Claude Code opens automated reviews while teams retain merge decisions
Anthropic has made Code Review generally available to Claude Team and Enterprise organizations. The service examines proposed code changes with several agents, checks potential findings and posts comments for developers. Administrators can choose review depth and timing for each repository. The tool reads code without running it, and it leaves approval and merging to the software team.
Artificial Intelligence··Morning
Automatic review reaches Team and Enterprise
Anthropic, the company behind Claude, made its Code Review service generally available to Team and Enterprise organizations on October 15. The service examines pull requests, the proposed code changes that developers submit for inclusion in a shared project. It had previously operated as a research preview. The release follows requests from preview users for more focused findings and reviews that fit the size of a change.[1], [2]
An administrator enables the service in Claude Code settings, installs its GitHub application and chooses the repositories it may examine. Developers then receive automatic reviews on new pull requests without configuring the service individually. Repository administrators can select Quick, Standard or Deep review levels, applying different depths of examination to routine work and more critical codebases.[1], [2]
Agents inspect code and check proposed findings
Several AI agents examine the changed lines and surrounding code in parallel. A separate verification step checks potential findings against the code, then duplicate findings are removed and the remainder ranked by severity. Results appear as inline comments, a summary and a check named Claude Code Review. The default focus is correctness problems that would break production, rather than formatting preferences or gaps in test coverage. Anthropic says findings now describe the problem, the conditions that bring it about and a proposed fix in a few sentences.[2]
Teams can supply REVIEW.md and CLAUDE.md guidance to tune what the service flags. Review instructions in REVIEW.md come from the base or default branch, rather than from the changes being proposed. The managed environment has read-only access to the repository and inspects code without building or executing it. Organizations cannot run this managed service on their own infrastructure.[2]
A review does not authorize a merge
Reviews may run when a request opens, after every push or only when requested. Anyone with write access can ask for a review through an @claude review comment; drafts and forks are excluded from automatic review. The check never returns a failure status, and the service neither approves nor blocks a pull request. Teams retain the merge decision. Anthropic warns that the AI can miss bugs and security vulnerabilities even when it returns no findings.[2]
