Code Review
Core Facts
Timeline (last 90 days)
Code Review fixed cases where delayed GitHub push events caused reviews of stale commits or no review at all; it now reviews the PR's current commit.
Implement Spec ultimately merges all tasks into one branch and automatically invokes the Code Review skill for review, generating a clean pull request after fixing issues.
Claude Code v2.1.307's Code Review adds the `@claude review stop` command (as well as off and pause), used to end a `@claude review always` subscription on a PR without triggering a new review.
The Code Review product only issues review comments and does not issue approvals; approving a PR is still decided by humans, whereas the Claude Code team's own repository allows low-risk changes to be merged on automated checks alone.
Matt Pocock's code review skill reviews against Martin Fowler's 12 refactoring principles in a fresh context window.
Claude Code v2.1.305 fixes an issue where Code Review sometimes did not review the latest commit when multiple pushes were received during PR review.
In v2.1.304, the Code Review feature allows organization owners to choose among three review levels—Quick, Standard (default), or Deep—and to set them individually for a single repository; the admin settings add an option that can cover all repositories in a GitHub organization (including those added later).
Code Review runs on Anthropic's infrastructure with read-only permissions and does not build or execute code.
The product documentation states explicitly: Code Review never approves or blocks a pull request, and check runs never end in failure.
According to The Next Web, Anthropic's claims about the product's performance have not been independently tested.
15 more timeline events
All Facts (20)
Google's engineering practice research indicates that code review can detect approximately 60% of defects, far exceeding the approximately 30% detection rate of unit testing.
80%UnverifiedCode Review fixed cases where delayed GitHub push events caused reviews of stale commits or no review at all; it now reviews the PR's current commit.
50%UnverifiedImplement Spec ultimately merges all tasks into one branch and automatically invokes the Code Review skill for review, generating a clean pull request after fixing issues.
50%UnverifiedThe Code Review product only issues review comments and does not issue approvals; approving a PR is still decided by humans, whereas the Claude Code team's own repository allows low-risk changes to be merged on automated checks alone.
50%UnverifiedMatt Pocock's code review skill reviews against Martin Fowler's 12 refactoring principles in a fresh context window.
50%UnverifiedIn Anthropic's own Claude Code repository, a PR cannot be merged as long as a review finding remains open; in the Code Review version sold externally, reviews never fail checks and never approve or block a PR.
50%UnverifiedCode Review lets administrators choose one of three review depths per repository: Quick, Standard (default), or Deep.
50%UnverifiedCompared with the preview stage, most Code Review reviews are now about 3 times faster, dropping from about 20 minutes to 6 to 10 minutes.
50%UnverifiedCode Review's mechanism is that multiple agents read the diff and its surrounding code in parallel, then a validation round discards false positives before results are published.
50%UnverifiedEach finding on a PR is labeled Important, Nit, or Pre-existing.
50%UnverifiedCode Review runs on Anthropic's infrastructure with read-only permissions and does not build or execute code.
50%UnverifiedThe product documentation states explicitly: Code Review never approves or blocks a pull request, and check runs never end in failure.
50%UnverifiedAccording to The Next Web, Anthropic's claims about the product's performance have not been independently tested.
50%UnverifiedOption one: GitHub branch protection rules can require all comments on a PR to be resolved before merging, thereby using unresolved inline threads to block the merge.
50%UnverifiedOption two: at the end of the details text of a check run there is a machine-readable verdict line, in which the normal key counts the number of Important findings, including those still unresolved from previous reviews; gh can parse this line.
50%UnverifiedCode Review is billed by token usage through usage credits, calculated separately from the usage included in the plan, and the cost is charged to the organization account.
50%UnverifiedDuring the preview, according to The Next Web, each Code Review review costs between 15 and 25 US dollars.
50%UnverifiedFor repositories newly added to Code Review, the default trigger is to review after every push, and the cost multiplies with the number of pushes.
50%UnverifiedEligible Enterprise organizations can apply for at least 10,000 US dollars in Code Review credits before January 15, 2027.
50%UnverifiedAfter improvements to the code review module, merging the base branch into a PR that contains Additional findings now triggers only a lighter follow-up review rather than a full re-review.
50%