Claude Code Review Find bugs at the pace of agentic code
Anthropic launched Code Review on GitHub for Team and Enterprise tiers, inspecting pull requests in 6 to 10 minutes with severity-ranked inline comments.
By Dillip Chowdary • Nov 17, 2026 • Source: claude.com

At a glance
- Product
- Claude Code Review
- Developer
- Anthropic
- Supported host
- GitHub and GitHub Enterprise
- Review duration
- 6 to 10 minutes
- Plan tier
- Claude Team and Enterprise
- Billing model
- Token usage credits
In this post
Anthropic launched Code Review, an automated system that inspects GitHub pull requests, verifies potential defects against surrounding repository code, and publishes findings directly into discussion threads, according to claude.com's report. The service evaluates incoming pull requests against production-breaking criteria and ranks detected issues by severity.
This report examines how the system flags bugs across pull requests, how engineers run local branch reviews from the terminal, how teams configure automated scanning levels, and which repository environments Anthropic supports. The feature is targeted at engineering organizations using GitHub on Claude Team and Enterprise plans seeking automated code inspection.
01Automated inspection beyond pull request diffs
Anthropic designed Code Review to inspect pull requests alongside the surrounding repository context rather than reading isolated diff snippets. Most reviews finish within 6 to 10 minutes, generating an overall summary and posting inline comments before an engineer reviews the code. Findings are evaluated against repository files prior to publication to limit noise, and comments resolve automatically on subsequent pushes once underlying problems are fixed.
By default, the automated review targets bugs that would disrupt production environments. The engine organizes identified problems into three distinct severity tags: Important, Nit, or Pre-existing. Engineering organizations can customize flagged issues and inspection priorities by defining rules inside a REVIEW.md file located in the target repository.
Note: Code Review access to repositories during an automated run is strictly read-only and does not build, execute, or push code back to GitHub.
02Internal dogfooding metrics across Anthropic repositories
Anthropic deployed Code Review internally to manage volume generated by developers writing code with Claude across its engineering organization. The internal deployment demonstrated that automated first-pass reviews altered merge activity and review quality metrics over a four-month measurement window in 2026.
| Metric | Historical baseline | Published 2026 result |
|---|---|---|
| Daily merged changes | ~2,500 in May | 30,000+ as of September |
| Pull requests with substantive comments | 16% before Code Review | 54% in research preview |
| Claude Code revert rate | Baseline rate in June | 40% drop through September |
Between June and September, merged pull requests doubled across internal repositories while the proportion of changes requiring reverts decreased.
03Terminal workflows and local Ultrareview execution
Developers who are not organization administrators or who operate on individual plans can initiate reviews locally using Claude Code. The CLI includes a dedicated review tool that inspects local git branches or remote pull requests without requiring repository-wide GitHub App deployment.
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
npm install -g @anthropic-ai/claude-codeThe review capability is accessed directly inside the terminal interface by initiating interactive commands against branches or pull request identifiers.
/code-review
/code-review ultraclaude --model sonnet
claude --dangerously-skip-permissionsUltrareview runs as a research preview feature that delegates deeper code inspection tasks to reviewer agents operating inside cloud sandboxes. Teams running terminal-initiated ultra reviews must enable usage credits on their Anthropic billing configuration.
04Repository triggers and tiered inspection depth
Repositories connected to the system default to running reviews on every push, but teams can adjust execution triggers to run only upon pull request creation or manual demand. Team members with write access can trigger an on-demand review on standard pull requests, draft pull requests, or forks by submitting a specific comment.
- GitHub App installation An organization owner installs the Claude GitHub application across target repositories.
- Claude admin enablement An organization owner turns on the Code Review setting inside Claude admin settings.
- Trigger configuration Teams select automatic scanning or comment-based execution using the @claude review mention.
Review depth is managed through configurable tiers named Quick, Standard, and Deep. Quick runs review routine modifications, Standard serves as the default baseline, and Deep provides extensive analysis for sensitive operational services.
05Platform compatibility boundaries and billing structure
Code Review supports cloud-hosted GitHub environments, including GitHub Enterprise Server and GitHub Enterprise Cloud configurations with data residency. The service does not support GitLab, custom per-user access control rules, or fully self-hosted infrastructures. Anthropic excludes organizations configured with zero data retention policies, HIPAA settings, or customer-managed encryption keys.
Teams accessing Claude Code through third-party platforms such as Amazon Bedrock, Google Cloud, or Microsoft Foundry can use Code Review if they hold an active Claude Team or Enterprise organization. Inspection workloads execute directly on Anthropic infrastructure and bill against Anthropic account credits based on consumed tokens rather than inclusive plan allowances.
Administrators monitor review usage by setting monthly organization spending caps and reviewing average per-run costs calculated across individual repositories.
Frequently asked questions
Which repository hosting platforms work with Code Review?
Code Review works with GitHub, GitHub Enterprise Cloud with data residency, and GitHub Enterprise Server. It does not support GitLab, per-user access controls, or self-hosted systems.
Can developers without a Team or Enterprise plan use Code Review?
Yes. Developers can run /code-review or /code-review ultra on individual branches directly inside Claude Code on any plan.
How is Code Review billed?
Code Review is billed separately from subscription allowances using usage credits based on token consumption, with configurable monthly spend limits.
Sources
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement