AgentSkillsCN

YouTube Transcriber Pipeline - Claude Code Skill

YouTube 字幕管道——Claude 代码技能

SKILL.md

YouTube Transcriber Pipeline - Claude Code Skill

Version: 1.0.0 | Status: Ready for Claude Code Integration

Overview

Complete 4-skill pipeline for extracting, formatting, organizing, and archiving YouTube transcripts. Integrates with Claude Code CLI using the system's configured API keys (no manual key management needed).

Capabilities

This skill provides a complete workflow:

  1. •Extract Facts - AI-powered fact extraction from transcripts (uses Claude API via Claude Code)
  2. •Format Notes - Convert facts to atomic markdown notes with wiki-links
  3. •Integrate Obsidian - Import notes into Obsidian vault with backlinks
  4. •Archive Transcripts - Safe archival with integrity verification

Usage in Claude Code

bash
# Run complete pipeline
/run-youtube-transcriber --transcript ~/Downloads/transcript.txt

# Just extract facts
/extract-transcript-facts ~/Downloads/transcript.txt

# Just format notes
/format-transcript-notes ~/Downloads/notes/ --output ./formatted

# Just integrate Obsidian
/integrate-obsidian-vault ./notes/ --vault-path ~/Documents/Obsidian/Aaron

# Just archive
/archive-transcripts ~/Downloads/transcript.txt

Features

Phase 1: Extract Facts (AI-Powered)

  • •Uses Claude 3.5 Sonnet via Claude Code CLI
  • •7 fact categories (scientific, quotes, methods, examples, statistics, warnings, connections)
  • •Hierarchical topic organization
  • •Confidence scoring (0.0-1.0)
  • •Automatic metadata (actionable, philosophical, controversial)
  • •No API key needed - Uses Claude Code's configured authentication

Phase 2: Format as Atomic Notes

  • •1 fact = 1 markdown file
  • •Wiki-link cross-references [[Topic Name]]
  • •YAML front matter with metadata
  • •Topic indices and hierarchies
  • •Obsidian vault compatibility

Phase 3: Integrate Obsidian

  • •Safe vault backup before changes
  • •Preserve directory structure
  • •Auto-generate backlinks
  • •Master index creation
  • •Link verification
  • •Dry-run mode for testing

Phase 4: Archive Transcripts

  • •tar.gz compression
  • •MD5 integrity verification
  • •Metadata tracking
  • •Optional cleanup
  • •Verify-only mode

Installation

The skill is ready to use once copied to Claude Code:

bash
# Copy to Claude Code skills directory
cp -r ~/github/astoreyai/claude-skills/ ~/.claude/skills/youtube-transcriber

# Or individual components
cp -r youtube-transcript-extractor ~/.claude/skills/
cp -r transcript-to-logseq ~/.claude/skills/
cp -r transcript-to-obsidian ~/.claude/skills/
cp -r transcript-archiver ~/.claude/skills/

Requirements

For Claude Code CLI

  • •Claude Code installed and configured
  • •API key already configured in Claude Code (no manual setup needed)
  • •Python 3.11+

For Obsidian Integration

  • •Obsidian vault at ~/Documents/Obsidian/Aaron/ (configurable)
  • •Read/write permissions to vault directory

Architecture

code
Claude Code CLI
    ↓
Transcript Extractor (Phase 1)
    ├─ Uses Claude Code's API key automatically
    ├─ Extracts facts via Claude API
    └─ Outputs Fact objects
    ↓
Logseq Formatter (Phase 2)
    ├─ Converts facts to atomic notes
    └─ Creates markdown files
    ↓
Obsidian Integrator (Phase 3)
    ├─ Imports notes to vault
    ├─ Generates backlinks
    └─ Creates indices
    ↓
Transcript Archiver (Phase 4)
    ├─ Compresses files
    ├─ Verifies integrity
    └─ Stores archive

Future: Ollama Integration

The architecture is designed to support local LLM models via Ollama:

bash
# Future upgrade (when Ollama integration ready)
/run-youtube-transcriber --transcript ~/Downloads/transcript.txt \
  --llm-backend ollama \
  --model mistral-7b

# Benefits:
# - No API costs
# - Runs completely locally
# - Works offline
# - Full privacy (data never leaves your machine)

Swapping backends will be as simple as changing a configuration parameter.

Testing

All 4 skills are fully tested:

  • •Phase 1: 18 tests (extraction logic)
  • •Phase 2: 21 tests (formatting, hierarchy)
  • •Phase 3: 6 tests (integration, backlinks)
  • •Phase 4: 18 tests (archival, verification)
  • •Integration: 6 tests (cross-skill compatibility)

Total: 69/69 tests passing (100%)

Performance

  • •Fact extraction: 50-100ms per fact (via Claude API)
  • •Note formatting: ~500ms per note
  • •Obsidian integration: <1s for 100 notes
  • •Archival: <2s for 1000 files

Data Flow

code
Raw Transcript (txt/md)
    ↓
Phase 1: Extract Facts
    → Fact objects with metadata
    ↓
Phase 2: Format Notes
    → Markdown files in ~/output/notes/
    ↓
Phase 3: Integrate Obsidian
    → Updated ~/Documents/Obsidian/Aaron/
    ↓
Phase 4: Archive
    → ~/youtube-transcriber/transcripts-archive/

Example Workflow

1. Extract Facts

bash
/extract-transcript-facts ~/Downloads/ml-lecture.txt

Output:

code
✓ Transcript loaded (5,234 chars)
✓ Extracting facts...
✓ Facts extracted: 12
✓ Topics identified: 4
✓ Accuracy: 92%

2. Format Notes

bash
/format-transcript-notes ~/Downloads/ml-lecture.txt --output ./notes

Output:

code
✓ Creating atomic notes...
✓ Notes created: 12
✓ Topics structured: 4
✓ Output: ./notes/

3. Integrate Obsidian

bash
/integrate-obsidian-vault ./notes/ --vault-path ~/Documents/Obsidian/Aaron

Output:

code
✓ Vault path: ~/Documents/Obsidian/Aaron
✓ Notes integrated: 12
✓ Backlinks created: 8
✓ Index created: Video-Transcripts/index.md

4. Archive

bash
/archive-transcripts ~/Downloads/ml-lecture.txt

Output:

code
✓ Files archived: 1
✓ Archive: ~/youtube-transcriber/transcripts-archive/transcripts_20251122_143200.tar.gz
✓ Size: 45.2 KB

Configuration

Edit config.yaml in the skill directory:

yaml
# Paths
vault_path: "~/Documents/Obsidian/Aaron"
archive_dir: "~/youtube-transcriber/transcripts-archive"

# Extraction
max_facts: null  # No limit
confidence_threshold: 0.75
extract_all_categories: true

# Formatting
notes_per_file: 1  # Atomic notes
add_wiki_links: true
create_indices: true

# Integration
backup_before_integrate: true
auto_generate_backlinks: true
dry_run_mode: false  # Set true to test without changes

# Archival
compression: "gzip"
verify_after_archive: true
cleanup_originals: false

Troubleshooting

API Key Issues (If Running Standalone)

If you run the scripts outside Claude Code and get API key errors:

bash
# Claude Code integration (recommended)
# Just use /extract-transcript-facts - no setup needed

# Or manually set API key for standalone use
export ANTHROPIC_API_KEY='sk-ant-...'
python workflow_runner.py --transcript your_file.txt

Obsidian Vault Not Found

bash
# Create vault directory if needed
mkdir -p ~/Documents/Obsidian/Aaron

# Set custom vault path
/integrate-obsidian-vault ./notes/ --vault-path /your/vault/path

Permission Errors

bash
# Ensure write access to vault and archive directories
chmod -R 755 ~/Documents/Obsidian/Aaron
mkdir -p ~/youtube-transcriber/transcripts-archive
chmod -R 755 ~/youtube-transcriber

Files Included

code
youtube-transcriber/
├── youtube-transcript-extractor/        Phase 1 Skill
│   ├── src/
│   │   ├── extractor.py
│   │   ├── formatters.py
│   │   └── metadata.py
│   └── tests/ (18 tests)
│
├── transcript-to-logseq/                Phase 2 Skill
│   ├── src/
│   │   ├── logseq_formatter.py
│   │   ├── obsidian_compat.py
│   │   └── hierarchy_builder.py
│   └── tests/ (21 tests)
│
├── transcript-to-obsidian/              Phase 3 Skill
│   ├── src/
│   │   └── integrator.py
│   └── tests/ (6 tests)
│
├── transcript-archiver/                 Phase 4 Skill
│   ├── src/
│   │   └── archiver.py
│   └── tests/ (18 tests)
│
├── workflow_runner.py                   Full pipeline runner
├── integration_test.py                  Cross-skill tests (6 tests)
└── config.yaml                          Configuration

Future Roadmap

v1.1.0 (Ollama Integration)

  • • Swap Claude API → Ollama for local LLMs
  • • Support multiple local models (Mistral, Llama2, etc)
  • • Offline-first workflow

v1.2.0 (Advanced Features)

  • • Batch transcript processing
  • • Custom fact categories
  • • Multiple output formats
  • • Web UI for management

v1.3.0 (Integration)

  • • Direct YouTube URL support (fetch transcripts automatically)
  • • Streaming support for long transcripts
  • • Multi-language support

Support

  • •Documentation: See README.md in each skill directory
  • •Tests: Run python -m pytest tests/ in each directory
  • •Issues: Check integration_test.py for validation

License

Same as parent project (astoreyai)

Status: ✅ Ready for Claude Code integration

Works With:

  • •✅ Claude Code CLI (uses configured API key automatically)
  • •✅ Ollama (local models - future version)
  • •✅ Custom LLM backends (extensible architecture)

No API key management needed when used with Claude Code CLI