admin-valentin/awesome-agent-skills-new
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
The definitive resource for Agent Skills - modular capabilities revolutionizing AI agent architecture
15 commits 1 branch 0 tags 1.4 MiB
Find a file
xxl007 930be18d43 Remove emoji from social preview to fix white rectangle rendering issue
2025-12-22 22:42:57 +00:00
assets Remove emoji from social preview to fix white rectangle rendering issue 2025-12-22 22:42:57 +00:00
.gitignore Initial commit: Populate Awesome Agent Skills list 2025-12-22 18:17:21 +00:00
CONTRIBUTING.md Initial commit: Populate Awesome Agent Skills list 2025-12-22 18:17:21 +00:00
README.md Add comprehensive FAQ section with 15 SEO-optimized questions covering what, how, why, comparisons, security, and best practices 2025-12-22 22:06:51 +00:00

README.md

🎯 Awesome Agent Skills Awesome

🚀 The definitive resource for Agent Skills - the breakthrough format revolutionizing how AI agents acquire knowledge, execute workflows, and scale their capabilities.

💡 What Are Agent Skills?

Agent Skills mark a fundamental shift from monolithic AI systems to composable agent architectures. Rather than embedding capabilities through expensive fine-tuning or bloating context windows with static documentation, skills provide modular, on-demand knowledge injection through standardized SKILL.md packages.

The Architecture

Progressive Disclosure at Scale: Skills leverage a three-tier context management strategy:

  1. Discovery (~50 tokens): Lightweight metadata (name + description) loads at startup
  2. Activation (~2-5K tokens): Full SKILL.md instructions load when task-relevant
  3. Execution: Referenced scripts and resources accessed dynamically

This architecture delivers:

  • 🧠 Infinite capability scaling - No context window trade-offs
  • ⚡ Sub-second skill loading - Zero startup latency penalty
  • 🔄 Cross-platform portability - Write once, deploy to any compatible agent
  • 📦 Frictionless distribution - Share via Git, install like packages

The Revolution

From specialized agents to universal platforms: Instead of building separate coding agents, research agents, and analysis agents, the industry is converging on general-purpose agents with skill libraries. One agent. Unlimited specializations. Dynamic composition.

"We used to think agents in different domains will look very different. The agent underneath is actually more universal than we thought."
— Barry Zhang, Anthropic Research

The open standard advantage: Major platforms (OpenAI, GitHub, Microsoft, Cursor) have adopted the Agent Skills specification, creating network effects: every skill you create works across the entire ecosystem. This is the npm moment for AI agents.

🏷️ Key Topics

agent-skills · ai-agents · llm · claude · anthropic · skill-md · progressive-disclosure · context-management · ai-development · mcp · agent-architecture · agentic-ai · automation · productivity

👉 New to Agent Skills? Start here: agentskills.io

📚 What's Inside?

🎁 Ready-to-Use Skills

Don't build from scratch - stand on the shoulders of giants. Tap into battle-tested skill libraries and specialized capabilities that power production agent systems today.

📚 Skills Collections & Repositories

Multi-skill repositories with diverse capabilities:

🎯 Individual Skills by Category

💻 Development & Programming

🔒 Security & Operations

🔧 Integration & Automation

🧠 AI & Context Engineering

🛠️ Tools & Platforms

The ecosystem is here. Industry-leading platforms have already integrated Agent Skills - your next agent build starts with choosing the right foundation.

🎨 IDEs & Code Editors

  • OpenCode - 🎯 AI development tool with built-in Agent Skills support
  • Cursor - ⚡ AI-powered code editor with native skills integration (docs)
  • Amp - 🚀 Next-gen AI coding assistant
  • Claude Code - 🤖 Anthropic's coding tool with first-class skills support
  • OpenAI Codex - 💻 OpenAI's CLI agent with Agent Skills support

🤖 AI Agent Platforms

  • Letta - 🧠 Build stateful LLM agents with memory
  • Goose - 🦆 Open source AI agent framework
  • GitHub Copilot - 💙 GitHub's AI coding assistant with Agent Skills support (docs, announcement)
  • VS Code - 💻 Microsoft's editor with Agent Skills in Insiders build

🔧 Developer Tools & Libraries

📖 Learn & Explore

From zero to expert in hours, not weeks. Curated resources that take you from understanding the fundamentals to architecting production-grade agent systems.

🎓 Getting Started

🏗️ Building Skills

📰 Featured Articles

📰 Industry News & Analysis

🎬 Videos & Tutorials

See it in action. Watch developers build real agent systems from scratch - from 1-minute quickstarts to hour-long engineering deep dives.

🎓 Beginner-Friendly Tutorials

🔧 Technical Deep Dives

🆚 Comparisons & Analysis

🏢 Real-World Applications

🎤 Conference Talks

💡 Know a great video? We're actively looking for more video content! Please contribute if you find or create relevant tutorials.

🔬 Research Papers

Understand the foundations. Cutting-edge research exploring the architecture, security implications, and theoretical underpinnings of skills-based agent systems.

🛡️ Security & Safety

🌐 Interoperability & Standards

📊 Progressive Disclosure & Transparency

🧠 Skill Learning & Architecture

💡 Note: Agent Skills (SKILL.md format) is a recent innovation (2024-2025), so academic research is still emerging. Most papers focus on related concepts like skill learning, tool use, and agent architectures.

❓ Frequently Asked Questions

What are Agent Skills?

Agent Skills are modular, standardized SKILL.md packages that provide AI agents with on-demand capabilities. Instead of fine-tuning models or bloating context windows, skills enable progressive disclosure: lightweight metadata loads at startup (~50 tokens), full instructions activate when needed (~2-5K tokens), and resources load dynamically during execution.

How do Agent Skills differ from fine-tuning?

Fine-tuning modifies model weights permanently (expensive, inflexible), while Agent Skills provide runtime knowledge injection that's instantly updatable, shareable across platforms, and requires zero retraining. Update a skill once, and every agent using it benefits immediately.

What's the difference between Agent Skills and MCP (Model Context Protocol)?

Agent Skills focus on capabilities and workflows (how to do X), while MCP focuses on data access (connecting to APIs, databases). They're complementary: use MCP to connect to external data sources, use Skills to teach agents how to process that data. Many developers use both together.

How do I create my first Agent Skill?

  1. Create a SKILL.md file with YAML frontmatter (name, description)
  2. Write clear instructions in markdown (what, when, how)
  3. Add optional scripts/ folder for code references
  4. Place in .github/skills/ or .claude/skills/ directory
  5. Test with compatible platforms (Claude Code, Cursor, GitHub Copilot)

👉 Full guide: How to create custom skills

Which AI platforms support Agent Skills?

Major platforms with native support include: Claude (Claude.ai, Claude Code), OpenAI (Codex CLI), GitHub Copilot, Cursor, VS Code Insiders, OpenCode, Amp, Letta, and Goose. The ecosystem is rapidly expanding as more platforms adopt the open standard.

Can I use Agent Skills with ChatGPT or other LLMs?

While ChatGPT doesn't natively support the Agent Skills format, you can use universal loaders like openskills to integrate skills with any LLM platform. The skills specification is open and platform-agnostic by design.

Are Agent Skills secure?

Agent Skills introduce new security considerations. Since skills can execute code and access files, always review skills before installation from untrusted sources. Use tools like skillcheck to scan for vulnerabilities. Research shows potential for prompt injection attacks (arxiv.org/abs/2510.26328), so treat skills like code packages—audit before trusting.

How do I share Agent Skills with my team?

Option 1: Commit to your repository's .github/skills/ or .claude/skills/ directory—agents automatically discover them.
Option 2: Publish to GitHub as a standalone repo (e.g., my-org/data-analysis-skill)—others can reference or clone.
Option 3: Use skillport or agent-skills-mcp for cross-platform distribution.

What makes a good Agent Skill?

Best practices:

  • ✅ Single responsibility - One clear capability per skill
  • ✅ Progressive detail - Brief metadata, detailed instructions, extensive resources
  • ✅ Context-aware - Include when/why to use (not just how)
  • ✅ Testable - Provide example inputs/outputs
  • ✅ Discoverable - Descriptive names and rich descriptions

When should I use Agent Skills vs custom tools/functions?

Use Agent Skills when:

  • Capabilities need to work across multiple platforms
  • Instructions are complex multi-step workflows
  • You want version control and collaborative editing
  • Knowledge needs frequent updates

Use custom tools/functions when:

  • Platform-specific integration required (e.g., API-specific features)
  • Real-time data access needed
  • Computational logic can't be expressed as instructions

Can Agent Skills call other Agent Skills?

Yes! Skills can reference other skills in their instructions, enabling skill composition. This allows building complex capabilities from simple, reusable building blocks. Some platforms also support skill libraries where skills automatically discover and leverage related capabilities.

How do Agent Skills impact token usage?

Dramatically reduce it. Traditional approaches load all documentation upfront (10K-100K tokens), while skills load only what's needed (50 tokens metadata + 2-5K when activated). For agents managing 50+ capabilities, this means 90%+ token reduction during idle time.

Are Agent Skills just for coding tasks?

No—Agent Skills work for any domain: data analysis, content creation, project management, research, security operations, automation, and more. The format is domain-agnostic. Think of them as "packages for knowledge" rather than "packages for code."

Where can I find production-ready Agent Skills?

Start with:

How often should I update my Agent Skills?

Update skills whenever:

  • Platform APIs or tools change
  • Better approaches emerge
  • Team workflows evolve
  • User feedback reveals gaps

Pro tip: Add version numbers and changelogs to your skills for better tracking.

💬 Join the Community

You're not alone in this. Join thousands of developers pioneering the future of AI agents - share breakthroughs, debug challenges, and shape the standards.

🤝 Contributing

Shape the ecosystem. This is a community-driven resource - your contributions help thousands of developers discover and build better agent systems.

  • 🔗 Add new skills, tools, or articles
  • 📝 Improve documentation
  • 🐛 Fix broken links
  • 💡 Suggest new sections

Please read our contribution guidelines before submitting.

Found this useful? Give it a ⭐ and share it with others!

Built with ❤️ by the Agent Skills community