code-review

Reviews code diffs for correctness, contract safety, and structural quality with cited findings.

Updated May 13, 2026
One-click install
npx skills add https://github.com/andrewcodesit/skills --skill code-review-andrewcodesit
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: code-review
Source: https://github.com/andrewcodesit/skills/tree/main/skills/engineering/code-review
Command: npx skills add https://github.com/andrewcodesit/skills --skill code-review-andrewcodesit

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Code reviews of agent-written or human-written changes tend to be either rubber-stamps or nitpick floods, missing real bugs while drowning reviewers in style noise. This Skill runs a strict, senior-engineer-grade review that focuses on correctness, contract breaks, and structural regressions, and saves every finding with exact file:line citations and pasteable fixes. ## Core Features & Use Cases - Diff-driven review: Pulls a PR/MR diff via gh or glab, or diffs the current branch against its merge-base, then reads every changed file in full plus surrounding context. - Structured findings report: Groups findings into Issues, Rules & Conventions, Refactor Opportunities, Quick Wins, and Bigger Picture, each with verbatim current code and a compilable suggested replacement, saved to ~/.agents/code-reviews/. - Resolution gate: Ends every run with a fixed lettered disposition menu (grill on ambiguous findings, hand off to planning, apply best judgment, or do nothing) and records all decisions in the review file. - Use Case: After finishing a feature branch, ask for a review before opening the PR. The Skill audits both sides of every touched boundary, flags a contract break in your API handler with a concrete fix, and lets you approve each remediation one by one. ## Quick Start Ask the agent to review the current branch or a specific PR number, for example: review the changes on this branch and flag any real bugs or design problems.

Frequently Asked Questions about code-review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a pull request with an AI agent?▼

Give the agent a PR or MR number and it fetches the diff with gh or glab, reads every changed file in full plus surrounding context, and produces a severity-ordered report. Findings are saved to a markdown file with exact file:line citations and suggested fixes.

How to review code changes on the current branch before merging?▼

The Skill diffs the current branch against its merge-base with main, master, or origin/HEAD, then audits the changes for correctness, contract breaks, and structural regressions. It stops cleanly if there is no diff to review.

What kinds of issues does an automated code review catch?▼

It prioritizes silent correctness and contract bugs, structural regressions, spaghetti growth from one-off branches, logic in the wrong layer, missing boundary validation, and file-size sprawl. Style-only nitpicks are deliberately excluded in favor of high-conviction findings.

Can the review be delegated to a subagent to save context?▼

Yes, when a delegation mechanism exists the review itself runs in a subagent that reads the diff and report-format reference, then returns only the saved path and a findings count. The resolution gate always stays with the main agent.

What happens after the code review findings are reported?▼

Every run ends with a resolution gate offering fixed dispositions: grill on ambiguous findings, grill on every finding, hand off to the planning skill, apply best judgment, or do nothing. All decisions are recorded in a Decisions section of the saved review.

When should I not use this strict code review approach?▼

It is not suited to reviewing its own output immediately after implementation, since an agent re-confirms its own assumptions when the plan is still in context. Run it as a separate pass over the finished diff instead.