Code review on command – the colleague who always has time

Pull request done, review requested. And then you wait. Your colleague is busy with a release, tomorrow they're on holiday, and the day after they've long forgotten the context.

For a while now we've had a first round before the human review: Claude Code.

🔍 /code-review

Checks the changes in the current branch for real bugs: wrong logic, forgotten edge cases, race conditions. The review runs as a subagent (part 1) in the background, so I can keep working in the meantime.

• /code-review low to /code-review max: from a few high-confidence findings to many, some of them uncertain.
• --fix applies the findings straight to the working tree.
• --comment posts them as comments on the pull request.

🔐 /security-review

The specialist: compares the branch with the main branch and looks specifically for security issues like injection, broken authentication or exposed data. For us it comes before every merge of code that handles user input.

☁️ And when it needs to be thorough

/code-review ultra starts an in-depth multi-agent review in the cloud. It takes longer and is mainly worth it for large or critical changes.

A word of caution:

An AI review doesn't replace a human review. Claude reliably finds forgotten null checks, but not whether the feature was meant to work that way in the first place. Also: a review uses tokens too. Not every typo commit needs max.

My rule of thumb:

• Before every pull request: /code-review.
• User input, permissions, data? Add /security-review.
• The question "Is this the right idea?" stays with a human.