If you've written a SKILL.md, the standard it follows now belongs to a neutral foundation. On Oct 22, 2026, Anthropic donated Agent Skills to the Agentic AI Foundation (AAIF), the Linux Foundation project that already looks after MCP and AGENTS.md.
New to skills? Start with What are agent skills? How to write your first SKILL.md, then come back.
What happened
Anthropic launched skills in October 2025 and published the open standard in December 2025. A year later, it says:
- more than 70 platforms have integrated Agent Skills, including ChatGPT, Cursor, GitHub Copilot and VS Code
- 1 million skills have been published
- skills have been installed nearly 280 million times
Anthropic is handing AAIF the Agent Skills name and logo, the GitHub organization and agentskills.io. The lead maintainers are David Soria Parra and Mahesh Murag. Maintainers from Anthropic, Google, OpenAI and other companies take part as individuals, not as company representatives.
The Linux Foundation's release says the news came out at AGNTCon + MCPCon North America in San Jose (Oct 22–23). The same day, AAIF took in three more projects:
| Project | From | What it is |
|---|---|---|
| Agent Skills | Anthropic | The SKILL.md standard |
| Open Responses | OpenAI | An open spec for multi-provider model APIs, based on the OpenAI Responses API |
| OpenEnv | Hugging Face | A framework and hub for reinforcement-learning environments |
| OpenSandbox | Alibaba | A sandbox platform for running agents and AI code |
AAIF now hosts 10 projects, up from 3 when it launched in December 2025. The others the release lists include A2A, Agent Router, agentgateway, AGENTS.md, goose and MCP.
What changes for you
Nothing breaks. The format hasn't changed. A folder with a SKILL.md that has name and description front matter is still a skill, and the tools that load skills today keep loading them.
The standard gets a process. Anthropic lists a proposed roadmap: versioned releases, a formal proposal process for changes, and security guidance. Versioned releases matter most. Right now you can't easily say "this skill targets version X of the spec." Once you can, tools can warn you about fields they don't support.
You don't need a membership. Anthropic says anyone can use the standard or contribute without joining AAIF.
What to do this week
- Validate what you have. Run your skills through the reference validator from the spec repo, or follow our starter skills pack, which we test this way.
- Keep skills portable. Stick to the fields in the spec. Tool-specific extras are fine, but put them where another tool can safely ignore them.
- Treat third-party skills like code. A skill can ship scripts, and your agent may run them. Read them before you install. Our cross-agent skills pack includes a security checklist and a scanner for exactly this.
- Watch the roadmap. The security guidance is the piece most likely to change how marketplaces and tools treat skills.
Why it matters
MCP, AGENTS.md and now Agent Skills all sit under one neutral home. For people building with AI, that means the three files you're most likely to commit next to your code (an MCP config, an AGENTS.md and a skills folder) are now governed in the same place, by the same rules. See What is AGENTS.md? and What is MCP? for the other two.
