# Agent Skills

> An open standard from Anthropic that packages procedural knowledge as a folder with a SKILL.md file an agent discovers and loads on demand.

_The Agentic Web Lexicon · /glossary/agent-skills · [JSON](/api/glossary/agent-skills) · [all The Agentic Web Lexicon](/glossary)_

- **term:** Agent Skills
- **category:** knowledge-memory
- **short_def:** An open standard from Anthropic that packages procedural knowledge as a folder with a SKILL.md file an agent discovers and loads on demand.
- **long_def:** A Skill is a directory containing a SKILL.md (metadata plus instructions), optional scripts and resources, which an agent loads dynamically only when relevant — keeping the context window lean. Anthropic unveiled Agent Skills on 16 October 2025 and released it as an open standard at agentskills.io on 18 December 2025; Microsoft, OpenAI, GitHub, Figma and Cursor adopted it.
- **see_also:** context-engineering, mcp, tool-use
- **etymology_origin:** Unveiled by Anthropic on 16 October 2025 and released as an open standard at agentskills.io on 18 December 2025; each Skill is a folder with a SKILL.md file of metadata and instructions.
- **related_to:** context-engineering, mcp, tool-use, prompt-caching, agentic-rag
- **contrast_with:** Unlike MCP, which connects an agent to external tools and data at runtime, Agent Skills package reusable procedural knowledge (how to do a task) that the agent loads into context on demand — know-how, not a tool connection.
- **example:** Anthropic released Agent Skills as an open standard on 18 December 2025, with Microsoft, OpenAI, GitHub, Figma and Cursor adopting it.
- **source:** https://claude.com/blog/skills
- **status:** active
- **why_it_matters:** Agent Skills let a service ship its own how-to knowledge to any compatible agent as a portable folder, so agents can perform domain tasks correctly without bespoke prompting.
- **sameAs:** —
- **bridge_entity:** agent-readiness
- **last_verified:** 2026-06-15
- **md_twin:** /glossary/agent-skills.md