Agent skill for extracting and summarizing YouTube transcripts with yt-dlp

SKILL.md

name	youtube-transcript
description	Load when user asks to fetch, extract, or summarize a YouTube video transcript. Contains yt-dlp workflow, subtitle parsing, and summarization patterns.

YouTube Transcript Extraction & Summarization

Reliable workflow for fetching and processing YouTube video transcripts using yt-dlp.

Prerequisites

• Python 3 with pip
• yt-dlp: Install with python -m pip install --user yt-dlp
• Verify: python -m yt_dlp --version

Step 1: Download Subtitles

MUST use python -m yt_dlp (not bare yt-dlp) to guarantee the correct installation is invoked.

Auto-generated subtitles (most common)

python -m yt_dlp --skip-download --write-auto-subs --sub-langs "en.*" --sub-format vtt -o "%(id)s.%(ext)s" "VIDEO_URL"

Manual/uploaded subtitles (if available)

python -m yt_dlp --skip-download --write-subs --sub-langs "en" --sub-format vtt -o "%(id)s.%(ext)s" "VIDEO_URL"

List available subtitles first

python -m yt_dlp --list-subs "VIDEO_URL"

Key flags

Flag	Purpose
--skip-download	Do NOT download the video file
--write-auto-subs	Fetch auto-generated captions
--write-subs	Fetch manually uploaded captions
--sub-langs "en.*"	English variants (en, en-US, en-GB, etc.)
--sub-format vtt	WebVTT format (cleanest for parsing)
-o "%(id)s.%(ext)s"	Output filename = video ID

Output

Produces a .vtt file, e.g., VIDEO_ID.en.vtt.

Step 2: Read and Summarize

1. Read the full VTT file using the Read tool. VTT auto-subs contain repeated/overlapping lines (rolling windows) — this is normal, just read through it.
2. Synthesize into a coherent summary.
3. Structure the output with clear headings, bullet points, and blockquotes for key phrases.

Recommended summary structure

## Core Thesis
One-paragraph overview of the speaker's main argument.

## Key Concepts
Bulleted list of the main ideas with brief explanations.

## Supporting Arguments / Evidence
Deeper detail on how the speaker supports their claims.

## Key Quotes / Phrases
Notable exact phrases worth preserving.

## Takeaways
Actionable or memorable conclusions.

Tips & Gotchas

• Auto-subs have no punctuation or speaker labels. Expect run-on text.
• Prefer --sub-format vtt over srt — VTT is cleaner to parse and yt-dlp handles it more reliably.
• Some videos have no English subs. Use --list-subs first if download fails, then try other languages.
• Age-restricted or private videos may require cookies: --cookies-from-browser chrome or --cookies /path/to/cookies.txt.
• Output directory: Use a temp directory, not the project root.
• If yt-dlp is not installed, install it with python -m pip install --user yt-dlp. Do NOT use pip install directly (may target wrong Python).