Ultrareview
Multi-agent code review running on Anthropic's cloud: a fleet of reviewer agents explores your branch or pull request in a remote sandbox, and every issue is reproduced and verified before it reaches you. The official command is now `/code-review ultra`, with `/ultrareview` kept as an alias. Expect 15 to 25 minutes per review, then 5 to 25 dollars in usage credits once the 3 free runs of the Pro and Max plans are gone, and those never refill. It runs from any git repository and you track it with `/tasks`; the `claude ultrareview` subcommand runs it from a script or CI. It requires a claude.ai login: unavailable through Bedrock, Vertex or Foundry, under Zero Data Retention or the HIPAA configuration (the command then runs a local review).
Strengths
- Each finding independently reproduced in a sandbox, eliminates false positives
- Anthropic cloud: runs without local machine, persistent results via /tasks
- 3 free Pro/Max runs, zero cost to evaluate on a real PR
Limitations
- 5 to 25 dollars per review after the 3 free runs, too expensive for every daily PR
- ~15-25 min latency, not suited for synchronous reviews
- Research preview: behavior and pricing may change without notice
- No documented settings: not the number of agents, the review angles or the models, and the intermediate reasoning stays hidden
Best for
- Critical PRs before a release or main merge, when cost is worth the certainty
- Teams wanting an independent Claude review without setting up a GitHub App