video-shots takes a finished video, gets cut points, durations and motion from ffmpeg scene detection, and has the model judge only shot size, category, camera move and frame content from contact sheets; 15 scripted quality gates check the result. It writes shots.json, a Markdown shot table and a single offline HTML report with a synced player and a filterable table. video-sync then takes that data and the source video and renders a copy with the current shot's details beside the picture, choosing the layout from the aspect ratio; a third skill, video-scrub, rebuilds a video without its metadata. Everything needs only node and ffmpeg (video-sync also needs a Chrome-family browser) and no API keys, which suits breaking down reference films or AI shorts for pacing and shot language.