Instructions and know-how

Agent Skills

Portable folders of instructions, scripts and references that an agent loads only when a task needs them.

Browse Agent Skills
Status
Supported by 45+ agent products
Governance
Open standard at agentskills.io, originally developed by Anthropic.
Sources
Project siteSpecification

What to know

  • A skill is a folder with a SKILL.md file: YAML frontmatter with name (≤64 chars, lowercase, digits and hyphens, matching the folder) and description (≤1024 chars).
  • Optional fields: license, compatibility, metadata and the experimental allowed-tools.
  • Progressive disclosure: only name and description (~100 tokens) load at startup; the body loads on activation; scripts/, references/ and assets/ load on demand.
  • Keep SKILL.md under 500 lines and validate with skills-ref.
  • Supported by Claude, Claude Code, Codex, Gemini CLI, Cursor, GitHub Copilot, VS Code, OpenCode, goose, Junie, Kiro, Factory, OpenHands, Hermes Agent, OpenClaw and many more.

When to use it

To teach any agent a repeatable workflow or house rules once and reuse it across tools. Review third-party skills before installing: they can run scripts.

Related standards

MCP · Model Context ProtocolA2A · Agent2Agent ProtocolAGENTS.md · AGENTS.mdACP · Agent Client Protocol

Reviewed Oct 5, 2026. Spotted something out of date? Email hello@agentatlas.dev.