Pdf2skills

pdf2skills

Convert PDF books into Skills callable to Claude Code.

Please see recent updates: https://github.com/kitchen-engineer42/Anything2Workspace

Quick Start

1. Install Dependencies

pip install requests python-dotenv python-Levenshtein numpy scikit-learn spacy jieba PyPDF2

# Download spaCy English model
python -m spacy download en_core_web_sm

2. Setup MinerU Client

Copy `mineru_client.py` to the parent directory of `pdf2skills/`:

your_project/
├── mineru_client.py      # MinerU API client
├── pdf2skills/           # This folder
│   ├── run_pipeline.py
│   └── ...
└── test_data/            # Your PDFs here

Get your MinerU API key at: https://mineru.net/

3. Configure Environment

Create a `.env` file in the `pdf2skills` directory:

# SiliconFlow API (Required - primary LLM provider)
SILICONFLOW_API_KEY=your_siliconflow_api_key
SILICONFLOW_BASE_URL=https://api.siliconflow.cn/v1

# MinerU API (Required - PDF to Markdown conversion)
MINERU_API_KEY=your_mineru_api_key
MINERU_BASE_URL=https://mineru.net/api/v4/extract/task

# Pipeline Configuration
CHUNK_MAX_TOKENS=30000
CHUNK_MAX_ITERATIONS=3
OUTPUT_LANGUAGE=English

# Rate Limiting (adjust based on your API tier)
GLM_RATE_LIMIT_SECONDS=3.0
FUSION_RATE_LIMIT_SECONDS=2.0
BUCKET_THRESHOLD=0.5

4. Run the Pipeline

# Basic usage
python run_pipeline.py your_book.pdf

# With custom output directory
python run_pipeline.py your_book.pdf --output-dir ./output

# For English PDFs
python run_pipeline.py your_book.pdf --language en

# Resume interrupted processing
python run_pipeline.py your_book.pdf --resume

5. Output Structure

After processing, you'll find:

your_book_output/
├── full.md                          # Extracted markdown
├── full_chunks/                     # Chunked documents
│   ├── chunks_index.json
│   └── chunks/
├── full_chunks_density/             # Semantic analysis
│   ├── density_scores.json
│   └── heatmap.html