Code Review Skill: Bugs First, Then Cleanups

A code review skill is the smallest agentic capability that pays for itself every PR. This one runs on the current diff, looks for actual bugs first, then surfaces reuse and cleanup opportunities — and bans the "looks good 👍" non-review reflex.

What It Does

• Reads only the diff — not the whole codebase, not the history
• Bugs before cosmetics — null/empty inputs, boundary conditions, swallowed errors, broken callers
• Reuse first — points to existing functions before suggesting new ones
• Flags premature abstractions — three similar lines beats one half-baked helper

Install in 30 Seconds

Pick your tool above and hit Download. Each platform gets a file pre-formatted for it:

• Claude Code: drop SKILL.md into ~/.claude/skills/code-review/ (or .claude/skills/code-review/ per project).
• OpenAI Codex CLI: append the downloaded file to your project's AGENTS.md.
• Cursor: append to .cursorrules at the project root.

Same content, three install paths. Reload and ask for a code review on your next change.

Why This Skill Beats "Just Ask"

When you ask an agent "review this code" without context, it tends to flatter the diff. This skill makes the agent earn the verdict: it must walk through the diff hunk by hunk, name file:line for every finding, and produce a clean "Clean — no findings" only when there genuinely is nothing. That's the difference between a review and a rubber stamp.

Install once, run before every push.