review-pr

A first round runs probe → raise → prove and ends in a task list. It never edits the code: whoever implements works the list. A later round, after fixes land, is fix-verify alone. Handed the task list of an earlier round, run a later round. Handed none, run a first round.

You are the carrier. You probe, pick the dimensions, run review.js through the Workflow tool, and write the report. The script runs every reviewing agent and sets each one's model and effort, so your own session needs no more than medium effort. This skill telling you to call Workflow is the opt-in that tool asks for, and the script decides how many agents run.

What each stage is for