Video transcribe

Gets the most faithful transcript a video can give you, and tells you when that transcript is not trustworthy. Emits transcript.json (millisecond-timed segments) and metadata.json. Everything here is deterministic — no model is involved in acquisition or in the quality verdict.

When to use

The user wants the words: "transcribe this", "get me the transcript", "what did they say", "grab the captions".

When not to: they want a written-up note or summary (video-notes), or they want images and on-screen text (video-keyframes). If they asked for notes, video-notes calls this skill for you — do not run it directly.

Workflow

1. Run the extractor