Before Showing It to People, Get Your Code Checked by AI First with /code-review

Coding alone, you get that anxiety of "isn't there a bug hiding somewhere in this?" Especially when there is no one to review it. There was no good way to check before showing it to people, but now there is. It is /code-review. AI reviews the bugs and improvements in your written code, and if you want, fixes them right there.

Definition

/code-review is a command that reviews the bugs and improvements in written code. Just as a human reviewer looks over a PR (a code change proposal), Claude scans the code and points out problems and parts to fix.

A distinctive feature is that you can choose the review intensity. Light review only for everyday work, deep detailed review for important changes, and so on. Adding an option can also have it fix the problems it finds right away.

How to Use (by difficulty)

Basics — Getting a Review Only

Type just this into the chat box.

/code-review

Once you enter it, Claude looks over the written code and reports the bugs and improvements it found item by item. It explains what is wrong and why, so you can decide whether to fix it.

Advanced — Review and Fix Right Away

/code-review --fix

Adding --fix (fix — an option meaning "to fix") does not stop at review but applies the found problems to the code and fixes them right away. It is handy for cleaning up small bugs and duplicate code in one go.

Deeper — Detailed Review in the Cloud

/code-review ultra

Adding ultra (ultra — detailed mode) has multiple agents (assistant AI workers) review the code deeply in the cloud (a server across the internet). It takes more time, but it is correspondingly thorough, so use it right before deployment or for important changes.

Common Pitfall — Security Checks Are Separate

/code-review is a quality-focused review. It looks at things like bugs, duplicate code, and improvements. But beginners easily misunderstand "since I ran this, security must be covered too," which is not the case.

To intensively check security vulnerabilities (hacking, information-leak risks, etc.), you have to run /security-review (security review — a security-only review command) separately. For important code, it is safer to check both quality with /code-review and security with /security-review.

Real-World Example

After adding a new feature to a blog backend, I wanted to check it once before showing it to people or deploying. So I ran this.

/code-review --fix

Claude found a few small bugs and some duplicate code and cleaned them up on the spot. It caught parts I would have missed on my own. Getting "filtered once" code even without a human reviewer put my mind much more at ease.

Using It Further

• Auto-fix: Adding --fix fixes the found problems right away.
• Cloud detail: For important changes, multiple agents review in depth with /code-review ultra.
• Leave as PR comments: Adding --comment (the comment option) leaves line-by-line comments on a GitHub PR.
• Double-check security too: Bundle with /security-review to see quality and security together.

Tip: Light and quick with /code-review for everyday work, detailed with ultra right before deployment or for important code — dividing the intensity this way is efficient. Try making it a habit to run code past the AI before showing it to people.

How Others Use It

• Lessons from real projects — a collection of cases that raised quality by putting review into the workflow. — Best Practices: Lessons From Real Projects (RanTheBuilder)
• Tips I wish I'd had from day one — real experience of making review and verification a habit. — Tips I Wish I'd Had From Day One (Marmelab)

Summary

The core fits in one sentence. Get your code checked by AI first before showing it to people. Point out bugs and improvements with /code-review, fix right away with --fix, and look deeply with ultra if important. Just note: security must be checked separately with /security-review.

Based on: Claude Code v2.1.154 (2026.05)