Use this skill when the user wants yt-dlp Superpowers: download, inspect, archive, transcribe, or extract media from YouTube or other yt-dlp-supported sites using local yt-dlp. Includes workflows for video, audio, subtitles, metadata, playlists, exact Markdown transcripts, and handoffs to watch-video, perfect-cuts, or re-light.
Relight or re-scene a short talking-head / B-roll clip by transferring lighting and environment from an AI-generated reference still onto the video. Use when the user drops a video path and asks to "relight this", "change the lighting", "put me in [a scene/place]", "drop me into [environment]", "re-scene this clip", "make this look cinematic", or invokes /re-light. Two Fal.ai calls: nano-banana-2 relights one frame, Kling O3 Pro pushes that look onto the whole clip with audio preserved. Clips must be 3-10 seconds.
Turn a raw talking-head recording into a frame-perfect edited timeline - retakes removed, false starts caught, zero gap space. Drops a care package with Premiere/Resolve XML, MP4, EDL, SRT, Remotion launcher files, and a cut log. Use when the user drops a raw clip and asks to "perfect cut this", "clean cut this", "gling this", "cut the retakes out", "rough cut this footage", or invokes /perfect-cuts.
Use ONLY when the user explicitly asks to "Watch Video", "watch-video", or run the Watch Video skill. Builds a multimodal evidence bundle from a video URL or local media file by extracting transcript/subtitles plus ffmpeg screenshots at timed intervals so the AI can inspect what was visible on screen.