Anthropic gives away Agent Skills: the AI agent standard moves to the Linux Foundation

Anthropic is handing over control of Agent Skills, the standard that teaches AI agents repeatable tasks. The standard moves to the Agentic AI Foundation (AAIF) under the Linux Foundation, where the company earlier placed the Model Context Protocol (MCP). For organisations that package their own procedures as "skills", that work becomes less tied to a single vendor. Security rules for those packages, however, are still only a proposal.

What Anthropic has done

Anthropic announced on 22 October that it is donating Agent Skills to the AAIF. According to Anthropic, the transfer covers the name and logo, the GitHub organisation and the agentskills.io domain. Anthropic co-founded the AAIF with Block and OpenAI last year.

A skill, as Anthropic describes it, is a folder containing a SKILL.md file that teaches an agent a specific task or domain, along with any scripts or reference material the agent needs. The company developed the format while working on Claude Code, because the agent handled general knowledge work well but lacked the procedural knowledge real business tasks require.

Anthropic says more than 70 agent platforms have adopted Agent Skills since the standard was published in December 2025, including ChatGPT, Cursor, GitHub Copilot and VS Code. The company also says 1 million skills have been published and installed nearly 280 million times. These are Anthropic's own figures and have not been independently verified.

Who governs it now

Anthropic says the standard will be open, community-driven and vendor-neutral. Engineers from Anthropic, Google, OpenAI and other companies will maintain it as individuals, with David Soria Parra and Mahesh Murag as lead maintainers. Organisations can use the standard or contribute to it without joining the AAIF.

Anthropic has proposed starting with a roadmap covering versioned releases, a formal proposal process and security guidance for people who write and install skills. That is a proposal, not adopted rules. The announcement gives no timeline for versioning or security guidance.

What it means for business leaders

Less lock-in. If your organisation has built internal workflows as skills, such as report templates, control procedures or customer service answers, the same folders can in principle be used across tools that support the standard. With a neutral foundation owning the standard, investing in the format carries less risk than if a single vendor could change it on its own.

The supply chain is not solved. Skills can include scripts that the agent runs. With more than a million skills published, the risk resembles open source: some packages may be poor or malicious. The security guidance is only proposed so far. Until it exists, organisations should decide for themselves who may install skills from external sources and which ones are approved.

Versioning brings predictability. Versioned releases and a formal change process make it easier to know which version of the standard a tool supports. That matters in procurement: once numbered versions exist, ask vendors which one they follow.

What the announcement does not say

Anthropic does not explain how the AAIF will make decisions about the standard in practice, beyond saying the community will have a say and the process will be transparent. The announcement also says nothing about changed licence terms, costs or requirements for platforms already using the format.

Sources and media

• Anthropic: Donating Agent Skills to the Agentic AI Foundation, published 22 October 2026.
• Agent Skills specification
• Agent Skills on GitHub

Illustration: AI-generated with Grok Bot / hogby.ai