code-review

Coordinates parallel specialist agents to review code and aggregates findings into a prioritized report.

Updated Nov 7, 2023
One-click install
npx skills add https://github.com/lstellway/nixfiles --skill code-review-lstellway
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: code-review
Source: https://github.com/lstellway/nixfiles/tree/main/modules/home-manager/claude/skills/code-review
Command: npx skills add https://github.com/lstellway/nixfiles --skill code-review-lstellway

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Reviewing code, diffs, or design proposals from a single perspective misses domain-specific issues like security flaws, accessibility gaps, or architecture problems. This Skill orchestrates multiple specialist review agents and merges their output into one prioritized report. ## Core Features & Use Cases - Agent Selection & Confirmation: Scans available domain and technology agents, proposes a relevant set based on the target's content, and waits for explicit user confirmation before dispatching. - Parallel Fan-Out: Invokes all confirmed agents simultaneously in a single parallel call, with prompts tailored to whether the target is existing code or a proposed design. - Unified Aggregation: Deduplicates cross-cutting findings cited by multiple agents, collects Critical and High findings into a Priority Findings section, and merges What's Working and Open Questions across agents. - Use Case: Paste a pull request diff and receive a single report combining security, testing, architecture, and technology-specific feedback, sorted by severity with per-agent detail sections. ## Quick Start Ask the assistant to review a file, diff, or design proposal using the code-review skill and confirm the proposed set of specialist agents.

Frequently Asked Questions about code-review

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

FAQPage Schema
How do I run a multi-agent code review on a pull request?

Provide the PR diff or file paths as the review target. The skill proposes relevant specialist agents based on the content, waits for your confirmation, then runs them in parallel and merges results into one prioritized report.

How are review agents selected for a target?

Selection starts with broad domain agents like security, code quality, testing, and architecture, then adds specialists based on what the target touches, such as accessibility for UI or technology agents matching file types like .tf or .tsx.

Can I change which agents run before the review starts?

Yes. After the proposed agent set is presented, you can modify it with natural language, such as asking to drop or add specific agents. No agents are dispatched until you explicitly confirm the set.

Does this work for design proposals or only existing code?

It handles both. For proposed designs, the skill favors architecture and API-design agents over implementation-focused ones, and prompts agents to evaluate the design rather than review code.

How are duplicate findings from multiple agents handled?

When two or more agents cite the same file and location, the finding is marked as cross-cutting with the contributing agent names, deduplicated, and shown once at the top of the Priority Findings section.