What problem does it solve? When tuning a game's look or sound, you often only have a vague direction like "warmer colors" or "shorter, sharper sound" and no safe way to compare options without overwriting live assets. This Skill drafts several candidate Docs in a sandboxed atelier/ directory, renders them headlessly, and reports a self-critique so a human can pick the winner without risking the production assets/ folder. ## Core Features & Use Cases - Sandboxed candidate drafting: Writes N candidate Docs to atelier/ with slot-compatible entityId and version, never touching assets/. - Headless rendering and review: Renders candidates via make targets (render, gallery-prologue, gallery-sounds) and inspects contact sheets, waveforms, and spectrograms in debug/. - Structured self-critique: Reports each candidate's intent, fit to the stated direction, harmony with existing assets, and a recommended pick with reasoning, leaving promotion to human judgment. - Use Case: You want a warmer color theme for a character. The Skill reads the slot's schema, writes three variants (a/b/c) to atelier/, renders them, critiques each against the direction, and recommends one for you to approve. ## Quick Start Ask the agent to draft three warmer-color candidates for the character theme slot and report which one fits best.
