The Claude Code team describes its internal review policy. Public Code Review finds issues but does not itself approve or block a pull request.
The Claude Code team has described how it divides review between agents and people. Its internal merge policy does not automatically come with the public Code Review service. That distinction matters when adding AI review to a GitHub repository.
In their October 15 post, Addy Osmani, Morgan Westlee Lunt and Qing Wang describe automated checks and human approval based on a change’s risk. An agent’s approval cannot fill a required human approval slot. This is Anthropic’s account of its own workflow, not a method NODVIEW has tested across teams.
The public Code Review documentation sets a different boundary: the service comments on pull requests but does not approve or block them. It reads the changes and surrounding code without building or running the project. Managed Code Review is available to Team and Enterprise organizations; the local /code-review command is a separate option available on any plan.
Our reading of the two sources is that an AI comment, a test result and permission to merge are separate decisions. A team wanting mandatory tests or human approval needs to establish those requirements in its own workflow. An empty findings list is not evidence that the code is correct or secure: the documentation explicitly says reviews can miss problems.