YouTube Transcript

Fetch YouTube captions with yt-dlp and save durable transcript artifacts. Use excerpts for narrow questions; read the full timestamp-free text when a complete summary or other whole-video task requires it, without asking for separate permission.

This skill is based on the workflow shape of Jim Liu's baoyu-youtube-transcript skill, but uses a compact Python wrapper around yt-dlp as the primary fetcher.

Workflow

Resolve this skill directory and run the bundled script. Always single-quote YouTube URLs in zsh because ? can be treated as a glob.
If the user did not specify a language, start with list when language choice matters. Otherwise default to English.
Fetch to local artifacts with fetch. Do not read transcript.md or transcript.json into context by default.
Use transcript.txt for summaries or other content-only work; it removes timestamps and is cheaper to read than transcript.md.
Use preview, search, excerpt, or bounded text output when later work needs only a small portion.
Read the full transcript only when the task truly requires full-document reasoning and no smaller excerpt/search path is enough.

Commands

python3 <this-skill-dir>/scripts/fetch-youtube-transcript.py list 'https://www.youtube.com/watch?v=VIDEO_ID'