What problem does it solve? Ambitious artifacts often ship with unverified quality because the same agent both builds and judges its own work. This Skill separates builders from fresh critics and loops until direct comparison evidence shows the artifact meets a concrete, inspectable bar. ## Core Features & Use Cases - Bar Definition: Establishes an observable goal, comparison method, materiality threshold, and stop policy before any building begins. - Builder/Critic Separation: Assigns builders isolated git worktrees while fresh critics judge artifacts blind, returning WIN, LOSE, or UNJUDGEABLE verdicts with evidence. - Convergence Ledger: Tracks rounds, verdicts, gaps, and repairs in a compact table, stopping only on a win, resource limit, or sub-threshold remaining gap. - Use Case: Iteratively refine a UI redesign by pinning a rendered reference and capture protocol, then running build-and-judge rounds until screenshots match the bar or the budget is exhausted. ## Quick Start Run a gauntlet loop on my landing page redesign against the reference screenshot, with a budget of six rounds and a stop threshold for pixel-level differences.
