Back to Blog

Claude Code PR review: four ways to review every pull request, and which to pick

Alex Kim
12 min read
Claude Code PR review: four ways to review every pull request, and which to pick

Last updated: November 20, 2026

TL;DR

To have Claude review every pull request, pick by plan and code host. On a Team or Enterprise plan with GitHub repositories, turn on Code Review: Anthropic's managed service reviews each pull request on its own infrastructure, with no workflow file to write. On any other plan, or on GitLab, run Claude in your own CI with GitHub Actions or GitLab CI/CD. Whichever you pick, run /code-review in your session before you open the pull request, on any plan (Anthropic's Code Review docs).

  • ๐Ÿข Managed Code Review: every pull request, automatically, on Anthropic's side. Team and Enterprise, GitHub only.
  • โš™๏ธ Your own CI: your pipeline's events start it, on your runners, with an API key or cloud credentials. GitHub or GitLab.
  • ๐Ÿ’ป /code-review in your session: you run it on your branch before anyone else sees it. Any plan.
  • โ˜๏ธ Ultrareview: a deeper pass in a cloud sandbox when one branch matters more than the rest.

The rules each review follows live in two files, CLAUDE.md and REVIEW.md, and that part is the same whichever you choose.

The four options side by side

Anthropic's docs compare the ways to get a review by what starts one, where it runs and who can use it (Code Review):

OptionWhat starts a reviewWhere it runsWho can use it
Code ReviewPull request activity, as each repository is set, or an @claude review commentAnthropic infrastructureTeam or Enterprise organizations with GitHub repositories, once an Owner turns it on
Claude in your own CIYour pipeline's events, such as every pull request or merge requestYour own CI runnersAnyone with a Claude API key or cloud provider credentials, on GitHub or GitLab
/code-reviewYou run the command on your branchYour Claude Code sessionAnyone, on any plan
UltrareviewYou start it from Claude CodeA cloud sandboxAnyone signed in with a claude.ai account, with exceptions

The docs list a fifth: your own reviewer, written as a subagent or a skill, which you or Claude invoke during a session. It is how you build a review the others don't cover, not a way to review every pull request.

Option 1: Managed Code Review, when you're on Team or Enterprise

What it does. Several agents read the change alongside the surrounding code, a verification step checks their findings against the code to filter out false positives, and what remains is deduplicated and ranked. The review posts inline comments on the lines concerned, a summary, and a check run named Claude Code Review. It focuses on correctness by default: bugs that would break production, not formatting or missing tests (Code Review).

What it never does. A review never approves or blocks a pull request, and its check run never concludes as a failure. Inside a review, access to the repository is read-only, it pushes no commits, and it works from reading the code without building or running it.

Severity. Each finding carries one of three markers (Read and request reviews):

MarkerSeverityMeaning
๐Ÿ”ดImportantA bug that should be fixed before merging
๐ŸŸกNitA minor issue, worth fixing but not blocking
๐ŸŸฃPre-existingA bug in the codebase that this pull request didn't introduce

Setup takes an Owner. An Owner of the Claude organization turns it on in Admin settings, Claude Code, in the Code review section: switch on Allow code review, connect GitHub, install the Claude GitHub App on the repositories, choose them, and click Finish. Installing the app needs an owner of the GitHub organization, and adding a repository needs write access to it. Then open a test pull request that isn't a draft: the check run appears within a few minutes (Set up Code Review).

Choose when reviews run, per repository. The Review runs setting has three options:

  • Once after PR creation: a review when a pull request is opened, reopened or marked ready for review.
  • After every push: a review on each push. A newly added repository starts here, and it runs the most reviews.
  • Manual (@claude review): nothing runs until someone comments.

Automatic reviews skip drafts and pull requests from forks. On any setting, someone with write access can comment @claude review for one review, or @claude review always to subscribe that pull request to a review on each push; @claude review stop ends the subscription.

Choose how thorough. The review level is Quick, Standard (the default) or Deep, set once for the organization and overridable per repository. A deeper level takes longer and costs more.

What it costs. There is no flat per-review price. Code Review is billed on token usage through usage credits, separately from your plan's included usage, and to the organization rather than to the developer. The cost of a review scales with the level, the size of the change, the complexity of the code around it and how many candidate issues need verifying, and the number of reviews comes from the Review runs setting (Code Review costs). Anthropic's suggested pilot: set a monthly spend limit, turn it on for a few repositories at Once after PR creation or Manual, read the Average cost column after several reviews, and multiply by your monthly pull requests (and by pushes per pull request if you'll use After every push). When a limit or the credit balance runs out, reviews are skipped and the pull request gets a comment naming the cause.

Where it isn't available. GitHub only, GitHub Enterprise Server included. Not for organizations with Zero Data Retention, the HIPAA configuration or customer-managed encryption keys. Teams that use Claude Code through Amazon Bedrock, Google Cloud or Microsoft Foundry still need a Claude organization on Team or Enterprise for it.

Gate a merge on it from CI. Since the check run never fails, a team that wants a hard gate reads the result itself. The docs show how with the gh CLI: list the commit's check runs, take the Claude Code Review run's ID, and parse the last line of its output, an HTML comment holding the severity counts as JSON. A normal count above zero means at least one Important finding; a missing line means no verdict, not zero findings.

Option 2: Claude in your own CI, when you're not

If your plan doesn't include Code Review, your repositories are on GitLab, or you want reviews on your own runners, run Claude from your pipeline: GitHub Actions on GitHub, GitLab CI/CD on GitLab. Your pipeline's events start a review, such as every pull request or merge request, and it runs on your CI runners with a Claude API key or cloud provider credentials (Code Review).

What you trade for that reach: you write and own the workflow, you decide its triggers and permissions, and you pay in CI minutes plus Claude usage rather than through usage credits. The setup and the controls on a GitHub workflow are in our guide to running Claude Code unattended, and a one-off review from a script is headless mode.

Running both on one repository? If the Claude Code GitHub Action listens for its default @claude trigger in interactive mode, an @claude review comment starts the Action as well as Code Review. Set the Action's trigger_phrase input to another phrase to keep the two apart (Read and request reviews).

Option 3: /code-review before you open the pull request

/code-review is a command you type at the Claude Code prompt. It reviews the changes on your branch for correctness bugs and reports in your conversation; /review is an alias. It works on any plan and counts toward your plan's usage or API billing like the rest of the session (Review locally with /code-review).

With no target it reviews your branch's commits ahead of its upstream plus uncommitted changes. Point it elsewhere by typing a target after it:

To reviewType
Your branch and uncommitted changes/code-review
A pull request by number/code-review 1234
One branch against another/code-review main...my-feature
A file or directory/code-review src/auth

Two flags turn the report into action: the fix flag applies the findings to your working tree, through your permission mode, and the comment flag posts them on a GitHub pull request as inline comments, or on a GitLab merge request as one note through the glab CLI. Type each with two hyphens before it, as on the command line. An effort level such as /code-review high trades confidence for coverage: low reports only what it's most sure of.

In a terminal session it runs as a background subagent with its own context window, so your conversation stays clean while it works. It follows your CLAUDE.md, not REVIEW.md.

Option 4: Ultrareview, for the branch that matters

/code-review ultra runs a deeper review in a cloud sandbox. It compares your current branch with the repository's default branch, plus uncommitted and staged changes, and on a github.com pull request it can post the findings to the pull request from your GitHub account. From a script or CI job, the claude ultrareview subcommand waits for the findings and prints them.

It needs a claude.ai sign-in. It isn't available on Amazon Bedrock, Google Cloud or Microsoft Foundry, or with Zero Data Retention or the HIPAA configuration; there, /code-review ultra runs a local review instead (Review locally with /code-review).

Tell every review what your team cares about

Two files steer what gets flagged (Customize Code Review):

  • CLAUDE.md: project rules Claude reads on every task. Both Code Review and /code-review read it, and a change that breaks one of its rules becomes a finding.
  • REVIEW.md: review-only rules at the repository root, read by Code Review alone. Use it to redefine what Important means for your codebase, cap nits, skip generated files and lockfiles, and add checks you want on every pull request.

Anthropic's example REVIEW.md does all four. This is it, shortened:

# Review instructions

## What Important means here

Reserve Important for findings that would break behavior, leak data,
or block a rollback. Style, naming, and refactoring suggestions are
Nit at most.

## Cap the nits

Report at most five Nits per review.

## Do not report

- Anything CI already enforces: lint, formatting, type errors
- Generated files under `src/gen/` and any `*.lock` file

## Always check

- New API routes have an integration test
- 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's own changes, so merge a change to it before you expect it to apply. Protect your default branch: lowering a severity, excluding paths and capping nits are honored from a protected default branch, and included files and subdirectory REVIEW.md files apply only when the branch is protected. Put guidance you want in both the managed review and /code-review in CLAUDE.md.

Frequently asked questions

How do I get Claude Code to review every pull request?

On a Team or Enterprise plan with GitHub repositories, have an Owner turn on Code Review and set each repository's Review runs to Once after PR creation or After every push. On other plans, or on GitLab, run Claude from your own CI with GitHub Actions or GitLab CI/CD.

Is Claude Code Review free?

No. Code Review is billed to the organization on token usage through usage credits, separately from your plan's included usage. There is no flat price per review: cost scales with the review level, the size of the change and how often reviews run. The local /code-review command counts toward your normal plan usage.

What is the difference between Code Review and /code-review?

Code Review is a managed service for Team and Enterprise organizations on GitHub that reviews pull requests on Anthropic infrastructure. /code-review is a command you run in your own Claude Code session, on any plan, to review your branch, a pull request or a path. Code Review reads CLAUDE.md and REVIEW.md; /code-review reads only CLAUDE.md.

Can Claude Code Review block a merge?

No. A review never approves or blocks a pull request, and its Claude Code Review check run never concludes as a failure. To gate a merge, read the severity counts from the check run's output with the gh CLI in your own CI job and fail the job when the Important count is above zero.

Does Claude Code PR review work with GitLab?

The managed Code Review service is GitHub only. On GitLab, run Claude in your pipeline with GitLab CI/CD, or run /code-review locally and post its findings to a merge request as a note through the glab CLI.

How do I request a Claude review on a pull request?

Comment @claude review as a new top-level comment on the pull request, with write access to the repository. It starts one review. @claude review always also subscribes the pull request to a review on each later push, and @claude review stop ends that subscription.

What is REVIEW.md?

A Markdown file at the repository root that holds review-only instructions for Code Review: what counts as Important, how many nits to post, which paths and issue types to skip, and which checks to run on every pull request. Code Review reads it from the base or default branch, never from the pull request's own changes.

Give every review the same rules

Whichever review runs, it is only as good as the rules it reads, and the one file every option reads is CLAUDE.md. Adding a rule each time a review misses something, so the next one catches it, is the method in our guide to compound engineering for Claude Code. For checks that run on every tool call rather than every pull request, see our hooks guide.

#claude-code#code-review#Pull Requests#ai-agents#CI
Free assessment

Stop guessing. Score your first workflow.

Score one workflow in about five minutes. Nine questions, a grade, and the specific gap holding it back - no call required.