What problem does it solve? Turning a raw screen recording into a watchable product demo normally requires manual video editing: finding the interesting moments, cutting dead air, adding titles, music, and zooms. This Skill automates that pipeline locally, with no SaaS tools or API keys. ## Core Features & Use Cases - Automated prep analysis: Builds a 720p proxy, transcribes narration with multilingual mlx_whisper (via a macOS host over SSH), and detects silence, scene changes, black frames, and dead spans where nothing happens on screen. - EDL-based editing: The orchestrating agent authors a JSON edit decision list mapping your prose description ("hook in first 10s, money shot at 2:30") to timed segments with speed ramps, zooms, fades, title cards, and logos. - Deterministic rendering: ffmpeg composites the final 1080p video with a music bed (file, URL via yt-dlp, or synth fallback) that ducks under narration via sidechain compression. - Use Case: Record a 15-minute screencast of your app, then ask for a 90-second demo with a lo-fi music bed — the Skill transcribes, finds the dead spans, cuts around them, and renders the reel. ## Quick Start Ask the agent to make a 90-second demo video from your recording file, describing the hook, the money shot timestamp, the vibe, and optionally a music track URL.
