Read Paper Skill for Claude Code

A Claude Code skill that reads arXiv papers and generates structured reading notes in both Chinese and English, with convincing figures extracted from the paper, then automatically pushes to GitHub.

Features

• Accepts arXiv ID or URL in multiple formats
• Automatically downloads paper PDF and saves to repo
• Extracts ALL figures first before reading paper
• Reviews figures to select the most convincing ones
• Generates structured notes in both Chinese and English
• Embeds relevant figures in notes to improve readability
• Auto-saves to GitHub with organized folder structure

Workflow

1. Download PDF from arXiv
2. Extract all figures from PDF (using PyMuPDF)
3. Review extracted figures to understand their content
4. Read the paper
5. Select convincing figures for each section of notes
6. Generate bilingual notes with embedded figures
7. Push to GitHub

Installation

# Clone this repository
git clone https://github.com/0xPabloxx/Paper-Skill.git

# Copy skill to Claude Code skills directory
mkdir -p ~/.claude/skills/read-paper
cp Paper-Skill/SKILL.md ~/.claude/skills/read-paper/
cp Paper-Skill/README.md ~/.claude/skills/read-paper/

Recommended: Install PyMuPDF for figure extraction

pip install pymupdf

Alternative: Install poppler-utils

# Ubuntu/Debian
sudo apt install poppler-utils

# macOS
brew install poppler

Usage

In Claude Code, use the /read-paper command:

/read-paper 2301.07041

Or with full URL:

/read-paper https://arxiv.org/abs/2301.07041

Output Structure

After running the skill, files are saved to:

Paper-Skill/
├── paper-2023-02-11-verifiable-fhe/
│   ├── paper.pdf          # Original paper
│   ├── notes_zh.md        # Chinese notes (with figures)
│   ├── notes_en.md        # English notes (with figures)
│   └── figures/           # ALL extracted figures
│       ├── fig-000.png    # Reviewed and categorized
│       ├── fig-001.png
│       └── ...
├── paper-2024-01-15-another-paper/
│   └── ...
├── README.md
└── SKILL.md

Note Template

Each note includes:

• Quick View: Title, authors, arXiv link, year
• Question: Research question addressed
• Task: Specific task being solved
• Challenge: Technical challenges of previous methods
• Insight: Core high-level idea + overview figure
• Contribution: Technical contributions + method figures
• Experiments: Ablation studies + result figures/tables
• Limitation: Failure cases + statistics figures

Figure Selection

Figures are selected based on their value:

Category	Where Used	Priority
Architecture/Overview	Insight	High
Method diagrams	Contribution	High
Results charts/tables	Experiments	High
Ablation figures	Experiments	High
Comparison charts	Experiments	Medium
Statistics	Limitation	Medium
Small icons (< 5KB)	Skipped	-

Requirements

• Claude Code CLI
• Internet connection (for arXiv access)
• Git configured with push access to your fork
• PyMuPDF (pip install pymupdf) or poppler-utils for figure extraction

License

MIT