← All posts
This post is written in English.
On 22 October, Anthropic donated Agent Skills, its standard for giving agents specialised knowledge and workflows, to the Agentic AI Foundation. The foundation operates under the Linux Foundation, and Anthropic, Block and OpenAI co-founded it in 2025. Model Context Protocol went there earlier.
What changes
The donation includes the name, logo, GitHub organisation and domain. Engineers from Anthropic, Google, OpenAI and others will maintain it, serving as individuals and not as company representatives. The proposed roadmap covers versioned releases, a formal proposal process, and security guidance for skill authors and installers.
Anthropic reports more than 70 agent platforms have integrated it, including ChatGPT, Cursor, GitHub Copilot and VS Code, with 1 million published skills and nearly 280 million installs. Those adoption figures are Anthropic's own.
What to watch
The security guidance may matter most. A skill is instructions and sometimes code that an agent will follow, which makes it a supply chain risk much like a package. With a million published, someone has to think about trust and review.
My read
Companies donate standards when they want others to feel safe adopting them, and that's generally a good thing. The real test comes when members disagree, for example when a change suits one vendor's product and hurts another's. Neutral governance only proves itself then. For now, the practical advice is simple: read a skill before installing it, as you'd read a script before running it.
