Agent Skills Moves to the Agentic AI Foundation
Anthropic donated Agent Skills to the Agentic AI Foundation on 22 October. What transferred, what changes for skill authors, and what stays the same.
Anthropic has donated Agent Skills, the open format behind SKILL.md files, to the Agentic AI Foundation. The announcement went up on 22 October 2026. The Agent Skills name and logo, the GitHub organisation, and the agentskills.io domain all transfer to the foundation, which sits under the Linux Foundation. Nothing about how you write or install a skill changes today. What changes is who decides where the format goes next, and the answer is no longer a single vendor.
If you have never written one, a skill is a folder with a SKILL.md file in it that teaches an agent a specific procedure. Our explainer on agent skills covers the format itself. This post is about the handover.
What actually transferred
The donation is of the standard and its assets, not of any product. Per the announcement, the foundation receives:
The Agent Skills name and logo.
The GitHub organisation that holds the specification and reference tooling.
The agentskills.io domain, where the spec is published.
Maintainership moves with it. Engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals rather than as company representatives, with David Soria Parra and Mahesh Murag named as Lead Maintainers. Anthropic's stated commitment is to keep the format "open, community-driven, and vendor-neutral".
The second standard to go the same way
This is the foundation's second Anthropic donation. The Model Context Protocol went first, and the MCP explainer describes what that protocol does. Anthropic co-founded the Agentic AI Foundation with Block and OpenAI, so the pattern is now established: a vendor publishes a format, adoption outruns the vendor, and the format is handed to a neutral body before competitors have to choose between adopting a rival's standard and fragmenting the ecosystem.
That is a different kind of body from the frontier-safety standards effort covered in our piece on the AI industry standards body. One governs interoperability formats. The other concerns pre-deployment testing commitments. They are easy to confuse because the same three company names appear in both.
The adoption numbers
Anthropic published figures alongside the announcement, and they explain the timing better than any statement about neutrality does.
Measure | Figure |
|---|---|
Agent platforms with Agent Skills integrated | More than 70, including ChatGPT, Cursor, GitHub Copilot and VS Code |
Skills published by developers | 1 million |
Skill installs | Nearly 280 million |
Standard first published | December 2025 |
A format that more than 70 platforms have implemented is no longer something one company can quietly revise. Anthropic says skills came out of work on Claude Code, where the tool handled general knowledge work well but lacked the procedural knowledge that specific business tasks need.
What changes for you, and when
Short term, nothing. Skills you have written keep working, the spec at agentskills.io is unchanged by the transfer, and you do not need to join the foundation to use the format or to contribute to it.
The proposed roadmap is where the practical changes sit:
Versioned releases. Today the spec is a living document. Versioning gives you something to pin against, which matters once a skill has to work on several clients at different release levels.
A formal proposal process. Changes would arrive through a documented route rather than a commit, so breaking changes become visible before they ship.
Security guidance for writing and installing skills. This is the gap worth watching. A skill is instructions plus optional executable scripts, and installing one from a stranger is closer to installing a package than to reading a document.
That last point already has a precedent. The same questions people now ask about untrusted skills were asked about untrusted MCP servers first, and our checklist for vetting an MCP server before you connect it transfers almost unchanged: read what it actually does, check what it can reach, and assume anything it loads can influence the agent.
Feedback on proposed specification changes runs through the project's Discord, with the spec itself at agentskills.io. For keeping track of moves like this one without reading every announcement, see our guide to following AI news without drowning in it.
Frequently asked questions
Do I need to change my existing skills?
No. The transfer covers ownership of the standard and its assets. The format itself is unchanged, and no migration is required.
What is the Agentic AI Foundation?
A body under the Linux Foundation, co-founded by Anthropic with Block and OpenAI, that acts as a neutral home for agent-related standards. It already holds the Model Context Protocol.
Do I have to join the foundation to contribute?
No. Anthropic states you can keep using the standard or contribute to it without joining, and feedback on spec changes is handled through the project's public channels.
Does this mean Anthropic stops working on Agent Skills?
No. Anthropic engineers remain among the maintainers, but they do so as individuals alongside engineers from Google, OpenAI and other companies, rather than as the owning vendor.
How did this land?
About the author
Senior Editor, AI & Product
Cecilia leads the Swarmz editorial desk. She has spent a decade turning complex AI and product topics into writing people actually finish, and she owns the blog's quality bar.
