Turn talking-head footage into an editable AI rough cut

The video-use Skill turns a folder of raw footage into an editable AI rough cut. Give a coding Agent an editing brief and it can transcribe speech, remove selected pauses and filler words, apply basic colour work, captions, transitions, and visual overlays, then write edit/final.mp4.

Best fit, input, and result

It fits talking-head, tutorial, and spoken videos where transcript and pauses are meaningful editing signals. It is less reliable for action-led, music-led, or mostly silent footage because transcript and sampled visual summaries can miss the important event.

Start with one to three minutes of non-sensitive footage. Ask only for obvious pauses to be removed and captions added. Check names and terms in the transcript, compare the cut with the original for changed meaning, then make a second natural-language request such as keeping a sentence or changing caption timing.

Setup and privacy boundary

The upstream workflow requires ffmpeg and Python tooling; yt-dlp is optional for online material. Its README uses ElevenLabs Scribe for transcription, so this is neither zero-configuration nor necessarily local-only: audio may go to a third-party service and incur cost or privacy constraints. The manual examples are macOS-oriented. Windows users need to verify PATH, ffmpeg, and Python in a test folder rather than assume identical steps.

Review before delivery

The Skill uses transcript plus selected frames or a visual composite, not continuous human-style viewing. Fast visuals, silent events, and subtle expressions can be missed. A person must review the full video, especially when automatic filler-word removal changes timing or emotion.

Source and license

GitHub currently provides no packaged Release, so installation and updates should follow the repository's install.md, SKILL.md, and README.