youtube-transcript-library

name

youtube-transcript-library

description

Narrow entry skill for YouTube transcript ingestion and library stub creation. Use when the user pastes a youtube.com or youtu.be URL and wants a transcript saved to disk and/or a library entry created. Fetches captions via fetch-transcript.py and walks the 4-step library ingest checklist. For full pipeline work (citation verification, multi-source analysis, batch gathering), hand off to the research-and-analyze skill after the transcript is on disk.

<essential_principles>

Scope

This skill handles one narrow path: YouTube URL → transcript on disk → library entry.

It does not run citation verification, claims analysis, or multi-source pipelines. If the user wants those, complete the transcript step here, then route to research-and-analyze.

Script

Use fetch-transcript.py — located at .cursor/skills/research-and-analyze/scripts/fetch-transcript.py.

Do not use yt-dlp directly for captions — the script handles format normalisation and saves structured markdown.

# Single video
python3 .cursor/skills/research-and-analyze/scripts/fetch-transcript.py \
  --url "https://www.youtube.com/watch?v=VIDEO_ID" \
  --out "research/{slug}/sources/"

# Output: research/{slug}/sources/{title-slug}.md  (timestamped transcript)

Pick a subject slug that will serve as the research directory name — e.g., karpathy-software2, rich-hickey-simple-made-easy.

</essential_principles>

Step 1 — Fetch transcript

Run fetch-transcript.py with the user-provided URL. Confirm the output file exists and is non-empty before continuing.

Step 2 — Gather metadata

From the transcript file or the video, collect:

• Title
• Speaker / channel
• Date published
• URL
• Wing tag (see below)
• One-paragraph summary (synthesise from transcript — do not fabricate)

Wing tags:

Tag	Topics
ai-engineering	agents, harness, context, memory, models, agentic-workflow
philosophy-practice	zen, karate, flow, solitude, martial arts
devops	git, openshift, rhacm, aap, ansible
leadership-org	career, AI organisational impact, team dynamics

Step 3 — Library ingest (all four steps required)

Every ingest must complete all four:

1. Create entry file — library/{slug}.md

   • Frontmatter: title, speaker, date, url, wing, tags
   • Body: summary, key ideas, quotes worth keeping, link to transcript in research/

2. Add row to library/catalog.md — slug, title, speaker, date, wing

3. Add entry block to library/README.md — Enriched Entries table row

4. Append dated entry to library/log.md — date, slug, one-line description

Step 4 — Confirm and hand off

Tell the user:

• Transcript saved at: research/{slug}/sources/…
• Library entry created at: library/{slug}.md
• Offer: "If you want citation verification or claims analysis, I can hand off to the full research-and-analyze pipeline."

Route to research-and-analyze when the user wants:

• Claim verification against cited sources
• Multi-source or batch gathering
• Full gather → analyze → synthesize pipeline
• Evaluation of the talk with evidence scoring

The transcript file produced here becomes the starting source for that pipeline — no re-fetching needed.