paper-reading - SKILL.md Agent Skill

name: paper-reading description: "Paper reading and analysis framework using Moonshot AI (Kimi). Actions: download, analyze, read, summarize, implement papers from arXiv, SIGGRAPH, etc. Features: intelligent paper fetching, AI deep analysis, personalized reading guidance, knowledge internalization, code generation. Supports: PDF parsing, multi-dimensional analysis (summary, innovation, methodology, implementation), structured notes, knowledge graphs, code framework generation."

Paper Reading - Academic Paper Analysis Framework

Complete framework for precision reading, internalization, and implementation of academic papers using Moonshot AI (Kimi).

Prerequisites

Check if Python is installed:

python3 --version || python --version

If Python is not installed, install it based on user's OS:

macOS:

brew install python3

Ubuntu/Debian:

sudo apt update && sudo apt install python3

Windows:

winget install Python.Python.3.12

Install required dependencies:

pip install requests openai python-dotenv beautifulsoup4 lxml PyPDF2 markdown pyyaml tqdm aiohttp

How to Use This Skill

When user requests paper reading work (download, analyze, read, summarize, implement), follow this workflow:

Step 1: Analyze User Requirements

Extract key information from user request:

Paper source: arXiv ID, URL, or local PDF path
Action type: download, analyze, full (download + analyze + code generation)
Analysis type: comprehensive, summary, methodology, innovation, implementation
Reader profile: amateur or professional (affects explanation depth)

Step 2: Execute Paper Processing

Use paper_skill.py to process the paper:

python3 .claude/skills/paper-reading/scripts/paper_skill.py <arxiv_id_or_url> [--action <action>] [--type <type>]

Available actions:

download - Only download the paper
analyze - Analyze existing paper
full - Complete workflow (download + analyze + code generation)

Available analysis types:

comprehensive - Full analysis (default)
summary - Brief summary
methodology - Methodology details
innovation - Innovation points
implementation - Implementation guide

Step 3: Access Results

After processing, results are saved in:

Notes: data/papers/<paper_id>/notes/
Summaries: data/papers/<paper_id>/summaries/
Code: data/papers/<paper_id>/code/
Knowledge Graph: data/papers/<paper_id>/knowledge/

Search Reference

Supported Paper Sources

Source	Format	Example
arXiv	ID or URL	`2301.12345` or `https://arxiv.org/abs/2301.12345`
SIGGRAPH	URL	`https://www.siggraph.org/...`
Local PDF	File path	`papers/paper.pdf`

Analysis Types

Type	Description	Use Case
`comprehensive`	Full analysis with all aspects	Deep understanding
`summary`	Brief overview	Quick review
`methodology`	Technical details and methods	Implementation focus
`innovation`	Key innovations and contributions	Research insights
`implementation`	Code generation guide	Practical application

Example Workflow

User request: "Analyze paper 2301.12345 and generate implementation code"

AI should:

# Execute full workflow
python3 .claude/skills/paper-reading/scripts/paper_skill.py 2301.12345 --action full

Then: Review the generated files and provide summary:

Notes location
Summary location
Code directory
Key findings

Tips for Better Results

Use full action for complete analysis - Downloads, analyzes, and generates code
Specify analysis type - Choose appropriate type for your needs
Check generated files - Review notes, summaries, and code structure
Iterate on analysis - Run different analysis types for different perspectives
Review implementation guide - Check code directory for implementation details

Configuration

The skill uses environment variables or config.yaml for configuration:

Required:

MOONSHOT_API_KEY - Your Moonshot AI API key

Optional (in config.yaml):

moonshot.model - Model to use (moonshot-v1-8k, moonshot-v1-32k, moonshot-v1-128k)
moonshot.temperature - Temperature parameter (0.0-1.0)
paper_reading.paper_workspace_dir - Output directory for results

Common Workflows

Quick Summary

python3 .claude/skills/paper-reading/scripts/paper_skill.py 2301.12345 --action analyze --type summary

Full Analysis with Code

python3 .claude/skills/paper-reading/scripts/paper_skill.py 2301.12345 --action full

Download Only

python3 .claude/skills/paper-reading/scripts/paper_skill.py 2301.12345 --action download

Output Structure

data/papers/<paper_id>/
├── notes/              # Structured notes
├── summaries/          # Paper summaries
├── code/               # Generated code framework
│   ├── README.md
│   ├── main.py
│   ├── algorithm.py
│   └── implementation_guide.md
└── knowledge/          # Knowledge graph
    └── knowledge_graph.json