Code review

Skill rokurokulab/dotai/shared/skills/code-review

Cross-tool AI coding agent config registry & installer (Claude Code + Codex). Skills, hooks, agents, marketplaces, all in one place.

Install

npx -y skills add rokurokulab/dotai --skill code-review

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Review a diff (staged, branch, or PR) for correctness, hidden state changes, missing tests, and convention drift. Use this when the user asks for a review, second opinion, or sanity check on changes they're about to commit or merge.

SKILL.md

2.3 KB, as published. Nobody here has run it

code-review

A lightweight peer-review pass. Aims for the kind of comments a thoughtful colleague would leave — not exhaustive linting (the linter already does that) and not full-redesign feedback (out of scope).

When to invoke

User says "review this", "any issues with this change?", "second opinion before I merge"
User asks you to look at git diff, a PR URL, or a specific file/range
Don't auto-invoke on every Edit — too noisy

What to do

Read the change. Get the diff from git diff, git diff --cached, git diff <base>...HEAD, or gh pr diff <num>. If a PR URL was given, also read the PR description for declared intent.
Read 1–2 nearby files to understand local convention (naming, error-handling style, log format).
Pass the change through this checklist (skip categories that don't apply):
- Correctness: Does it do what the message claims? Edge cases? Off-by-one, nil/None, empty input, concurrent access.
- Hidden state changes: Touches global state, env vars, file system, network outside its declared scope?
- Tests: New behavior without a test? Removed behavior with a now-dead test?
- Backward compat: Public API broken without a major bump or deprecation note?
- Convention drift: Departs from AGENTS.md rules or local patterns without justification?
- Comment hygiene: Comments explain what (redundant) instead of why (useful)?
- Naming: Names readable a year from now? Abbreviations that won't be obvious?
- Errors: Swallowed errors? Lost context in re-raise? Generic error type for actionable failure?

Output format

Group findings into Must fix (correctness, security, broken tests) and Consider (style, naming, cleanups). Include file:line references. End with a one-sentence overall take ("Looks good apart from the two must-fix items" / "Suggest reworking the error handling, then re-review").

Don't pad with praise. If there's nothing significant to flag, say so in one line and stop.