ytube

name	ytube
description	Transcribe a YouTube video to markdown using yt-dlp auto-captions. Use when a YouTube URL appears, or the user asks to transcribe/summarise a video.

ytube

Transcribe a YouTube video to a markdown file using yt-dlp's auto-captions.

Arguments

A YouTube URL. If not provided, ask the user.

Your Task

1. Get video metadata

yt-dlp --print "%(title)s" --print "%(channel)s" --print "%(duration_string)s" --print "%(webpage_url)s" --skip-download "$URL" 2>/dev/null

Output is 4 lines, in order: title, channel, duration_string, webpage_url

2. Generate filename

Sanitize video title (replace / : " with -, trim to 80 chars)

Ensure filename ends with the normal markdown extension: ".md"

3. Download auto-captions

yt-dlp --write-auto-sub --sub-lang "en-orig,en" --sub-format vtt --skip-download -o "/opt/clones/github/jalanb/tmp/ytube_$$" "$URL"

The VTT file will be at /opt/clones/github/jalanb/tmp/ytube_$$.en-orig.vtt or /opt/clones/github/jalanb/tmp/ytube_$$.en.vtt

Note that "en-orig" captions may not be available, only "en"

4. Clean and convert to markdown

clean_vtt.py (in this skill's ytube/ package) cleans VTT formatting from the transcript, unwraps caption line-breaks into proper paragraphs, and writes a markdown file.

The call looks like this example if the "en-orig" captions are available:

python3 -m ytube.clean_vtt \
    "/opt/clones/github/jalanb/tmp/ytube_$$.en-orig.vtt" \
    "$filename" \
    "$title" \
    "$channel" \
    "$duration_string" \
    "$webpage_url"

The script will output the full path to the markdown file it wrote, under /opt/clones/github/jalanb/clawdrs/clawdr/skills/ytube/transcripts/

5. Cleanup

rm -f /opt/clones/github/jalanb/tmp/ytube_$$.*

6. Report

Show the user the output from the script, which is expected to be a path.

User will normally copy/paste that path into another tool, e.g. vim

7. Review

Give user a chance to read transcript, then ask user about the quality of the transcript. Then consider whether any changes to this text/skill might improve the next transcript for user If there are such changes, propose them to user.