youtube-transcript

Goal

Pull the transcript from a YouTube video and produce the most useful artifact from it — clean transcript, structured notes, or a Slack-ready summary.

Steps

1. Receive the YouTube URL. If none is provided, ask for it.
2. Run the transcript script (scripts/get_transcript.py) with the URL. The script uses yt-dlp to fetch auto-generated or manually-uploaded subtitle tracks (prefers manual/English; falls back to auto). Output: {title, channel, duration_seconds, transcript_text, language}.
3. Choose the output mode based on the request:
   • Full transcript → upload as a Slack file (plain text).
   • Summary → 3-4 sentence paragraph: what the video covers, the core argument or demo, key takeaway.
   • Structured notes → sections matching the video's natural breaks, with timestamps if present in the transcript.
   • Action items → for talks/webinars: decisions mentioned, questions raised, next steps named.
4. Include the metadata header: video title, channel, duration, language, URL.
5. Flag transcript quality: if the transcript is auto-generated (noted in output), mention that names and technical terms may be misheard.

Rules

• No Whisper audio transcription — use only the subtitle tracks yt-dlp retrieves. If no subtitles exist, say so rather than transcribing manually or refusing silently.
• Do not fabricate transcript content.
• For videos over 2 hours, default to structured notes rather than full transcript unless the user asks explicitly.
• Private or geo-restricted videos will fail extraction — report the error clearly.