Claude Code's ultrareview: automated PR review, run live on this Mac

A cloud command that runs several reviewer AIs on your diff — tested live on a real repo.

What ultrareview is

`ultrareview` reviews the diff between your current branch and a base (or a PR number) using a fleet of reviewer agents in a remote sandbox — unlike the local, single-pass `/review`. Per the official docs (code.claude.com/docs/en/ultrareview, fetched 2026-07-29), it uses a Find→Verify two-stage pipeline to cut false positives.

Two ways to run it

`claude ultrareview [options] [target]` from the shell, with `--json` and `--timeout <minutes>` (default 30), verified via `claude ultrareview --help` on this Mac. Or `/ultrareview` inside a session.

$ claude ultrareview --help
Usage: claude ultrareview [options] [target]

Run a cloud-hosted multi-agent code review of the current
branch (or a PR number / base branch) and print the findings

A real run on this Mac

On a scratch repo with a `feature/subtract` branch ahead of `main`, `claude ultrareview main --timeout 3` returned:

$ claude ultrareview main --timeout 3
Free ultrareview 1 of 3.
Ultrareview launched for feature/subtract → main (~5-10 min, runs in the cloud).
Scope: 1 file changed, 1 insertion(+)
Waiting for findings (~5-10 min)…
  finding — 0 found, 0 verified, 0 refuted
  verifying — 5 found, 1 verified, 4 refuted
  synthesizing — 5 found, 1 verified, 4 refuted
Ultrareview failed: cloud session exceeded 3 minutes

"No changes to review" errors

Running it with no target against a branch with no diff produced: "Ultrareview could not launch: It doesn't look like you have any new commits or changes to review against your main branch." You need an actual diff against a base branch.

Pricing: 3 free runs on Pro/Max

The real output showed "Free ultrareview 1 of 3." — confirming 3 free runs before usage-credit billing kicks in.

vs. permissions/hooks

Permissions/PreToolUse hooks block risky actions before they happen; ultrareview reviews the finished diff afterward, in the cloud. Use both as complementary layers.

FAQ

What does ultrareview review?

The diff between your current branch (or a PR/base branch) and its base, via multiple cloud reviewer agents.

How many free runs?

Verified live: "Free ultrareview 1 of 3." — 3 free runs on Pro/Max.

What if it says there's nothing to review?

There's no diff against the base branch — reproduced live. Specify a target or run it on a branch with real changes.

Where do I see results?

The printed claude.ai/code/session_... URL shows live progress in a browser.

How long does it take?

About 5-10 minutes per official info and our live test; --timeout caps the wait.

Find review-focused subagent configs →Learn systematically at School →See the All-Access plan →