You describe a video (brief, duration, aspect ratio) and can add a brand lock, a markdown file fixing palette, type and voice, which brand-lock-extractor can build from a site, PDF or screenshots; without one the skills fall back to a flagged template. storyboard-architect writes run.json, storyboard.md, shots.json and a separate on-screen text spec, visual-prompt-forge writes copy-paste prompts for six image and four video generators, and storyboard-html-preview builds one shareable page. After you generate frames elsewhere, visual-asset-critic checks each against its shot spec and brand lock, and the prompts are re-written only for shots marked for revision, with content hashes recorded so later changes can be detected. It stops at specs and prompts: it calls no generator APIs and renders no video.