How the Claude Code team does code review

The Claude Code team has optimized their code review process to handle the high volume of changes produced by agents. Merged pull requests to the Claude Code codebase more than doubled from June to September, while the share that had to be reverted fell by more than 40 percent. Rather than reducing human review time, they moved most of the reading to machines, optimizing for human time and attention, and invested in making it fast and easy to undo mistakes.

When engineers write code, the number of changes increases rapidly, with changes landing per day going from about 2,500 in May to more than 30,000 in September. Claude Code lives in its own repository, separate from the monorepo, but still faces significant pressure. To address this, review runs continuously, similar to spellcheck while writing. Most changes have been reviewed and fixed by Claude at least once before a human needs to look.

The review process now covers multiple layers before a PR exists. Claude runs tests, builds, or creates screenshots to compare against designs, helping catch issues early. Engineers also run internal skills to hunt for bugs, simplify the diff, and check end-to-end behavior. UI changes are checked against design guidelines. After these steps, a Claude review checks the diff against the entire repo and the PR's stated intent, determining if the change solves the problem, fits the system, or breaks something else.

For bug fixes, the Claude review looks for the root cause, while for pull requests over 1,000 lines, findings are generated to highlight potential issues. These findings are verified by other agents before humans review them, with style, diagnostic, or cleanup findings being dropped. The review process includes security checks, with a PR not merging until any security findings are resolved.

Once a finding is confirmed as real, it is posted as inline comments sorted by severity. Each finding must be answered, either by fixing the issue or explaining why it was not fixed. Security reviews ensure pull requests can't merge until all findings are addressed, and if a later commit fixes an issue, the finding clears on its own.