lean-pr-review
Review a pull request using Lean waste categories and structured severity ratings
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Review a pull request using Lean waste categories and structured severity ratings
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Audit VS Code extensions against the current project stack and recommend keep/add/remove actions
Configure and manage Model Context Protocol servers for external tool access
Set up browser testing — dual path with built-in browser tools (interactive) or Playwright (CI); detect framework, scaffold, verify
| name | lean-pr-review |
| description | Review a pull request using Lean waste categories and structured severity ratings |
| compatibility | >=1.4 |
Skill metadata: version "1.1"; license MIT; tags [review, pull-request, lean, kaizen, code-review]; compatibility ">=1.4"; recommended tools [codebase, githubRepo].
Perform a structured pull request review using §5 Review Mode conventions and §6 waste categories.
Get the diff — Read the PR diff or the set of changed files. If working locally, use git diff main...HEAD or the equivalent for the target branch.
Scan each changed file — For every file in the diff, read the full file (not just the diff hunk) to understand context.
Classify each finding — For every issue found, record:
[severity] | [file:line] | [waste category] | [description]
Severity levels:
critical — blocks merge; security flaw, data loss risk, or broken functionalitymajor — should fix before merge; logic error, missing test, or significant smellminor — nice to fix; style issue, naming, minor inefficiencyadvisory — informational; suggestion for future improvementWaste categories (§6) — full list W1–W16; most common in PR review:
| Code | Name | Typical PR signal |
|---|---|---|
| W1 | Overproduction | Dead code, unused exports, features not yet needed |
| W2 | Waiting | Blocking sync calls, missing timeouts |
| W3 | Transport | Unnecessary data copying, prop drilling 3+ levels |
| W4 | Over-processing | Abstraction for its own sake, premature generalisation |
| W5 | Inventory | Large WIP; changes that could be split into smaller PRs |
| W6 | Motion | Logic scattered across many files without justification |
| W7 | Defects | Bugs, type errors, missing error handling, test failures |
| W8 | Unused talent | Missing tests, missing automation, repetitive manual patterns |
| W11 | Hallucination rework | Phantom API usage, methods that don't exist, incorrect assumptions |
| W14 | Model-task mismatch | Overly complex solution to a trivial problem |
For W9–W10, W12–W13, W15–W16 definitions, see §6 of .github/copilot-instructions.md.
Check test coverage — Verify that new or changed behaviour has corresponding tests. Flag untested paths as major | W7 Defects.
Check for baseline breaches — Compare against §2 baselines:
Produce the report — Format as:
## PR Review — <PR title or branch name>
### Summary
<1–2 sentence overview of the changes and their quality>
### Findings (<N> total: <critical> critical, <major> major, <minor> minor, <advisory> advisory)
#### Critical
- [critical] | [file:line] | [W7] | <description>
#### Major
- [major] | [file:line] | [W4] | <description>
#### Minor
- [minor] | [file:line] | [W1] | <description>
#### Advisory
- [advisory] | [file:line] | [W8] | <description>
### Verdict
<APPROVE / REQUEST CHANGES / COMMENT>
Wait — Do not apply fixes. Present the report and wait for the user to decide what to address.