Curated resources for building and using Agent Skills.

• What Are Agent Skills
• Start Here
• Phase 1: Learn the Fundamentals
• Phase 2: Use Existing Skills
• Phase 3: Build and Integrate
• Phase 4: Benchmarks and Research
• Frequently Asked Questions

Agent Skills are modular, standardized SKILL.md packages that give agents on-demand capabilities via progressive disclosure: lightweight metadata can load early, full instructions load only when relevant, and supporting resources are accessed when needed.

agent-skills · ai-agents · skill-md · progressive-disclosure · context-management.

If you are new to Agent Skills, start with these quick primers.

• What are skills - Guide: Beginner-friendly introduction to Agent Skills.
• Using skills in Claude - Guide: Quick start for enabling skills in Claude.

Concepts, comparisons, and explainers to build a solid mental model.

• Equipping agents for the real world with Agent Skills - Original announcement from Anthropic Engineering.
• Claude Skills vs MCP: Complete Guide - Comparison of Agent Skills and Model Context Protocol.
• The Great AI Agent Configuration Confusion - Overview of SKILL.md, AGENTS.md, and related standards.
• Using skills with Deep Agents - How frameworks apply the skills pattern.
• 5 Agent Skill design patterns every ADK developer should know - How to structure the logic inside Agent Skills.
• Lessons from Building Claude Code: How We Use Skills - Lessons learned about using skills at Anthropic.

• Claude's new Agent Skills - One-minute overview.
• Don't Build Agents, Build Skills Instead - Anthropic talk on skills as a scalable abstraction.
• Intro to Agent Skills - By Google's Anti-Gravity team on X.
• Claude Agent Skills Tutorial and Demo - Intro tutorial and hands-on demo.
• Claude Code Skills built me an AI Agent Team - Extended beginner guide.

• Agent Skills with Anthropic - Short course from DeepLearning.AI

Platforms that support skills today, plus ready-to-use skill catalogs.

• Claude Code - Platform: Anthropic's coding tool with skills support (docs).
• OpenAI Codex - Platform: OpenAI's CLI agent with skills support (docs).
• Gemini CLI - Gemini in the terminal with skills support (docs).
• Cursor - AI-powered editor with native skills integration (docs).
• VS Code - Editor with Agent Skills support (docs).
• GitHub Copilot - Coding assistant with Agent Skills support (docs).
• Mistral Vibe - CLI coding agent with Agent Skills support by Mistral (docs)
• Manus - Autonomous AI agent, now supporting Agent Skills (blog)
• OpenCode - AI development tool with built-in Agent Skills support (docs).
• Amp - AI coding assistant with Agent Skills support (docs).
• Goose - Open source agent framework with extensions support (docs).
• Letta - Stateful LLM agents with memory (docs).
• Roo Code - VS Code extension and cloud agents with skills integration (docs).

• Anthropic skills - Official skills catalog from Anthropic.
• OpenAI skills - Official skills catalog from OpenAI.
• Microsoft skills - Skills for AI coding agents working with Azure SDKs and Microsoft AI Foundry.
• Google Workspace skills - Agent Skills for using Google Workspace products.
• Google Agent skills - Agent Skills for Google products and technologies
• Vercel skills - Collection of skills reflecting best practices in web development from Vercel.
• Supabase skills - Agent Skills to help developers using AI agents with Supabase.
• Hugging Face skills - Community skills catalog with broad compatibility.
• OpenClaw skills - Skills for OpenClow agents (formerly Clawd and Moltbot).

• muratcankoylan/Agent-Skills-for-Context-Engineering - A collection of Agent Skills for context engineering (from Anthropic).
• Orchestra-Research/AI-research-SKILLs - A collection of AI/ML research and engineering skills.
• phuryn/pm-skills - A collection of PM Skills

• SkillsMP - Marketplace for discovering and sharing Agent Skills.
• agentskill.sh - Directory of 44k+ skills with two-layer security scanning and /learn installer.
• Skillstore - Curated marketplace for Agent Skills.
• SkillsDirectory - Directory of popular Agent Skills.
• skills.sh - A directory and leaderboard for Agent Skills.

Guides and tools for authoring, validating, and distributing skills.

• A Complete Guide to Building Skills for Claude - Guide: comprehensively covers how to build, test, and distribute skills.
• How to create custom skills - Article: Step-by-step instructions for authoring skills.
• Skills API Quickstart - Docs: API reference for implementing skills.
• How I Built Agent Skills for Claude Code - Tutorial: Practical walkthrough for building custom skills.
• Claude Agent Skills Tutorial - Video: End-to-end walkthrough with examples.

• LangChain Multi-Agent Skills - Docs: Implementing skills in LangChain.
• SkillCheck - Tool: Scanner for common risks in skill packages.
• OpenSkills - Tool: Universal loader for integrating skills with many agents.
• LangChain Deep Agents - Framework: Agent harness with a skills-oriented workflow.
• IntentKit - Framework: Intent-driven agent building.
• Agentica - Framework: TypeScript function-calling utilities for agents.

Evaluation frameworks and deeper technical reading.

• huggingface/upskill - Library: Use Upskill to Generate and evaluate agent skills for code agents.
• benchflow-ai/SkillsBench - Library: Use Benchflow for Measuring skills performance on real workflows.
• Practical Guide to Evaluating and Testing Agent Skills - Blog: A guide for testing agent skills

• Claude Agent Skills: A First Principles Deep Dive - Article: Detailed architecture analysis.
• I finally CRACKED Claude Agent Skills - Video: Comparison of Skills, MCP, and subagents.
• Claude Agent Skills - Video: One-hour deep dive into domain-specific usage.
• muratcankoylan/Agent-Skills-for-Context-Engineering - Repository: Context-engineering reference materials.
• jakedahn/pomodoro - Repository: System skill pattern reference.
• yzfly/Mind-Cloning-Engineering - Repository: Techniques for building reusable skill knowledge.

• SkillFlow:Benchmarking Lifelong Skill Discovery and Evolution for Autonomous Agents (2026) - A benchmark for evaluating autonomous agents' ability to discover, repair, and maintain skills over time
• Graph of Skills: Dependency-Aware Structural Retrieval for Massive Agent Skills (2026) - Enabling efficient inference in large skill libraries by constructing executable skill graphs
• SkillClaw: Let Skills Evolve Collectively with Agentic Evolver (2026) - Aggregating user interactions to autonomously improve reusable skills
• SkillNet: Create, Evaluate, and Connect AI Skills (2026) - An open infrastructure designed to create, evaluate, and organize AI skills at scale
• SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks (2026) - Benchmark consisting of 86 tasks across 11 domains
• Agent Skills Enable a New Class of Realistic and Trivially Simple Prompt Injections (2025) - Security analysis of skill-file prompt injection risks.
• A survey of agent interoperability protocols (2025) - Survey of MCP, Agent Cards, and related protocols.
• Reinforcement Learning for Self-Improving Agent with Skill Library (2024) - Maintaining and improving skill libraries.
• PolySkill: Learning Generalizable Skills Through Polymorphic Abstraction (2024) - Learning transferable skills via abstraction.

Agent Skills are modular SKILL.md packages that provide on-demand capabilities without loading all knowledge up front.

Fine-tuning changes model weights, while skills provide runtime knowledge and workflows that you can update instantly.

Agent Skills focus on workflows and capabilities, while MCP focuses on secure, structured data and tool access.

See the How to Build Skills section for a step-by-step authoring guide.

Support varies by platform, but major tools include Claude (Claude.ai and Claude Code), OpenAI Codex, GitHub Copilot, Cursor, VS Code, and more.

If a platform does not support the format natively, you can often integrate skills via a loader or by adapting the SKILL.md instructions into that platform's prompt workflow.

Treat skills like code: review them before using, avoid installing from untrusted sources, and prefer audited skill libraries.

The most common approach is to version skills in Git (in a shared repo) and let your supported tools discover them from a standard directory.

Please read our contribution guidelines before submitting changes.