Code Review Skill for Claude

<!--

Source: anthropics/claude-code (official)

File: https://github.com/anthropics/claude-code/blob/main/plugins/code-review/commands/code-review.md

Also: https://github.com/anthropics/claude-code/blob/main/plugins/code-review/README.md

Used by: reviewer agent

-->

Code Review

When to trigger

• Immediately after new or modified code
• Before commit / PR
• When user asks for /code-review

Review approach

Multi-agent review with confidence filtering:

• CLAUDE.md compliance — does the code follow project rules?
• Bug detector — obvious bugs in the diff
• History analyzer — git context to understand intent

Each issue is scored 0–100. Findings below threshold (default 80) are filtered out.

What to check

Clarity

• Names (variables, functions, files) — do they say what they do?
• Structure — is it easy to scan?
• Cognitive load — how much do you hold in your head to understand it?

Correctness

• Edge cases: null, undefined, empty array, boundary values
• Error paths: are they handled? Do they fail loudly or silently?
• Async: await missing? Race conditions? Promise rejections caught?
• Off-by-one errors in loops / slicing

Performance

• N+1 queries (especially around ORM usage)
• O(n²) algorithms on large collections
• Unnecessary re-renders (React)
• Missing memoization for expensive computations
• Synchronous I/O in hot paths

Consistency

• Matches project style guide (CLAUDE.md, rules/)
• Immutable patterns used
• Function size under 50 lines
• Nesting 4 levels max

Tests

• Is there a test for the new behavior?
• Edge cases covered?
• Mocks realistic (no "happy path only")?
• Tests independent (no shared state)?

Dead code

• Unused imports
• Unreachable branches
• TODO comments that should be done or deleted
• Commented-out code

Severity levels

• CRITICAL — do not commit. Breaks production, security hole, data loss risk.
• HIGH — fix before next release. Real bug users will hit.
• MEDIUM — fix when you're in the area. Code smell, minor inefficiency.
• LOW — nice-to-have. Style preference, optional refactor.

Output format

## CRITICAL (must-fix)
- [`<file>:<line>`] <issue>
  - Why: <impact>
  - Fix: <specific suggestion>

## HIGH
- ...

## MEDIUM / LOW
- ...

## Summary
<N CRITICAL · N HIGH · N MEDIUM · N LOW>
<Verdict: PASS | FAIL>

Rules

• Every finding has: severity, file + line, what's wrong, why it matters, suggested fix.
• Don't just flag — propose a fix.
• Don't be timid on CRITICAL. If it's critical, say so clearly.
• If you see a pattern (same issue in multiple places), flag it once with a list.
• Use --comment mode for PR review, default for local output.