youtube-knowledge-extractor_skill

Overview

This skill performs multimodal analysis of YouTube videos by combining timestamped transcripts with extracted video frames. It synchronizes what is said with what is shown to produce accurate step-by-step guides, visual anchors, and discrepancy detection. It is optimized for HowTos, tutorials, demos, and explainer videos where visual detail matters as much as narration.

How this skill works

The skill fetches video metadata and downloads a reliable transcript (preferring manual subs then auto-captions). It extracts frames at adaptive intervals or via scene-change detection, assigns timestamps to each frame, and pairs frames with corresponding transcript segments. Finally it analyzes images for UI elements, code, text, diagrams, and physical actions, synthesizing both channels into structured outputs like step-by-step guides or comprehensive summaries.

When to use it

• You want a step-by-step guide from a tutorial or demo video.
• A user shares a YouTube URL and asks "What happens in this video?"
• You need visual anchors (screenshots) linked to spoken instructions.
• You need to extract code, error messages, or UI details not spoken aloud.
• You require a synchronized summary of audio and visual content for documentation.

Best practices

• Prefer transcript-first: scan timestamps for phrases like "as you can see" to target frames.
• Use scene-change detection for UI/software demos; tighter intervals for physical HowTos.
• Extract at 720p for a good balance of text readability and download size.
• Flag and report visual-only information explicitly (button names, on-screen code).
• Ask the user to narrow time ranges for very long videos (>60 min).

Example use cases

• Generate a step-by-step installation guide from a 10-minute software tutorial with screenshots.
• Summarize a product demo, highlighting UI flows and exact button labels shown on screen.
• Extract and transcribe on-screen code snippets that the presenter does not read aloud.
• Produce a visual-only report when subtitles are unavailable or unreliable.
• Detect mismatches where the narrator describes one thing but the screen shows another.