Best Code Review Skills for AI Coding Agents in 2026

Anthropic's code-review is the best code review skill for AI coding agents in 2026. It fans several specialized reviewer agents across a pull request and uses confidence-based scoring to filter the false positives that make single-pass AI review so noisy, and it ships from the anthropics/claude-code repository (135,651 GitHub stars as of July 4, 2026), the most-adopted first-party distribution channel in the ecosystem. The full top 7, spanning Claude Code, Codex, and OpenClaw, with verified install methods, is below.

This list is task-specific. For a general ranking across every category, see our best Claude Code skills hub. This post only covers skills whose job is reviewing code: catching bugs, security issues, and design problems before a change lands.

How We Ranked Code Review Skills

A code review skill is a packaged instruction set, either a SKILL.md file or a Claude Code plugin, that tells an AI coding agent how to review a diff for bugs, security flaws, and design problems. We ranked these by three verifiable signals: official first-party distribution, the parent repository's GitHub stars (checked July 4, 2026), and honest community adoption. Anthropic introduced Agent Skills in October 2025, described in its engineering post on equipping agents for the real world, and both Anthropic and OpenAI now ship review skills through official marketplaces.

We did not invent install counts. Where a skill lives inside a large monorepo, the star count reflects that parent repo, not the individual skill, and we say so. Community skills from ClawHub are listed with their real star counts even when those are in the single digits, because honest ranking matters more than inflated numbers.

1. code-review: Best Overall Code Review Skill

code-review is Anthropic's own multi-agent review capability, and it is the best code review skill for AI coding agents in 2026. Instead of a single model pass, it dispatches specialized reviewer agents over a pull request and applies confidence-based scoring, which directly attacks the biggest complaint about AI review: too many low-value comments. It installs from the anthropics/claude-code marketplace in two commands:

/plugin marketplace add anthropics/claude-code
/plugin install code-review@claude-code-plugins

Once installed you invoke it on a diff or PR and it returns grouped, scored findings rather than an undifferentiated wall of comments. It is the default we recommend for anyone who wants an automated review gate without hand-tuning a prompt.

2. pr-review-toolkit: Best Deep Review

pr-review-toolkit is the heavyweight sibling of code-review, built for thorough passes on large pull requests. It splits the job across agents that each specialize in one lens: comments, tests, error handling, type design, general code quality, and simplification. That separation is why it catches design-level issues a single reviewer skips, at the cost of more time and tokens per run. It ships from the same Anthropic marketplace, so the marketplace add above already covers it:

/plugin install pr-review-toolkit@claude-code-plugins

Use code-review as a fast gate on every PR and reserve pr-review-toolkit for the big, risky changes where a multi-angle read earns its cost.

3. Superpowers Review Skill: Best Built-Into-Workflow Review

Superpowers embeds code review inside the development loop rather than treating it as a separate step. The skills library by Jesse Vincent (obra/superpowers, 245,152 stars as of July 4, 2026) teaches Claude to delegate implementation to subagents and run a review before code is accepted, so problems surface during the build instead of after. It installs from Anthropic's official marketplace in one command:

/plugin install superpowers@claude-plugins-official

If your problem is not "review my PR" but "stop shipping code that needs a PR review in the first place," this is the higher-leverage install. We cover the full library in our best Claude Code plugins guide.

4. gh-address-comments: Best Codex Review Skill

gh-address-comments is OpenAI's curated Codex skill for the other half of review: resolving the comments a reviewer already left. It reads a pull request's review threads and works through them systematically instead of one reply at a time, and it pairs naturally with gh-fix-ci. It lives in the openai/skills catalogue (roughly 19,000 stars and 38 curated skills as of July 2026) documented at the official Codex skills docs. Add it to Codex from the curated set:

codex skills add gh-address-comments

Directory entry: gh-address-comments in our Codex skills index.

5. security-guidance: Best Real-Time Review Layer

security-guidance reviews code as it is written, not after. It is a hook-based Claude Code plugin that warns in real time when an edit introduces a risky pattern like command injection, cross-site scripting, or unsafe deserialization, so the embarrassing class of bug never reaches a pull request. It is not a replacement for a full review, but as an always-on layer it costs nothing and ships from the same anthropics/claude-code marketplace:

/plugin install security-guidance@claude-code-plugins

6. critical-code-reviewer: Best Adversarial Community Skill

critical-code-reviewer is a community SKILL.md skill built for one thing: adversarial review with zero tolerance for medium-and-above issues. Authored by ziad-hsn on ClawHub, it is deliberately harsh, which is useful when you want a second opinion that assumes the code is wrong until proven otherwise. Adoption is modest and honest at 1 star, so treat it as a specialist prompt rather than a battle-tested tool. Install it into OpenClaw:

clawhub install ziad-hsn/critical-code-reviewer

Directory entry: critical-code-reviewer.

7. audit-code: Best Secrets-and-Danger Review

audit-code narrows review to the highest-severity findings: hardcoded secrets, dangerous calls, and unsafe patterns. Built by itsnishi on ClawHub (2 stars as of July 2026), it is a focused, security-leaning review skill rather than a general reviewer, which makes it a fast pre-commit sweep before a heavier pass. Install it into OpenClaw:

clawhub install itsnishi/audit-code

Directory entry: audit-code. For the security angle in depth, see our best security and pentest skills guide.

Comparison Table

Star counts are for each skill's parent repository, checked July 4, 2026. Community skills show the individual skill's own count.

Limitations and Tradeoffs

AI code review is a second reader, not a merge authority. Even the confidence scoring in code-review reduces false positives rather than eliminating them, and every reviewer skill will occasionally miss a subtle logic bug or flag a harmless pattern, so a human still owns the merge decision. Skills also run with your agent's privileges and can execute code, so install community review skills only from sources you trust and read the SKILL.md first, as we cover in where to find Claude Code skills safely. Finally, deep skills like pr-review-toolkit consume real tokens; run the fast gate on every change and the heavy pass only where the risk justifies it.

Related Guides

• Best Claude Code Plugins in 2026
• Best Claude Code Subagents
• Best Security and Pentest Skills for AI Agents
• Best Places to Find Claude Code Skills