Claude Code Ultrareview: When the Deep Cloud Review Is Worth Paying For

Claude Code ultrareview is a deep code review that runs in the cloud instead of on your computer. You type /code-review ultra, and Claude Code launches a group of reviewer agents in a sandbox on Anthropic's infrastructure. They work through your changes in parallel, and every bug they report has been reproduced and checked first, so what comes back is a short list of real problems rather than style notes. It takes roughly 5 to 10 minutes, it runs in the background while you keep working, and after three free runs it costs usage credits.

The plain way to think about it: the regular /code-review is a quick second pair of eyes you use all day. Ultrareview is the careful senior reviewer you book before a release that matters. Both look for bugs. They differ in depth, where they run, and who pays.

If you have not used the regular review yet, start there. Claude Code code review explains the local command and the managed pull request reviewer, and in the desktop app the local review comes back as a card you can step through and fix line by line. Ultrareview sits on top of that, for the moments when a quick pass is not enough.

A few useful variations from the docs. To compare against a branch other than the default, pass it: /code-review ultra develop. To review a pull request on GitHub instead of your local branch, pass its number: /code-review ultra 1234. And you can add a note in plain words, such as /code-review ultra check my auth changes, which still reviews your current branch but tells the reviewers what you care about.

Ultrareview does not come out of your plan's normal usage. Pro and Max accounts get three free runs, and those three are a one-time allowance per account that does not refresh. After that, each review bills as usage credits. The official docs say a typical review lands between $5 and $25 depending on the size of the change, and the launch dialog shows the estimate for your specific diff before you confirm.

My rule on this: spend the three free runs on real releases, not on testing what the feature does. And before you turn on usage credits, know how they are capped on your account. A $5 to $25 review is a good deal before a client launch. The same review fired off out of habit on every small change adds up fast.

On a GitHub pull request you can have Claude post the finished findings as one plain comment from your own GitHub account. It is not an approval, and it ends with a note that Claude Code generated it. Nothing is posted unless you choose it for that run: pick Run and post the findings to the PR as me in the launch dialog, or add --post to the command to preselect it. Keep the session open until the review finishes, because if the session ends first, nothing gets posted.

For client work this is the part I like most. The findings sit on the pull request where the client or their developer can see them, which makes handing off the project cleaner and shows the work was checked, without you writing a review summary by hand.

Here is how I would use it. Keep /code-review as the everyday habit, the thing you run before every commit. Book ultrareview for the changes where a missed bug costs real money or trust:

There is also a claude ultrareview command for scripts and CI that waits for the review and prints the findings. It is tempting to run it on every pull request. I would not. Each run past the free three bills usage credits, so an automatic review on every push is an open-ended cost. If you want automatic reviews on every pull request, the managed Code Review service is built for that, and Claude Code security review covers the security-focused check. Treat ultrareview as the expensive specialist you call in on purpose.