Use Claude Code /insights to build a multi-model review gate

Watch one /insights suggestion become a shipping workflow with AI review built in.

Who it's for: Developers who use Claude Code daily and want evidence about what to improve.

The problem: Shipping code and running an AI review by other models were separate steps. Free-tier reviewers also ran out of usage on bigger reviews.

The /insights command analyzes 84 past sessions and suggests an adversarial multi-model review gate. Claude Code builds it in plan mode, fixes a bug it introduced, and tests the gate against past pull requests.

Takeaways:

  • Run /insights to see patterns across your own sessions.
  • Size each review to how critical the change is, since one line can affect 20 files.
  • Ask the agent to check its own work before you ship.
  • Judge the result by outcomes when you cannot read every line.

Resources