The handover
Anthropic has transferred the Agent Skills standard to the Agentic AI Foundation (AAIF), the Linux Foundation arm dedicated to artificial intelligence agents. According to the source, the announcement was made on October 22, 2026, and along with the standard, the name, logo, GitHub organization, and the agentskills.io domain are also moving to the foundation. This is a change in governance, not in format: the technical structure remains unchanged.
A skill is a folder containing a SKILL.md file with a name, description, and instructions for performing a specific task, possibly accompanied by scripts and reference materials. At startup, the agent reads only names and descriptions, loads the full file when the task matches, and then executes any included code. The source distinguishes this mechanism from the Model Context Protocol (MCP), which connects agents to external tools and data: skills explain how to complete a specific job.
The stated figures and the precedents
The cited data are attributed to Anthropic: more than 70 platforms are said to have integrated the standard since December 2025, including ChatGPT, Cursor, GitHub Copilot, and VS Code; published skills are said to number one million, and installations nearly 280 million. The format originated within Claude Code, launched in October 2025, and was opened as a cross-platform standard two months later. These are figures stated by the company itself, not independently verified in the available material.
The transfer follows a path already taken: the source recalls that Anthropic had also handed MCP over to the same foundation. The AAIF, created in December 2025 by Anthropic, Block, and OpenAI, now also hosts OpenAI's Open Responses, goose, and AGENTS.md. Governance therefore becomes shared: engineers from Anthropic, Google, OpenAI, and other companies will maintain the standard in an individual capacity, with David Soria Parra and Mahesh Murag as leads. Anthropic no longer leads alone.
What remains open for developers and security
For developers, nothing changes in the immediate term, according to the source: the format remains identical, with a SKILL.md directory and only two mandatory fields, name and description. Developers can continue to use the standard or contribute without joining the AAIF; the code is under the Apache-2.0 license, and the documentation under CC-BY-4.0. The proposed roadmap includes releases with version numbering, a formal process for proposals, and security guidelines for writing and installing skills.
The security chapter is the most delicate: Snyk is said to have identified 76 malicious skills in public directories, and Anthropic is said to have called for shared rules. A neutral standard facilitates the exchange of skills among strangers, but as exchange grows, so does the risk of executing unverified instructions. With Anthropic, OpenAI, and Google at the same table, SKILL.md can position itself as the reference format for agent instructions, just as MCP is for connections: the test, according to the source, will come from the first numbered versions and from the security rules the foundation manages to impose. The actual adoption of the stated figures and the effectiveness of the announced guidelines remain to be verified.
