youtube-fetcher（YouTube 抓取器）

YouTube Fetcher

Overview

Fetches transcripts from YouTube videos and saves them as structured Markdown files with YAML frontmatter. Uses two tools:

• yt-dlp — video metadata (title, channel, description, duration, chapters)
• youtube-transcript-api — transcript/captions extraction

Each file includes full video metadata, the video description (with links and chapters), and the transcript. Files are saved to ~/yt_transcripts/ with the naming convention YYYY-MM-DD_video-title-slug_[VIDEO_ID].md.

When to Use This Skill

• User shares a YouTube URL (any format) or video ID
• User wants a transcript, captions, subtitles, or "the text" from a video
• Note-taking or summarization workflows involving YouTube content
• User pastes a YouTube link — even without explicit instructions, default to fetching

How Claude Should Use This Skill

The script path is: ~/.config/skillshare/skills/youtube-fetcher/scripts/fetch_transcript.py

Always use --force when running from Claude — the duplicate check uses an interactive prompt that doesn't work in non-interactive contexts. If you need to check for duplicates first, look for matching files in ~/yt_transcripts/ yourself.

Typical workflow:

1. Run the script with --force to fetch the transcript
2. Tell the user where the file was saved
3. If the user wants a summary or analysis, read the saved file and process it

Example invocation:

python3 ~/.config/skillshare/skills/youtube-fetcher/scripts/fetch_transcript.py "https://www.youtube.com/watch?v=VIDEO_ID" --force

If the user wants the transcript in the conversation (not just saved to a file), add --stdout:

python3 ~/.config/skillshare/skills/youtube-fetcher/scripts/fetch_transcript.py "URL" --force --stdout

If deps are missing, the script exits with code 2 and prints install instructions. Run the pip/brew commands it suggests.

Prerequisites

The script checks dependencies automatically on each run and will guide you through installation if anything is missing. You can also run a manual check:

python3 ~/.config/skillshare/skills/youtube-fetcher/scripts/fetch_transcript.py --check-deps

Required:

pip install youtube-transcript-api requests

Recommended (for video description, chapters, duration):

brew install yt-dlp  # or: pip install yt-dlp

Usage

Default (saves structured Markdown to ~/yt_transcripts/)

python3 ~/.config/skillshare/skills/youtube-fetcher/scripts/fetch_transcript.py <youtube_url_or_id>

This creates a file like ~/yt_transcripts/2026-03-04_video-title-here_[dQw4w9WgXcQ].md.

Duplicate Detection

If a video was already transcribed, the script will:

• Interactive terminal: prompt Re-transcribe anyway? [y/N]:
• Non-interactive (Claude, pipes, scripts): auto-skip and suggest --force

Use --force to always re-fetch regardless of duplicates.

Options

# With timestamps in transcript body
python3 .../fetch_transcript.py <url> --timestamps

# Specific language
python3 .../fetch_transcript.py <url> --lang es

# Override source project name
python3 .../fetch_transcript.py <url> --source "my-project"

# Custom output path
python3 .../fetch_transcript.py <url> --output ~/notes/my-transcript.md

# Skip video description (transcript only)
python3 .../fetch_transcript.py <url> --no-description

# Force re-fetch (skip duplicate check)
python3 .../fetch_transcript.py <url> --force

# Print to stdout instead of saving
python3 .../fetch_transcript.py <url> --stdout

# JSON or SRT format (raw, no Markdown wrapping)
python3 .../fetch_transcript.py <url> --format json
python3 .../fetch_transcript.py <url> --format srt

# List available transcripts for a video
python3 .../fetch_transcript.py <url> --list

# Check dependencies
python3 .../fetch_transcript.py --check-deps

Exit Codes

Output Structure

File Location

~/yt_transcripts/YYYY-MM-DD_video-title-slugified_[VIDEO_ID].md

File Content

---
title: "Video Title"
channel: "Channel Name"
url: "https://www.youtube.com/watch?v=VIDEO_ID"
video_id: "VIDEO_ID"
fetched: "2026-03-04"
source_project: "JARVIS-Obsidian-Setup"
language: "en"
caption_type: "manual"
duration: "36m 26s"
upload_date: "2024-01-15"
tags:
  - yt-transcript
---

# Video Title

## Video Details

| Field    | Value |
|----------|-------|
| URL      | https://www.youtube.com/watch?v=VIDEO_ID |
| Channel  | Channel Name |
| Duration | 36m 26s |
| Uploaded | 2024-01-15 |
| Fetched  | 2026-03-04 |
| Source   | JARVIS-Obsidian-Setup |
| Language | en (manual) |

## Video Description

The video description text with links, summaries, etc.

### Chapters

- `00:00` Intro
- `02:23` My Testimony
- ...

## Transcript

The full transcript text here...

Accepted URL Formats

• https://www.youtube.com/watch?v=VIDEO_ID
• https://youtu.be/VIDEO_ID
• https://youtube.com/embed/VIDEO_ID
• Just the video ID directly

Limitations

• Only works on videos that have captions (manual or auto-generated)
• Cannot transcribe videos without any captions — use Whisper for that
• Some videos have captions disabled by the uploader