Originally developed by Anthropic for Claude, Agent Skills are now an open standard adopted by multiple AI tools including GitHub Copilot.
Like MCP (Model Context Protocol), it’s a way to extend LLM capabilities by providing specialized instructions and additional resources.
A skill is simply a folder containing a SKILL.md file with:
The agent loads these skills dynamically and on-demand based on the conversation context.
See an example: https://agentskills.io/what-are-skills
Unlike custom instructions or MCP tools that are always injected into the context, skills use a 3-level progressive loading system:
Result: You can have dozens of skills installed without wasting tokens. Only what’s relevant to the current task gets loaded into context.
Skills are placed in specific folders depending on your AI assistant:
| Assistant | Project Location | Global Location |
|---|---|---|
| Claude Code | .claude/skills/ |
- |
| GitHub Copilot | .github/skills/ |
~/.github/skills/ |
| Antigravity | .agent/skills/ |
~/.gemini/antigravity/global_skills/ |
| Codex | .codex/skills/ |
~/.codex/skills/ |
Enable in VS Code: chat.useAgentSkills (currently in preview)
Many websites offer ready-to-use skills in the standard format:
You can create custom skills tailored to your team’s workflows!
Basically, anything you find yourself explaining repeatedly to the team can become a skill!