UltraReview
Run a deep multi-agent code review in the cloud with /code-review ultra before you merge.
UltraReview runs a deep code review on Claude Code on the web: many reviewer agents work in a remote sandbox and independently verify findings so you get high-signal bugs rather than only style nits. The preferred command is /code-review ultra; /ultrareview remains an alias when the feature is available.
Official guide: Find bugs with ultrareview.
Research preview and availability
UltraReview is a research preview; pricing and behavior may change. It requires a Claude.ai sign-in (use /login if you only use an API key today). It is not available through Bedrock, Google Cloud’s Agent Platform, or Foundry, or for organizations with Zero Data Retention enabled. When cloud UltraReview is unavailable, /code-review ultra runs a local review instead.
Run from the CLI
These /code-review ultra lines are typed inside Claude Code (terminal or VS Code), the same on macOS, Linux, and Windows.
Current branch (diff vs default branch, including uncommitted and staged changes). The CLI bundles the repo and uploads it to the sandbox:
/code-review ultraGitHub pull request by number (sandbox clones the PR; needs a github.com remote):
/code-review ultra 1234If the repo is too large to bundle, Claude Code may ask you to open a draft PR and use PR mode instead.
You get a confirmation dialog (scope, free runs remaining, estimated cost) before launch. The review runs in the background; use /tasks to track, open details, or stop (stopping archives the session; partial findings are not returned).
Example (set expectations before you confirm)
Pass a plain-language note after the command so reviewers weight the right risks:
/code-review ultra Prioritize SSRF-prone outbound fetches, authz on workspace routes,
and unsafe deserialization. Deprioritize formatting and naming unless they hide
a bug. Assume production deployment with untrusted clients.Compared to /code-review
/code-review | /code-review ultra | |
|---|---|---|
| Where | Local session | Cloud sandbox |
| Depth | Scales with the selected effort | Multi-agent with independent verification |
| Time | Seconds to a few minutes | Often 5 to 10 minutes |
| Cost | Normal usage | Three lifetime free runs on Pro and Max, then usage credits; Team and Enterprise have no free runs |
Use /code-review for fast iteration; use /code-review ultra when you want pre-merge confidence on a substantial change.
After the review
Finished reviews surface as notifications with file locations and explanations so you can ask Claude to fix issues. Typical turnaround is 5 to 10 minutes; you can keep working or even close the terminal while it runs.
After the free runs, reviews typically cost $5 to $25 in usage credits, depending on change size. A stopped or failed free run still consumes that run; paid reviews bill only for the work performed. Check the launch estimate and use /usage-credits for the current account setting.