[No.136] 3 God-tier Claude Code Review Prompts — PR Diff Template Practical Pack
For those who find PR reviews ending in a diff nightmare, I have compiled 3 types of prompts for Claude that you can use by copy-pasting.
By specifying the perspective, output format, and reader, you can easily standardize the quality and speed of your feedback.
What this article includes
Prompts ①②③ — Full text ready to use by just replacing the square brackets `[…]`
This week's PR checklist — 5 items to fill out before pasting
Snippets save labels — Short names for IDE / Notion
3 common mistakes — Dumping, no diffs, no perspective
One move to try this week
This week's PR — Paste Prompt ① and use the output directly as a draft for your review comments.
Prompt ① — Security, Bugs, Performance (3 points each)
あなたはシニアエンジニアです。以下のコード/差分をレビューしてください。
【観点】
- セキュリティ: 重大度順に最大3点
- バグ: 再現しうるものを最大3点
- 性能: ボトルネックになりうるものを最大3点
【出力形式】
各観点を見出し＋箇条書き。各項目に「重大度（高/中/低）」と「修正案1行」を付ける。
【コード/差分】
[ここに diff または該当ファイルを貼る]Prompt ② — Inferring change intent → Listing risks
以下は git diff のみです。コミットメッセージはありません。
1. 変更の意図を2文で推測する
2. 意図と実装のズレがありうる箇所を列挙
3. 本番で起きうるリスクを優先度順に5つ
【diff】
[ここに diff を貼る]Prompt ③ — Explanation for Juniors + Improvement suggestions
以下のレビュー指摘を、Juniorエンジニア向けに説明してください。
- 専門用語は平易語に置き換える
- 「なぜ問題か」「どう直すか」を各指摘2文以内
- 最後に「学びの要点」を3行
【指摘または diff】
[ここに貼る]Checklist before pasting
[ ] Diff only or did you add one line of necessary context (file name, purpose)?
[ ] Secret information (API keys, personal data) masked?
[ ] Perspective written in the prompt? (No dumping allowed)
[ ] Is it based on the premise that a human will do a final check instead of merging as-is?
[ ] Have you saved good outputs to Snippets?
Snippets save names (example)
`claude-review-sec-bug-perf`
`claude-review-diff-risk`
`claude-review-junior-explain`
Reference
Disclaimer
This is an introduction to general-purpose prompt techniques. Please follow your company's policy regarding the external transmission of confidential code. Claude's output must always be verified by a human.
Edo Tech Kawaraban — Survival Strategy in the AI Era