How to use the Claude Code review skill (/code-review)

How to run /code-review in Claude Code

1. Make your changes. Work in a branch as usual. With no target, the skill reviews your commits ahead of the upstream branch plus your uncommitted changes.
2. Run the review. Type /code-review (or /review) in Claude Code. It runs as a background subagent with its own context window and follows your CLAUDE.md, so you can keep working while it reads.
3. Tune it. Pick an effort level from low to max, point it at a pull request, branch, path or commit range, and use --max-findings to change how many findings it reports.
4. Fix or share. --fix applies fixes for what it finds. --comment posts the findings as inline comments on a GitHub pull request, or as a GitLab merge request note through glab.
5. Go deeper when it matters. /code-review ultra runs an ultrareview in the cloud; claude ultrareview runs it from scripts or CI.

The skill is fastest while you’re still working: run it before you push, fix what it finds, then open the pull request. It reviews what Claude Code can see in your checkout, so it’s a check on your own changes rather than a team-wide review step. The review runs in your own session, so it counts toward your plan’s usage or API billing like the rest of the session.

Choose a target and an effort level

Anything you type after the effort level and flags becomes the review target. A few useful forms:

• /code-review 1234 reviews pull request 1234.
• /code-review main...my-feature reviews one branch against another.
• /code-review HEAD~15..HEAD reviews your last 15 commits.
• /code-review src/auth reviews a file or directory.

Effort trades coverage for confidence. At low, the review reports only the findings it’s most confident in, so you see fewer false positives. From medium through max, it widens coverage and reports more, including some it’s less sure of. Use low for a quick pass on a small change, and a higher level before you merge something risky, such as auth or a migration. The command remembers the last level you typed, even across sessions, and says so when it reuses it.

Two things to know before you use --fix. Its edits go through your permission mode, and a background review makes them outside your session’s checkpoints, so /rewind won’t undo them; use git instead. Ultrareview needs a claude.ai login and isn’t available through Amazon Bedrock, Google Cloud’s Agent Platform or Microsoft Foundry, or to organizations with zero data retention or the HIPAA configuration. In those cases /code-review ultra runs a local review instead.

How to set up Claude Code Review on GitHub

Code Review is Anthropic’s hosted reviewer for GitHub. It left research preview on October 15, 2026, and is available to Claude Team and Enterprise organizations, with reviews taking about 6 to 10 minutes after GA, down from about 20.

Setup needs an Owner of your Claude organization, an owner of the GitHub organization to install the Claude GitHub App, and write access to each repository you add. No workflow files go into your repositories.

1. Turn it on. In Admin settings > Claude Code on claude.ai, click Allow code review and connect your GitHub account.
2. Install the Claude GitHub App on the GitHub organization and pick which repositories it can access.
3. Choose repositories and a review level: Quick, Standard (the default) or Deep, set for the organization and overridable per repository.
4. Decide when reviews run in each repository: once after the pull request opens, after every push (the default for new repositories) or only on request.
5. Test it with a non-draft pull request. A check run named Claude Code Review appears within a few minutes.

Anyone with write access can comment @claude review for one review, or @claude review always to review that pull request on every later push. Automatic reviews skip drafts and pull requests from forks, but a comment works on both. It supports GitHub, including GitHub Enterprise Server, and isn’t available with zero data retention, HIPAA or customer-managed keys.

How to read a review

Each finding is an inline comment on the lines concerned, marked with one of three severities:

• 🔴 Important: a bug that should be fixed before merging.
• 🟡 Nit: a minor issue, worth fixing but not blocking.
• 🟣 Pre-existing: a bug already in the codebase that this pull request didn’t introduce.

Expand Why this was flagged on a finding to see how Claude verified it. To dismiss one, resolve its thread; a reply alone doesn’t. To fix one, push the change, and the next review resolves the thread when it judges the issue fixed. Later automatic reviews focus on what changed and post fewer nits, so comment @claude review when you want the whole pull request covered again.

The Claude Code Review check run concludes as neutral, never as a failure, so it can’t block a merge on its own. To gate merges, Anthropic documents reading the last line of the check run’s Details text, which holds severity counts as JSON; a normal count above zero means at least one Important finding.

Write rules both reviewers follow

The two tools read different files. /code-review follows CLAUDE.md only. The hosted service reads CLAUDE.md as project context and REVIEW.md as review-only instructions. Put rules you want everywhere in CLAUDE.md, and keep REVIEW.md for what to flag, at what severity and what to skip:

# Review instructions

## What Important means here
Important is for broken behavior, leaked data or a migration
that isn't backward compatible. Naming and style are Nit at most.

## Cap the nits
Report at most five Nits per review.

## Do not report
- Lint, formatting and type errors. CI enforces them.
- Generated files under src/gen/

## Always check
- Database queries are scoped to the caller's tenant.

Code Review reads REVIEW.md from the base or default branch, never from the pull request, so merge a change to it before you expect results. Protect that branch: lowering severity, excluding paths and capping nits only apply when the file comes from a protected default branch. You can add a REVIEW.md in a subdirectory for checks on that path, and pull other root files in with an @file.md line.

What Claude Code Review costs

There’s no flat price per review. Code Review is billed on token usage through usage credits, separately from your plan’s included usage, and charged to the organization rather than the developer. Each review’s cost grows with the review level, the size of the change, the complexity of the surrounding code and how many candidate issues need checking. During the research preview, Anthropic said reviews averaged $15–25. At GA it offered eligible Enterprise organizations at least $10,000 in Code Review credits, to be claimed by January 15, 2027.

How often reviews run matters as much as the level: after every push multiplies cost by the number of pushes. Anthropic suggests a pilot on a few repositories set to once per pull request or manual, then reading the Average cost column in admin settings and multiplying by your monthly pull requests. Set a monthly spend limit for the service; when it, the organization limit or your credit balance runs out, reviews are skipped and the pull request gets a comment saying why.

/code-review vs Claude Code Review

When to add an independent reviewer

When Claude Code writes most of the code, a reviewer built on a different model can catch what the author model missed. Model diversity exposes blind spots; it doesn’t guarantee them away. An independent reviewer is worth adding when:

• Several agents and people write code in the same repositories.
• You want every pull request reviewed the same way, whoever opened it.
• Your team isn’t on Claude Team or Enterprise, or needs a predictable monthly cost instead of token billing.
• You want fixes run until the pull request is ready to merge, not only comments.

HeyDeer reviews every pull request, double-checks each finding against the code with a different model, reads your REVIEW.md and AGENTS.md, and can fix what it finds. See HeyDeer vs Claude Code Review for pricing, or the best AI code review tools. New to the topic? Start with how AI code review works.

A Claude Code REVIEW.md works in HeyDeer as it is, so you can run both reviewers from one file. Important maps to HeyDeer’s Medium severity or above and Nit to Low. HeyDeer reads the file from the pull request rather than the base branch, and doesn’t follow @file includes, so move that guidance into REVIEW.md. The REVIEW.md docs cover the rest.

Sources

1. Claude Code Docs: Review code locally (/code-review)
2. Claude Code Docs: Code Review
3. Claude Code Docs: Code Review costs
4. Claude Code Docs: Set up Code Review
5. Claude Code Docs: Customize Code Review (REVIEW.md and CLAUDE.md)
6. Claude Code Docs: Read and request reviews on a pull request
7. Anthropic: Code Review is now generally available (October 15, 2026)
8. Anthropic: Code Review launch (March 9, 2026)
9. GitHub: anthropics/claude-code-security-review