requesting-code-review ยท git:20260519.40bb110 ยท 2026-05-19 ยท sha256 9cb9e31cbcbc6671

requesting-code-review git:20260519.40bb110A

Immutable. This exact content never changes and is served at /api/v1/blob/9cb9e31cbcbc6671.

No tool to install first
claude plugin marketplace add https://markdownregistry.com/mp/a/art_z7wv2os74zg2ejwl/9cb9e31cbcbc6671679999120e640bb5d3044f698b41d9f7581e5aeae3e9c232.json

Then /plugin install requesting-code-review@markdownregistry-z7wv2os74zg2ejwl in Claude Code. It installs this exact version, the one graded A here, pinned to the commit it was graded at, and leaves it there.

---
name: requesting-code-review
description: Self-reviews work against the plan before sign-off.
tier: practical
category: workflow
created_by: human
platforms: [windows, macos, linux]
tags: [self-review, quality, gate]
author: Andreas Wasita (@andreaswasita)
---

# Requesting Code Review Skill

Self-reviews the agent's own output against the original plan before declaring done or opening a PR. Surfaces issues by severity; critical issues block progress. Does NOT replace human review โ€” it ensures the human review starts from a defensible baseline.

## When to Use

- A task or batch of tasks is complete.
- Between major implementation phases.
- Before creating a pull request.
- Before telling the user "done" on any non-trivial change.
- After a sub-agent returns delegated work.

## Prerequisites

- An approved plan in `tasks/todo.md` (or equivalent design doc).
- The `view` and `grep` tools to read the diff and surrounding code.
- `git` for the diff inspection.
- A green test run (the `verify-before-done` skill should already have run).

## How to Run

```text
1. Compare implementation to the plan spec, criterion by criterion.
2. Walk the self-review checklist.
3. Tag each finding ๐Ÿ”ด / ๐ŸŸก / ๐ŸŸข / โœ….
4. Act on findings โ€” ๐Ÿ”ด blocks, ๐ŸŸก fixed before PR, ๐ŸŸข logged.
5. Emit a short summary before claiming completion.
```

## Quick Reference

| Severity | Meaning | Action |
|---|---|---|
| ๐Ÿ”ด Critical | Wrong behavior, security risk, broken test | STOP. Fix before next task. |
| ๐ŸŸก Major | Should fix before merge | Fix before PR; OK to continue current task |
| ๐ŸŸข Minor | Style, nit, future improvement | Log to `tasks/todo.md` follow-ups |
| โœ… Passing | Criterion met | Note in summary |

| Checklist item | Look at |
|---|---|
| Matches plan spec | `tasks/todo.md` |
| Tests pass (including new) | Latest test run output |
| No debug artifacts | `grep` for `console.log`, `print`, `TODO` |
| Edge cases covered | The new test file |
| No security regressions | Auth, input validation, secrets |
| Follows project conventions | Adjacent files |
| Diff scope = task scope | `git diff main --stat` |

## Procedure

### Step 1: Compare to Plan

Open `tasks/todo.md` with `view`. For the task just completed, read its acceptance criteria. Tick each one explicitly โ€” do not summarize.

### Step 2: Walk the Checklist

For each row in the Quick Reference checklist, mark โœ… or flag a finding. Use `grep` to hunt for debug artifacts:

```bash
grep -nE "console\.log|print\(|TODO|XXX|FIXME" <changed files>
```

(Use the `grep` Copilot tool, not bare shell `grep`.)

### Step 3: Tag by Severity

Group findings under ๐Ÿ”ด / ๐ŸŸก / ๐ŸŸข / โœ…:

```markdown
## Self-Review

### ๐Ÿ”ด Critical
- `api/users.ts:42` โ€” DELETE endpoint missing authorization check.

### ๐ŸŸก Major
- `api/users.ts:88` โ€” Returns 500 instead of 404 for missing user.

### ๐ŸŸข Minor
- `api/users.ts:120` โ€” Consider pagination on the list endpoint.

### โœ… Passing
- 4/4 endpoints match spec.
- 12 new tests, all green.
- Diff scope = task scope.
```

### Step 4: Act on Findings

- ๐Ÿ”ด โ†’ STOP. Fix immediately. Re-verify. Re-review.
- ๐ŸŸก โ†’ Fix before the PR. Safe to continue the current task.
- ๐ŸŸข โ†’ Log in `tasks/todo.md` under follow-ups.
- โœ… โ†’ Note in the summary and proceed.

### Step 5: Present the Summary

Always emit a short visible summary before "done":

```markdown
**Review: Step 3 โ€” Implement API endpoints**
- โœ… 4/4 endpoints per spec
- โœ… Input validation on all routes
- ๐ŸŸก Missing rate limit on POST /users (logged for Step 6)
- โœ… 12 new tests, all passing
Proceeding to Step 4.
```

## Pitfalls

- **DO NOT** skip self-review and say "it works, ship it".
- **DO NOT** downgrade severity to avoid extra work. A missing auth check is ๐Ÿ”ด, not ๐ŸŸข.
- **DO NOT** review only the happy path. Walk the error and edge paths too.
- **DO NOT** proceed past a ๐Ÿ”ด finding. The whole point is that criticals block.
- **DO NOT** rubber-stamp ("โœ… all good") without actually checking each item.

## Verification

- [ ] Each acceptance criterion in the plan is explicitly ticked or flagged.
- [ ] Self-review summary was emitted with severity buckets.
- [ ] Any ๐Ÿ”ด findings were fixed and re-verified before claiming done.
- [ ] ๐ŸŸข findings were logged to `tasks/todo.md` follow-ups.
- [ ] User saw the summary before any "done" claim.