YouTube Summarizer

Overview

YouTube hosts an incredible wealth of knowledge — but watching a 45-minute technical talk to find the 3 key insights is an expensive use of time. YouTube Summarizer extracts the signal from the noise, giving you structured summaries with timestamps so you can jump directly to what matters.

The skill fetches video transcripts (auto-generated or uploaded), processes them through an LLM to identify key themes and talking points, and outputs a multi-level summary: a one-paragraph TL;DR, bullet-point key takeaways, and a detailed chapter-by-chapter breakdown with timestamps.

It handles technical content particularly well — coding tutorials, conference talks, product demos — where you need accuracy, not just gist.

How It Works

1. Fetch — Extract transcript from YouTube video (supports 50+ languages)
2. Chunk — Segment transcript into logical sections based on topic shifts
3. Summarize — Generate multi-level summaries at each granularity
4. Timestamp — Link key points back to specific video timestamps
5. Format — Output as structured markdown, JSON, or plain text

Use Cases

• Learning acceleration — Get key points from educational videos in seconds
• Meeting recaps — Summarize recorded meetings and webinars
• Content curation — Screen videos before adding to reading lists
• Research — Extract quotes and claims with timestamps for citation
• Accessibility — Provide text summaries for video content

Getting Started

# Summarize a YouTube video
yt-summarize "https://youtube.com/watch?v=dQw4w9WgXcQ"

# Detailed mode with timestamps
yt-summarize --detail full --timestamps "https://youtube.com/watch?v=..."

# Output as JSON
yt-summarize --format json "https://youtube.com/watch?v=..."

Example

📺 "The Future of AI Agents" — Andrej Karpathy (42:18)

TL;DR: Karpathy argues that AI agents will evolve from
single-model to multi-agent systems, with specialization
driving performance improvements.

Key Points:
• [2:15] Current agents are "one brain doing everything" — specialization is the next leap
• [8:40] Multi-agent systems need better communication protocols
• [15:22] Tool use is the breakthrough — agents that can code, search, and execute
• [28:10] Safety considerations: agents need sandboxing and audit trails
• [35:45] Prediction: 2027 will see widespread agent-to-agent commerce

Alternatives

• Otter.ai — Meeting transcription and summarization
• Descript — Video editing with transcript-based navigation
• NoteGPT — YouTube video summarizer with notes