YouTube Transcript Skill

Pulls captions (manual or auto-generated) for a YouTube video. The video must have captions available — most do.

Setup (one-time)

pip install youtube-transcript-api

Usage

# Print transcript to stdout
python "skills/tool-youtube-transcript/youtube_transcript.py" "https://www.youtube.com/watch?v=v4F1gFy-hqg"

# With timestamps
python "skills/tool-youtube-transcript/youtube_transcript.py" v4F1gFy-hqg --with-timestamps

# Save to file (recommended for long videos to keep context small)
python "skills/tool-youtube-transcript/youtube_transcript.py" <url> --out /tmp/transcript.txt

# Specific language
python "skills/tool-youtube-transcript/youtube_transcript.py" <url> --lang zh-Hans

Workflow when user pastes a YouTube link

1. Run the script, piping to file if the video is >~20 min: --out /tmp/yt-<id>.txt.
2. Read the file with the Read tool (using limit/offset if huge).
3. Summarize / answer the user's question against the transcript.

Notes

• WebFetch on a YouTube watch page returns only the page chrome, not the transcript — always prefer this skill.
• If a video has no captions in any language, the API raises an error. There's no fallback to audio transcription here; for that, use yt-dlp to download audio and Whisper to transcribe.
• The script accepts either a full URL or a bare 11-character video id.