Anthropic donates Agent Skills standard to the Agentic AI Foundation

Anthropic announced on 22 October that it would donate Agent Skills to the Agentic AI Foundation (AAIF), which operates under the Linux Foundation. The donation covers the standard’s name and logo, its GitHub organization and the agentskills.io domain. The company said the goal is for the format to develop in a community-driven and vendor-neutral home. Anthropic, Block and OpenAI co-founded AAIF last year; Anthropic had previously donated the Model Context Protocol to the foundation.

Agent Skills is a portable folder format for teaching an AI agent how to perform a specific task. Its core file is SKILL.md, which contains instructions and at least name and description metadata. A skill may also bundle scripts, reference materials, templates and other resources. In the progressive-disclosure model described in the documentation, an agent first discovers a skill from its name and description, loads the full instructions when a task matches, and then uses additional files or code if needed.

The design aims to make task-specific procedural knowledge available on demand rather than supplying every instruction before every task. A team might package an analysis or document workflow in a skill folder, and compatible agents can read the same structure. A shared folder format does not mean every agent executes every skill in the same way or safely. Permissions that a host product gives to files, networks and scripts continue to shape both the experience and the risks.

Anthropic says more than 70 agent platforms had integrated Agent Skills since its publication in December 2025, and that one million skills had been published and nearly 280 million installed. These are company-reported figures: the announcement does not provide a cross-platform counting method, unique-user count or active-use frequency. An installation may describe a download or copy; it does not show that a person uses a skill regularly or rule out the same person installing it in multiple products.

Anthropic said engineers from Google, OpenAI and other companies would maintain the standard as individual contributors, with David Soria Parra and Mahesh Murag as lead maintainers. The company proposed an initial roadmap covering versioned releases, a formal proposal process and security guidance for people who write and install skills. This roadmap was described as a proposal for community discussion, not as governance policy already in force.

The effect of moving to AAIF will depend not only on an open file format but also on how changes are approved, security issues handled and compatibility across agents verified. Anthropic says people can continue using and contributing to the standard without joining the foundation. That keeps participation open, but does not by itself prove neutrality in day-to-day implementation. Governance, compatibility and version decisions will show over time how the new stewardship works.