Name, GitHub and domain: How Anthropic is transferring Agent Skills to the Linux Foundation's AAIF

Name, GitHub and domain: How Anthropic is transferring Agent Skills to the Linux Foundation's AAIF

The company is handing over the name, the GitHub organization, and the agentskills.io domain to the Agentic AI Foundation, which it co-founded with Block and OpenAI. The adoption figures — more than 70 platforms, one million skills, and nearly 280 million installations — come, for now, only from Anthropic itself.

Anthropic announced on 22 October 2026 that it is donating the open Agent Skills standard to the Agentic AI Foundation (AAIF), which sits under the Linux Foundation. According to the company's own announcement, the donation covers the standard's name and logo, the GitHub organization, and the domain agentskills.io (Anthropic).

According to the company, this is the second time Anthropic has handed over a major agent standard to the same foundation: the Model Context Protocol (MCP) has already been donated there. AAIF was founded by Anthropic together with Block and OpenAI "last year" — that is, 2025, the same year Agent Skills was launched.

What is actually changing

Before the donation, Agent Skills was in practice a vendor-owned standard, originally launched in connection with Anthropic's own tool, Claude Code. After the transfer to AAIF, the name, logo, GitHub organization, and domain are owned by a foundation, and the standard is, according to the announcement, to be maintained by engineers from Anthropic, Google, OpenAI, and other companies — as individuals, not as representatives of their employers. David Soria Parra and Mahesh Murag are named as Lead Maintainers.

This is a move Anthropic has made before: MCP was previously donated to the same foundation. Now the pattern is repeated with the skills standard.

What a skill actually is

Anthropic describes Agent Skills as a mechanism for giving AI agents specialized knowledge and workflows. Technically, a skill is a folder containing a file named SKILL.md — which teaches the agent a specific task or domain — plus any scripts or reference material it needs. The format means the same skill can be read and interpreted by different agent platforms, without each vendor having to invent its own file format.

The ecosystem's growth — as Anthropic itself counts it

All the growth figures in the announcement are Anthropic's own and are not independently confirmed in the available material:

• The standard was published in December 2025. Since then, more than 70 agent platforms are said to have integrated Agent Skills — among them ChatGPT, Cursor, GitHub Copilot, and VS Code.
• One million skills are said to have been published, with nearly 280 million installations.
• After the launch in October 2025, developers reportedly published thousands of skills on GitHub within weeks and wrote loaders for competing tools such as Codex, Cursor, and OpenCode.

The figures come exclusively from the party with the greatest interest in presenting the standard as a de facto industry standard, and should be read with that in mind.

What is only proposed

Anthropic writes that it has proposed starting with a roadmap for the standard's further development within AAIF, with versioned releases, a formal proposal process, and security guidance for those who write and install skills. This is a proposal, not an adopted policy: there is no confirmation in the available material that AAIF or the Linux Foundation has endorsed either the donation or the roadmap.

The security component could prove the most significant. Now that installations are, according to Anthropic, approaching 280 million, and skills can contain executable scripts, questions about where skills come from and what they actually do are becoming increasingly practical.

What remains

Anthropic stresses that you can continue to use the standard and contribute to it without becoming a member of AAIF. The company's own announcement is, for now, the sole source for both the donation, the maintenance structure, and the usage figures. Open questions include:

• Confirmation from AAIF, the Linux Foundation, Block, and OpenAI.
• Independent verification of the 70-plus-platform figure, the number of published skills, and the installations.
• Whether the roadmap is adopted, and if so, how the security guidance is designed.

With both Agent Skills and MCP placed in the same foundation, Anthropic has, according to its own announcement, moved much of its agent infrastructure out of its own house. How much of the day-to-day governance actually ends up with people other than those who wrote the standard will be decisive for whether the transfer results in genuinely shared control.