review-squad brings the team back to solo Claude Code work. It dispatches a panel of reviewer subagents, each carrying its own pet concern the way teammates do, against whatever you are about to ship. It runs inside the session, at a checkpoint, before you push your mistakes out for the world to see.
Install
/plugin marketplace add 2389-research/claude-plugins
/plugin install review-squad@2389-research
There’s no slash command to remember. Just say “throw all the review squad panels at this,” or ask for the squad you want, and Claude routes to the right skill:
review-squad:experts
review-squad:normies
review-squad:regulars
review-squad:well-actually
What it does
Four squads, each a different kind of reviewer:
review-squad:experts are technical specialists: SEO, accessibility, security, performance, mobile UX, copy editing, web standards. This is where the ship-blockers live.
review-squad:normies are cold visitors at different sophistication levels (a senior dev, a small-business owner, a retired teacher) landing on your work with no context and reporting first impressions.
review-squad:regulars are faux users trying to complete real tasks: the gift shopper trying to buy, the RSS subscriber trying to subscribe. Each task ends in pass, partial, or fail.
review-squad:well-actually is a panel of insufferable pedants. Each one delivers a roast and a fixlist. The roast is performative, and frankly hilarious. The fixlist is frustratingly helpful.
The personas are examples, not hardcoded. review-squad looks at what you are asking it to review and builds an appropriate panel, then asks who is missing. Add an i18n expert, a recruiter, a grandparent, or a cat if you want a cat to tell you about your CSS.
How it works
Each persona is a separate subagent, blind to the other reviewers. Experts reads code, so its reviewers go out in parallel and finish fast. The other three can drive a real browser, so they dispatch one at a time when they’re sharing a browser instance. A synthesis step consolidates everything into one report tagged by priority. When reviewers from different panels independently converge on the same finding, the noise floor drops to near zero. Those are the ones to fix first.
Running it against 2389.ai surfaced 50 findings in one pass, including a frame-ancestors * CSP header and our own company name misspelled in two languages on two legal pages. The launch post has the full story, including the failure modes we know about.
Finding does not equal fix. The panel surfaces what you would miss; you still decide what ships.
Requirements
Claude Code with the plugin system enabled. Part of the 2389 plugin marketplace.