Skip to content
TakoAPI
Back to skills

youtube-knowledge-extractor

by sdrabent

Coding Agents & IDEs
512Downloads reported by clawskills.sh
2Stars
0Views
0 likesSign in to like this skill
—Not yet rated
Your rating

Brief

Multimodal YouTube video analysis through both audio (transcript) and visual (frame extraction + image analysis)

Description

What This Skill Does

Analyzes YouTube videos through both audio and visual channels by extracting timestamped transcripts and key frames from the video. Pairs spoken content with on-screen visuals, so UI actions, code, and diagrams shown but not described in speech are captured accurately.

Standard YouTube transcript tools miss anything shown on screen but not spoken; this skill matches frames to transcript timestamps to close that gap.

When to Use It

  • Building step-by-step guides from software tutorial videos
  • Extracting code shown on screen during live coding demos
  • Summarizing long explainer videos with visual anchors
  • Following UI navigation in product walkthrough videos
  • Capturing diagram details from architecture or design videos

Example Workflow

INPUT

User asks: 'Create a step-by-step guide from this YouTube tutorial: https://youtube.com/watch?v=...'

AGENT
  1. 1

    Fetches video metadata (title, duration, ID) using yt-dlp

  2. 2

    Extracts timestamped transcript via yt-dlp JSON dump and curl

  3. 3

    Downloads video at 720p and extracts frames at adaptive intervals using ffmpeg

  4. 4

    Synchronizes each frame to its corresponding transcript segment by timestamp

  5. 5

    Reads each frame for on-screen text, UI elements, and code, then combines with audio to generate a structured guide

OUTPUT

Timestamped step-by-step guide with descriptions combining spoken instructions and visible UI details, including button names, menu paths, and code shown on screen

Requirements

  • ffmpeg installed on system
  • yt-dlp installed on system

Installation

clawhub install sdrabent/youtube-knowledge-extractor

Run the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.

Links