Turns benchmark videos, contact sheets, or rough ideas into four structured objects: hook essence, viewer question, must-copy visual grammar, and forbidden drift. The point is to extract reusable prompt structure before you write a storyboard or generation request, not to summarize a reference as "good vibe" or "nice pacing." It separates what you copy (camera grammar, shot order, timing logic) from what you don't (faces, exact wardrobe, creator identity). If you have no reference at all, it runs in proxy mode and infers structure from your segment pattern and brief. Stops cleanly when source evidence is missing instead of guessing.
npx skills add https://github.com/postplusai/postplus-skills --skill reference-decode