How to use the Code Review Excellence agent skill

What does the Code Review Excellence skill make the agent do?

When the skill is installed, the agent follows a structured code review process when you ask it to review a pull request or code change. It works through phases: first context gathering, such as reading the PR description, checking PR size and CI status; then a high-level review of architecture, file organization and testing strategy; then a line-by-line pass covering logic, security, performance and maintainability; and finally a summary and decision.

The skill also shapes how feedback is written. It asks for comments that are specific and actionable, that explain the reason behind them, and that label severity so blocking issues are separated from optional suggestions. It also covers handling disagreements with an author and offers a PR review comment template.

• Phased review: context, high-level design, line-by-line, summary and decision
• Severity labels such as blocking versus optional
• Feedback framed as questions or reasons rather than verdicts
• Reference checklists for security, performance and testing, loaded when needed

How do you install the skill?

Download the ZIP file from our page and unzip it. You can install it in one of two ways. For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. Alternatively, in the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

The skill's folder contains SKILL.md, which holds the main instructions, and two reference files: references/review-patterns.md and references/review-techniques.md. Keep these together so the agent can load the references when it needs them.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into your project's .claude/skills/
• Claude apps: Settings → Capabilities → Skills, then upload

How do you trigger the skill with a natural request?

You do not need a special command. Ask for a review in plain language that matches the skill's purpose: reviewing a pull request, establishing review standards, or mentoring a developer. The skill's description covers these cases, so requests such as the ones below should lead the agent to use it.

Be specific about what you want reviewed. Pasting a diff, naming a file, or pointing to a branch gives the agent something concrete to work from.

• Review this pull request for logic errors, edge cases and security issues.
• Write a review checklist for our team's backend pull requests.
• Help me give constructive feedback to a junior developer whose PR has a race condition.
• Draft review comments for this diff and label which ones are blocking.

What does a review look like?

A good request produces a review organised by the phases above, with each comment tied to a file and a reason. Comments that matter are marked as blocking, and minor points are marked as optional. The output can follow the PR review comment template in the skill, with sections for summary, strengths, required changes, suggestions, questions and verdict.

Expect the agent to ask questions rather than issue flat statements where the skill recommends it, for example asking what happens when a list is empty instead of declaring the code broken.

What does the skill need, and what should you know about safety?

The skill lists no requirements. It does not run scripts and does not use the network, according to its listing. It is made of instructions and reference documents only.

The skill itself says formatting and linting should be handled by tools such as Prettier or Black, so it will not check style well on its own. Its security checklist is a general list of prompts, not an audit. For sensitive code, a human security review is still needed.

• Requirements: none listed
• Runs scripts: no
• Uses network: no
• Use linters and formatters for style; use a human reviewer for security-critical sign-off

What tips get better reviews?

Keep pull requests small. The skill suggests 200 to 400 lines as a practical maximum and recommends asking to split larger PRs. Review in blocks of about 60 minutes at most. Run your linters and tests first so the agent's attention goes to logic and design.

Give the agent context: the PR description, the linked issue and the business requirement. The skill's context phase depends on this information.

• Split PRs above roughly 400 lines
• Run automated checks before asking for a review
• Include the PR description and linked issue
• Ask for a second pass on specific concerns, such as security or tests