YouTube transcript

Captions only. Never download the video.

The fetch runs in a subagent, and that is the whole point: a 25-minute video is ~8k tokens of transcript and a two-hour one is 40k, and writing even a three-line summary means reading all of it. The subagent does that read and hands you a report. Do not read the transcript here to check its work.

Process

1. Source. Accept youtube.com/watch?v=, youtu.be/, youtube.com/shorts/, youtube.com/embed/, youtube.com/live/, or a raw 11-char id [A-Za-z0-9_-]{11}. Normalize to https://www.youtube.com/watch?v=<id>. Anything else: stop, without spawning anything. Done when you have that watch URL.