Research Extract — Communication skill for Claude Code

research-extract

Extract structured insights from YouTube videos, podcasts, blogs, PDFs, and audio files using Claude Code's parallel agent teams.

What it does

Feed it any content source and it will:

1. Ingest the content (download captions, transcribe audio, fetch articles)
2. Chunk long transcripts into processable pieces
3. Extract insights in parallel using agent teams (3 agents processing chunks simultaneously)
4. Merge and rank all findings into a consolidated analysis
5. Generate variants — Show Notes and Cheat Sheet HTML pages

What gets extracted

• Key insights with quotes and attribution
• Notable quotes
• Recurring themes with frequency counts
• Challenges and problems discussed
• Solutions and approaches proposed
• Concrete action items
• Frameworks and mental models
• External resources (books, tools, people, podcasts, courses)

Install

npx skills add katyella/research-extract

This installs into `.claude/skills/research-extract/` in your project.

First-time setup

bash .claude/skills/research-extract/scripts/setup.sh

Checks for required system tools (yt-dlp, pdftotext, whisper).

**Required:** `yt-dlp` (for YouTube). Install with `brew install yt-dlp` or `pip install yt-dlp`.

**Optional:** `whisper` (for audio without captions), `pdftotext` (for PDFs).

Usage

In Claude Code, use the `/research-extract` command:

/research-extract ingest https://youtube.com/watch?v=abc123 as my-talk
/research-extract analyze my-talk
/research-extract variants my-talk
/research-extract list sources

Full workflow example

> /research-extract ingest https://youtube.com/watch?v=abc123 as startup-advice

Ingested: startup-advice (How to Start a Startup - Sam Altman)
Transcript: