YouTube Video Transcription

Overview

Use yt-transcribe to get the actual spoken content from YouTube videos. Web search and fetch tools cannot access video content - they only see metadata.

When to Use

Use this skill when:

• User shares a YouTube URL and asks what it says/contains
• User wants information that likely exists in a video
• User asks "what does this video talk about?"
• You need to extract spoken content from YouTube

Do NOT use:

• For video metadata only (title, description) - web search is faster
• When user explicitly wants just the link, not content

Quick Reference

Flag	Purpose
-q	Required for LLMs - clean stdout, no progress noise
-m tiny	Faster transcription (less accurate)
-m medium	More accurate (slower, larger model)
-o file	Save to file instead of stdout
-f srt/vtt	Include timestamps

Core Pattern

# Standard usage for LLM consumption
yt-transcribe "https://youtube.com/watch?v=VIDEO_ID" -q

# Faster but less accurate
yt-transcribe "https://youtube.com/watch?v=VIDEO_ID" -q -m tiny

# With timestamps
yt-transcribe "https://youtube.com/watch?v=VIDEO_ID" -q -f srt

Always use -q flag - suppresses progress output, gives clean transcript to stdout.

Workflow

1. User provides YouTube URL or asks about video content
2. Run: yt-transcribe "URL" -q
3. Read and summarize the transcript for the user
4. Answer their specific question using the transcript content

Common Mistakes

Mistake	Fix
Trying WebFetch on YouTube URLs	YouTube blocks bots, use yt-transcribe
Forgetting -q flag	Progress output pollutes response, always use -q
Web searching for "what does video say"	Search finds metadata, not content - transcribe instead
Using without setup	Run yt-transcribe --setup first if binaries missing (exit code 5)

Exit Codes

• 0 - Success
• 5 - Binary not found (run --setup)
• 3 - Network error
• 4 - Transcription error