{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreiaf7idipwamou2p2qbjqidkjtabhtnjbl2qrao3hocbwokymqz4lm",
    "uri": "at://did:plc:ws6dhxzqnqxu5aqxt4kd27oc/app.bsky.feed.post/3mjrmqivhah32"
  },
  "coverImage": {
    "$type": "blob",
    "ref": {
      "$link": "bafkreibw5jutbf2ml3xgfwde67ipa77tzsy6wdedhvt7bxf6wy5dd6amze"
    },
    "mimeType": "image/jpeg",
    "size": 49989
  },
  "description": "A multi-agent remote review that verifies findings, runs in the background, and bills as extra usage after three free runs.",
  "path": "/claude-code-ultrareview-deep-cloud-bug-hunt-before-you-merge/",
  "publishedAt": "2026-04-18T14:13:50.000Z",
  "site": "https://allthings.how",
  "tags": [
    "extra usage"
  ],
  "textContent": "Ultrareview is a premium slash command in Claude Code that kicks off a deep, multi-agent code review in a remote sandbox. Instead of a single-pass scan in your terminal, it spins up a fleet of reviewer agents on Claude Code's web infrastructure to hunt for real bugs in a branch or pull request, with each finding independently reproduced before it lands back in your session.\n\n⚡\n\nQuick answer: Run /ultrareview in the Claude Code CLI to review your current branch against the default branch, or /ultrareview 1234 to review GitHub PR #1234. It requires Claude Code v2.1.86 or later, a Claude.ai login, and extra usage enabled after three free runs on Pro and Max.\n\n* * *\n\n### What ultrareview actually does\n\nWhen you invoke the command, Claude Code bundles your repository state (or, in PR mode, points the sandbox at GitHub directly) and dispatches multiple reviewer agents in parallel. Each agent looks at the diff from a different angle. Before anything is reported, a verification pass tries to reproduce the issue against the real code, so style nitpicks and hallucinated bugs get filtered out.\n\nThe tradeoff is time and money. A run takes roughly 5 to 10 minutes and costs about $5 to $20 once you exhaust free runs, which is why it's positioned as a pre-merge check rather than a while-you-type linter.\n\n* * *\n\n### Requirements and where it won't run\n\nRequirement| Detail\n---|---\nClaude Code version| v2.1.86 or later\nAuthentication| Claude.ai account (API-key-only sessions must run `/login` first)\nExtra usage| Must be enabled to run paid reviews; check with `/extra-usage`\nPR mode| Repository needs a `github.com` remote\nUnsupported| Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, and any org with Zero Data Retention enabled\n\nIf your repo is too large to bundle and upload, Claude Code will prompt you to switch to PR mode. The workaround is to push the branch, open a draft PR, and rerun with the PR number.\n\n* * *\n\n### Running it from the CLI\n\n**Step 1:** Open a terminal inside any git repository that Claude Code recognizes, and make sure you're signed in with a Claude.ai account. If you've only authenticated with an API key, run `/login` first.\n\n**Step 2:** Run `/ultrareview` on its own to review the diff between your current branch and the default branch, including uncommitted and staged changes. Pass a PR number (for example, `/ultrareview 1234`) to have the remote sandbox clone that pull request from GitHub instead.\n\n**Step 3:** Confirm the scope in the dialog that appears. It shows what's being reviewed, how many free runs you have left, and the estimated cost. Once you accept, the review runs in the background, so you can keep coding, start other commands, or close the terminal.\n\n🧭\n\nUltrareview only runs when you invoke it explicitly. Claude won't trigger one on its own during a normal session.\n\n* * *\n\n### Tracking and stopping a review\n\nUse `/tasks` to see running and completed reviews, open the detail view, or stop a review that's still in progress. Stopping archives the cloud session and discards any partial findings, so it's an all-or-nothing call.\n\nWhen the review finishes, verified findings land as a notification in your session. Each one includes the file and line, plus an explanation you can hand straight back to Claude with a \"fix this\" follow-up.\n\n* * *\n\n### Pricing and free runs\n\nPlan| Free runs| After free runs\n---|---|---\nPro| 3, one-time allotment| Billed as extra usage\nMax| 3, one-time allotment| Billed as extra usage\nTeam and Enterprise| None| Billed as extra usage\n\nThe three free runs on Pro and Max do not refresh. After that, every run is billed as extra usage, typically $5 to $20 depending on how large the change is. If extra usage isn't enabled on your account or org, Claude Code blocks the launch and links out to billing so you can flip it on.\n\n* * *\n\n### Ultrareview vs /review\n\n| /review| /ultrareview\n---|---|---\nWhere it runs| Locally, in your current session| Remote cloud sandbox\nDepth| Single-pass review| Multi-agent fleet with independent verification\nDuration| Seconds to a few minutes| Roughly 5 to 10 minutes\nCost| Counts toward normal usage| Free runs, then about $5 to $20 per review as extra usage\nBest for| Fast feedback while iterating| Pre-merge confidence on substantial changes\n\nThe short version: `/review` is for the inner loop, `/ultrareview` is for the gate before you hit merge on something meaningful.\n\n* * *\n\n### If the command doesn't appear\n\nUltrareview is a research preview, and availability can lag behind the docs. Some users on recent Claude Code builds, including v2.1.111 on macOS with a Max subscription, have reported that `/ultrareview` and `/ultraplan` don't show up in `/help` or resolve as slash commands. The feature sits behind a feature flag and an entitlement check on the server side, so being on the right version and plan isn't always enough on its own.\n\nIf the command isn't visible:\n\n  * Confirm you're on Claude Code v2.1.86 or later with `claude --version`.\n  * Make sure you're signed in through `/login` with a Claude.ai account, not just an API key.\n  * Check that your organization doesn't have Zero Data Retention enabled and isn't routing Claude Code through Bedrock, Vertex AI, or Foundry.\n  * Verify extra usage is on via `/extra-usage`, since a disabled billing path can gate paid features.\n\n\n\nRollout is staged, so eligibility can change without a client update.\n\n* * *\n\n### When to reach for it\n\nUltrareview earns its cost on substantial, pre-merge changes, such as refactors that touch concurrency, auth, or data handling, long-running feature branches, or PRs where a single-pass review has a decent chance of missing a subtle regression. For a quick sanity check on a small diff, the local `/review` is faster, cheaper, and usually good enough. Save the multi-agent fleet for the moments where a missed bug would be expensive.",
  "title": "Claude Code ultrareview: Deep cloud bug hunt before you merge",
  "updatedAt": "2026-04-18T14:13:52.459Z"
}