Anthropic's /ultrareview in Claude Code Sends an Agent Fleet to Hunt Bugs in the Cloud

You are about to merge an important PR and worried a hidden bug might slip through? Anthropic just shipped /ultrareview, a command that calls in a whole fleet of bug hunting agents running in parallel in the cloud to review your code for you.

What is /ultrareview?

This is a new command in Claude Code, currently in research preview. Instead of a single agent reviewing your code sequentially like before, /ultrareview splits the work and spins up multiple agents running in parallel in the cloud.

Each agent focuses on a different angle: logic bugs, security vulnerabilities, rare edge cases, and regression risk. The findings are then merged into one report and sorted by severity, so you know what to fix first.

Why running it in the cloud actually matters

The best part is that the whole process uses none of your local machine's resources. Your laptop stays fast and you keep coding normally, while an entire fleet of agents grinds through your code on Anthropic's servers.

The tool automatically bundles your code, uploads it to the cloud, runs the agent fleet, then sends the results straight back to your CLI or Claude Desktop. All you have to do is read the report and decide whether to merge.

When should you reach for /ultrareview?

Think of it as an automatic "second opinion", not something you run on every small PR. Save it for changes where a single bug could cost you data, money, or trust.

Good use cases include merging a PR that touches your auth flow, a database schema migration, payment logic, or any other critical path in your system, basically anywhere you want to sleep soundly after hitting merge.

How to use it

Inside the Claude Code CLI, there are two simple ways to call it. Type /ultrareview to review the branch you are currently on, or type /ultrareview <PR#> to review a specific PR on GitHub.

Wait a few minutes and you will get a detailed report listing bugs, vulnerabilities, and potential risks, complete with suggested fixes.

Pricing and free quota

Good news for Claude Pro and Claude Max users: you get 3 free /ultrareview runs, available until May 5, 2026. It is a great window to try it out before deciding whether to stick with it.

After that date, /ultrareview will be billed through the usage plan, meaning you pay for what you use. Given that it is catching critical bugs before they ever reach production, that cost is well worth it compared to firefighting an outage at 2am.

Have you tried /ultrareview yet, and which kind of PR would you run it on first?