Adversarial Review

The free adversarial review skill for Claude Code

One command, run when your AI says it’s finished. A separate agent that never saw the conversation reads the actual changes, assumes they’re wrong, and tells you in plain English whether they’re safe to ship.

We'll send occasional new posts and skill updates. Unsubscribe any time. Privacy.

What you get

  • A fresh reviewer. It runs in its own agent with no access to the chat that produced the code, so it can’t inherit the builder’s blind spots.
  • Read only. The review agent isn’t allowed to edit files. It reports; your builder fixes.
  • One of three verdicts. SHIP, FIX FIRST, or STOP, with the reason in one sentence at the top.
  • Findings in plain English. Each one says what a real customer would experience, where it is, and how to confirm it.
  • A fix list to paste back. Copy it straight into your main chat.

It checks the places AI-written changes most often go wrong: work that was quietly left unfinished, money and payments, who can see whose data, deleting or overwriting records, errors that fail silently, and tests that don’t prove anything.

Install it in two minutes

  1. Unzip the download. You’ll get a folder called adversarial-review with one file inside, SKILL.md.
  2. Move that folder into ~/.claude/skills/ to use it in every project, or into .claude/skills/ inside a single project. (Anthropic’s skills documentation covers both.) If you’d rather not touch folders, open Claude Code, attach the file, and ask it to install the skill for you.
  3. Use a recent version of Claude Code (2.1.218 or later).

Use it

When your AI finishes a change, type:

/adversarial-review <one sentence describing what you asked for>

For example: /adversarial-review add a cancel button that stops the subscription and emails the customer.

You can also point it at a GitHub pull request by number: /adversarial-review PR 42, the new invoice export.

New to the idea? Start with the plain-English guide.