| name | video-transcript |
| description | Pull a YouTube video's transcript into context for discussion — no notes written. Invoke with `/video-transcript <youtube-url>` or when user says "pull the transcript", "transcript this video", "let me read this video first", "what does this video say". Downloads and cleans the transcript via yt-dlp, then summarizes it for conversation. Hand off to `/video-to-zettelkasten` when ready to make notes. |
Video Transcript
Fetch a YouTube video's transcript, clean it, load it into context, and discuss
it. This skill stops at "here's what the video says" — it writes no notes. When
the user is ready to turn the conversation into permanent notes, hand off to
/video-to-zettelkasten, which reuses the same cleaned transcript already on
disk.
Workflow
-
Download metadata — keep only the useful fields:
yt-dlp --dump-json --no-download "<url>" | jq '{title, uploader, channel, upload_date, duration, webpage_url, id, description: (.description | .[0:500])}'
-
Download transcript — manual and auto subs in one call:
yt-dlp --write-subs --write-auto-subs --sub-langs "en" --skip-download --sub-format vtt -o "/tmp/yt-zk-%(id)s" "<url>"
Note the yt-zk- prefix and /tmp path are shared with video-to-zettelkasten on purpose — that skill looks for exactly this file to skip its own fetch.
-
Clean the transcript — reuse the bundled script from the ZK skill:
python3 /Users/joec/.claude/skills/video-to-zettelkasten/scripts/clean-vtt.py /tmp/yt-zk-<id>.en.vtt
This writes a cleaned .txt next to the .vtt and prints the word count. Then use the Read tool on the cleaned .txt — do not pipe it through cat/sed/head.
-
Summarize for discussion — present:
- Video title, channel, duration, upload date
- A 3-5 sentence summary of what the transcript covers and its main argument
- 3-6 bullet points of the most notable ideas, claims, or moments
Then stop and let the user drive the conversation. Answer questions against
the transcript already in context. Do not write any files.
Handoff to notes
When the user says "ZK this", "make notes", "atomize it", or similar, invoke
/video-to-zettelkasten <url>. The cleaned /tmp/yt-zk-<id>.en.txt is already
on disk, so that skill skips its own fetch and goes straight to extraction.
Leave /tmp/yt-zk-* in place until the notes are written — that's the handoff
artifact. Only clean up if the user ends without making notes.
Example Invocations
/video-transcript https://www.youtube.com/watch?v=dQw4w9WgXcQ- "Pull the transcript for https://youtu.be/abc123, I want to read it first"
- "What does this talk actually say?" (with URL in message)
