Code Review

Why It Made the Cut

Across 8 code-review skills, this one skips the pleasantries and filler: it only pinpoints lines with logic flaws and returns the improved code directly. It grades by severity and forces code citations, which makes it perfect for pre-commit automated review in CLI and agent workflows.

Problem & Use Cases

• Input: A piece of code or a git diff (single file or a small changeset).
• Output: A severity-graded list of issues, each with a minimal fix.
• Good for: Pre-commit automated review, quality gates in CLI / agent workflows, and teams that want "less talk, just fixes".
• Not for: Large-scale architecture reviews that need business context, or pure style-preference feedback.

How to Use

1. Paste the code or diff into a Claude Code / Codex session.
2. Prefer a diff over the whole repo; include language version and key constraints.
3. P0/P1 issues must be reproduced by a human before blocking a merge.

You are a rigorous senior code reviewer.

Review dimensions: correctness, security, performance, maintainability, test coverage.
Report only issues with clear evidence; ignore pure personal style.

Output each issue in this format:
[Severity P0-P3] Issue title
- Location: file and line
- Impact: what happens under which conditions
- Evidence: quote the relevant code
- Fix: the minimal change required

End with: merge recommendation (block / merge after fixes / mergeable)

Code or diff:
{paste here}

In-Scenario Comparison

• Runner-up A: floods the output with "this code looks great" pleasantries, burying real issues.
• Runner-up B: only gives style advice without locating lines, so fixes cannot be applied.

Boundaries & Known Issues

• For AI-generated code, beware plausible-but-wrong snippets; reproduce with tests.
• P0 security issues must be confirmed by a human before merging.
• Split very long diffs to avoid context truncation and missed findings.

Changelog