Claude Code

Verified against Claude Code · 2026-07-25

Turn Claude into a consistent code reviewer instead of a vibes check

A fixed-rubric code review prompt with six named categories and a forced pass or fail verdict, so every pull request gets the same systematic review instead of whatever the model happens to notice first.

Claude CodeCursorGitHub CopilotWindsurf

The prompt

Ready to copy — highlighted parts are example details you can swap.

You are reviewing this change as a senior engineer whose name goes on the approval. Cite exact line numbers or hunks. Do not soften an actual defect into a "just a suggestion" — say plainly whether it blocks merge.

REVIEW CATEGORIES
1. Correctness — does the logic do what the pull request claims? Any off-by-one, race condition, null or undefined path, or missed edge case?
2. Security — injection risk, auth or authorization gaps, secrets in code, unsafe deserialization, unvalidated input crossing a trust boundary.
3. Tests — do the tests actually exercise the changed behavior, or just re-assert the implementation? Is there a case the diff clearly needed but did not add?
4. Readability and naming — would a new engineer understand this in six months without asking the author?
5. Scope — does this diff do only what the description says, or is there an unrelated drive-by change bundled in?
6. Performance — any new N+1 query, unbounded loop, or blocking call on a hot path?

OUTPUT FORMAT
For each category: PASS, CONCERN, or BLOCKING, one line of reasoning, and a file or line reference if not PASS.
End with one verdict: APPROVE, APPROVE WITH COMMENTS, or REQUEST CHANGES. No verdict is valid without a one-sentence justification.

Paste the diff or files under review below this line, unedited.

Why this works

A generic review this pull request prompt reliably produces a different-shaped answer every time, sometimes a paragraph on style and nothing on security, because the model has no fixed frame to fill in. Giving it six named, closed categories forces systematic coverage of the same ground on every review, the same mechanic behind Claude Code's own built-in security review command scanning by category rather than freeform. Requiring a pass, concern, or blocking rating per category, not prose, also prevents the single most common failure mode of AI code review: a vague looks good overall that never actually commits to a position on any specific line. Forcing a final verdict with a one-sentence justification closes the last gap — a model asked only to review will often describe the code back to you instead of judging it, while a forced choice between approve, approve with comments, or request changes requires it to take a stance.

What you get back

1. Correctness: PASS. 2. Security: PASS. 3. Tests: CONCERN — the new discount path in lib/checkout/discount.ts has no test for an expired code; add one before merge. 4. Readability and naming: PASS. 5. Scope: BLOCKING — this diff also reformats unrelated-file.ts with no stated reason; split it out or explain it. 6. Performance: PASS. Verdict: REQUEST CHANGES — the missing expired-code test and the unexplained unrelated file change should be resolved before this merges.

Verified against

Claude Code Sonnet 4.6 · 2026-07-25

GitHub Copilot Copilot Chat 1.250 (VS Code) · 2026-07-18

Changelog

  • 2026-07-25 Initial publish, verified against Claude Code (Sonnet 4.6) and GitHub Copilot Chat.

Building this for real?

This is a free starting point. If you'd rather have custom software built and running for your business, that's Scult's day job.

EXPLORE CUSTOM SOFTWARE
All Claude Code prompts

Check your AI visibility

One URL in, a 0–100 score and the exact fixes out.

RUN THE CHECK

Browse all the tools

15 tools across six categories
13 of them never send your data anywhere

Free · No signup · No trial clock

SEE THE DIRECTORY