Write your Skill once, run everywhere - from Claude Code to Agent SDK, OpenClaw, and beyond.

Skill Universal demonstrates how declarative Skills (Claude Code Skills format) can be loaded and executed across different Agent platforms.

Traditional agent development: N tools = N! possible workflow combinations

When you have N tools, creating agents for every possible workflow becomes unmanageable.

Skills provide intelligent aggregation: N tools + 1 router = all possible combinations

Skills act as:

• Tool Aggregators - Combine existing capabilities
• Decision Engines - Replace human judgment with AI decisions
• Execution Coordinators - Optimize workflows dynamically

Just create a .claude/skills/my-skill/SKILL.md file!

---
name: research
description: Research expert
triggers:
  - "research"
  - "study"
tools:
  - WebSearch
  - WebFetch
---

## Research any topic efficiently...

### 1. Search
Search for "[topic] tutorial 2026"

### 2. Fetch Documentation
Read official documentation

### 3. Summarize
Create comprehensive summary

from skill_universal import SkillLoader

loader = SkillLoader()
skill = loader.load("skills/my-skill/SKILL.md")
agent = skill.to_agent_sdk()

# Use with Anthropic Agent SDK
# agent.create() returns properly configured agent

import { SkillLoader } from 'skill-universal';

const loader = new SkillLoader();
const skill = await loader.load('skills/my-skill/SKILL.md');
const openclawSkill = skill.to_openclaw();

// Deploy to OpenClaw platform
// openclawSkill contains proper tool mappings

• Python (loaders/python/skill_loader.py) - Load Skills in Python/Agent SDK
• TypeScript (loaders/typescript/SkillLoader.ts) - Load Skills in TypeScript/Node.js/OpenClaw

pip install skill-universal

npm install skill-universal

git clone https://github.com/claude-world/skill-universal.git
cd skill-universal

# Python
pip install -e .

# TypeScript
npm install
npm run build

Automate research across web, GitHub, and community discussions.

Process PDF, Word, and Markdown documents with consistent interface.

Analyze GitHub repositories for tech stack, dependencies, and metrics.

Aggregate knowledge from multiple sources for intelligent support.

skill-universal/
├── skills/           # Example Skills (SKILL.md format)
│   ├── tech-research/
│   ├── document-processor/
│   ├── github-analyzer/
│   └── customer-support/
├── loaders/          # Platform-specific loaders
│   ├── python/       # Python/Agent SDK loader
│   │   ├── skill_loader.py
│   │   └── tests/
│   └── typescript/   # TypeScript loader
│       ├── SkillLoader.ts
│       └── tests/
├── examples/         # Usage examples
│   ├── claude-code/
│   ├── agent-sdk/
│   └── openclaw/
├── tests/            # Integration tests
├── docs/             # Documentation (EN/ZH-TW/JA)
├── .github/          # GitHub Actions, issue templates
├── pyproject.toml    # Python packaging
├── package.json      # TypeScript packaging
└── README.md

# Python tests
pytest tests/

# TypeScript tests
npm test

# All tests
npm run test:all

• English
• 繁體中文
• 日本語

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

• Core skill loading (Python, TypeScript)
• Example skills
• Claude Code format support
• Agent SDK format support
• OpenClaw format support
• PyPI package publishing
• npm package publishing
• More platform support (LangChain, LlamaIndex, etc.)
• Skill marketplace
• Web UI for skill creation

MIT License - see LICENSE

• Claude Code - Skills format inspiration
• Anthropic Agent SDK - Python/TypeScript SDKs
• OpenClaw - Open-source agent platform
• Claude World - Community and documentation

• Documentation
• Examples
• Issues
• Discussions

Made with ❤️ by the Claude World Community