
body=<approval body — see format below>
)
Then immediately merge (squash):
mcp__github__merge_pull_request(
method="squash" (or the equivalent squash parameter)
)
FAIL → Request changes
Post a REQUEST_CHANGES review:
mcp__github__pull_request_review_write(
method="create",
event="REQUEST_CHANGES",
commitID=<head SHA>,
body=<findings body — see format below>
)
Do not merge.
9. Review Body Format
Approval body (PASS)
LGTM — all quality checks passed.
<One sentence describing what the PR does and why it looks correct.>
<If any Warn items exist:>
**Notes (non-blocking):**
- [domain] file:line — <note>
Request-changes body (FAIL)
<If Block items:>
## Required (blocking)
- [ ] [domain] `file:line` — <description and exact fix>
<If Warn items:>
## Informational (non-blocking)
- [domain] `file:line` — <note>
10. Report to User
After posting to GitHub, report back:
PASS:
PR #N approved and merged. Warn notes (if any):
FAIL:
PR #N: changes requested — blocking issue(s).
Rules
DO:
- Fetch the full diff before running any agent
- Launch all 8 review agents in a single parallel call and wait for all 8
- Post a GitHub review (APPROVE or REQUEST_CHANGES) — never just a comment
- Merge immediately and automatically after approving — this is the contract
- Be specific: cite file path and line number for every finding
DO NOT:
- Approve a PR that has any Block finding
- Merge without first posting an APPROVE review
- Merge a draft PR or a non-open PR
- Post REQUEST_CHANGES when the PR passes all checks
- Poll GitHub with sleep — act and report, then end the turn
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |