Plan review and proof

Two checks keep agent work honest: you approve the plan before code is written, and you watch a recorded test before you merge.

Review plans with Plannotator

Ask an agent to write its plan first. Start Plannotator review opens the plan inside Moneypenny, where you can read it, comment on it, and approve it before any code is written.

Recorded browser tests

Ask for a browser test along with a change. The agent runs it with Playrust on your machine, and each run lands in the chat as a result card: whether it passed, which step failed if it didn't, and the screen recordings.

Only recordings from passing runs count as proof. A failed run is shown with its failing step, never as evidence that the change works. Every recording is kept in the chat's Proof panel (/evidence).

Diffs and pull requests

/diff shows what changed in the chat's worktree. When you are happy with it, open a pull request from the /pr panel.

A second opinion

Mention a reviewer bot in any coding chat. It reads the work so far and points out what was missed. See getting a second opinion from another model.

Next: Software factories