What are Agent Skills (Claude Skills)?

Agent Skills, also known as Claude Skills, are folders containing instructions, scripts, and resources that AI agents load dynamically to improve performance on specialized tasks. They package expertise into discoverable capabilities that help AI assistants complete specific tasks more effectively.

How do Agent Skills work?

Skills are model-invoked—AI agents autonomously decide when to use them based on your request and the Skill's description. They use only ~100 tokens for scanning and less than 5,000 tokens when activated, making them efficient and powerful.

Do ChatGPT and Codex support skills?

Yes! As of December 2025, OpenAI adopted the skills format for ChatGPT and Codex CLI. Skills are now an open standard that works across Claude, ChatGPT, Codex, Cursor, and other MCP-enabled AI platforms.

How do I install an Agent Skill?

Skills can be installed in several ways depending on your platform. For Claude: Add to ~/.claude/skills/ for personal use, or .claude/skills/ for project sharing. For ChatGPT: Use the /home/oai/skills folder. For Codex CLI: Add to ~/.codex/skills/.

Where can I use Agent Skills?

Skills are portable—they use the same format everywhere. Build once, use across Claude, ChatGPT, Codex, Cursor, and any MCP-enabled platform. This open standard means skills work consistently across different AI assistants.

What's the difference between Skills and MCP servers?

Skills extend AI agent capabilities through instructions and scripts that agents load when needed for specific tasks. MCP (Model Context Protocol) servers are standalone protocol implementations that connect AI agents to external tools, services, and data sources. Skills teach agents how to do things, while MCP servers connect agents to things—they work best together.