Code Review Master

Code Review Master turns Claude into a senior reviewer for your pull requests. It reads a diff, reasons about each changed file, and produces inline, line-anchored comments covering correctness, security, performance, test coverage, and style drift — plus a summary you can paste straight into a GitHub review. It's built for engineers who want a thorough first pass before a human teammate spends time, and for solo maintainers who don't have a reviewer on call.

When to use it

• Before requesting human review, to catch the obvious stuff (missing null checks, unhandled errors, untested branches) so reviewers focus on design.
• On a teammate's PR when you're the reviewer and want a structured starting point.
• After a large refactor, to flag behavior changes and places where tests weren't updated.
• When touching auth, input parsing, SQL, or file/network I/O and you want a security-leaning read.

How it works

Point it at a diff and it works file by file. It runs git diff (or reads a gh pr diff <number> against the base branch) and analyzes only the changed hunks, with surrounding context pulled from the working tree so comments aren't made blind. Findings are grouped by category — correctness, security, performance, tests, style — and each carries a file path, line range, severity, and a concrete suggested fix. The output is GitHub-flavored Markdown: a top-level summary plus per-file sections, so you can drop it into a review or post individual comments.

Usage

curl -fsSL --create-dirs -o ~/.claude/skills/code-review-master/SKILL.md https://claudeskil.com/skills/code-review-master/SKILL.md

# Review your current branch against main
git checkout feature/checkout-refactor
claude "review this branch against main with code-review-master"

# Or review a specific PR
gh pr checkout 482
claude "run code-review-master on PR #482, focus on security and tests"

A finding looks like:

src/api/orders.ts:88-94  [security · high]
User-supplied `sortBy` is interpolated into the SQL string — injectable.
Suggestion: whitelist column names or use a parameterized ORDER BY.

Tips & gotchas

• Give it the base branch. Reviewing a diff against main produces far sharper results than reviewing a single commit, which lacks context.
• It reviews the diff, not the whole repo. If a bug lives in unchanged code that the diff merely calls, ask it explicitly to widen scope.
• Scope the categories ("security and performance only") on large PRs — an unfocused review on 50 files dilutes the high-severity findings.
• Treat severity as a hint, not gospel. It errs toward flagging; triage before pasting everything into a thread.

FAQ

Does it post comments to GitHub automatically? No. It generates review-ready Markdown; you paste it or post comments yourself via gh. This keeps a human in the loop on what actually lands on the PR.

Will it run my tests or linter? It reasons statically about coverage and style from the diff. It doesn't execute your suite — pair it with CI for actual pass/fail, and use it to find the cases CI isn't testing yet.

What languages does it handle? Any language Claude reads well — TypeScript, Python, Go, Java, Rust, and more. Quality tracks how much surrounding context the working tree provides.