What it does
Turns assets, copy and a plan into a finished video, then polishes it. The working backend is local FFmpeg, driven by one Python script with four commands: render (a simple text-card video from a JSON spec), jumpcut (detect silences and cut them out of a talking-head or lecture video), burn-subtitles (burn an SRT file into the picture) and polish (H.264 and yuv420p, volume normalisation and optional fades), plus a probe that checks for ffmpeg, node and an optional VectCutAPI service. Remotion, VectCutAPI and an MCP wrapper are described as optional or planned backends. It lists quality checks before delivery and says it does not publish to any platform.
Output
A finished MP4 under outputs/video-studio/final in your working folder.
Good for
Quick, scriptable edits of talking-head and course videos: remove pauses, add subtitles, normalise audio. The Skill text is in Chinese.
Medium risk: not verified as a whole. It is one Python script that runs local ffmpeg and writes files under outputs/video-studio in the current folder; it does not send anything over the network (it only contacts a service if you set a VectCutAPI address). The script was run directly, not driven by an agent, on a generated test clip with Homebrew ffmpeg on macOS: probe, jumpcut and polish worked, but render and burn-subtitles failed because that ffmpeg build lacked the drawtext and subtitles filters, so check your build. SKILL.md hard-codes the author's personal Windows paths, so adjust them. By foryourhealth111-pixel, Apache-2.0.