Anthropic announced on October 22 that it will transfer Agent Skills to the Agentic AI Foundation (AAIF), an organization under the Linux Foundation created to host open technologies for the agent ecosystem. The donation includes the Agent Skills name and logo, the project’s GitHub organization, and the agentskills.io domain. Its stated purpose is to give the specification a neutral home and prevent its evolution from depending exclusively on one vendor. [1]

Agent Skills defines a simple way to add specialized knowledge and reusable procedures to an agent. Each “skill” starts with a folder containing a `SKILL.md` file, which includes at least the task’s name, description, and instructions. The folder can also carry scripts, reference documents, templates, and other resources. Project documentation describes a progressive-loading model: an agent first discovers minimal metadata and brings the complete instructions into context only when a task calls for that capability. [2][3]

The transfer matters because the format is no longer limited to Claude. Anthropic says more than 70 platforms—including ChatGPT, Cursor, GitHub Copilot, and Visual Studio Code—have integrated it since the company published the standard in December 2025. Anthropic also reports that one million skills have been published and nearly 280 million installations recorded. Those figures come from Anthropic and are not independently audited in the materials reviewed for this article. [1]

The AAIF already houses the Model Context Protocol (MCP), another Anthropic-originated standard donated to the foundation after Anthropic formed it with Block and OpenAI. Adding Agent Skills puts two distinct layers of agent architecture under the same organizational structure: MCP supports connections to tools and data sources, while Skills packages instructions, procedures, and resources for completing specific work. They are complementary rather than equivalent, and neither component by itself guarantees that an agent will perform a task correctly or safely.

According to the announcement, engineers from Anthropic, Google, OpenAI, and other companies will maintain the specification as individuals. David Soria Parra and Mahesh Murag will serve as lead maintainers. Developers can continue using and contributing to the format without joining the AAIF. The public repository uses the Apache 2.0 license for code and CC BY 4.0 for documentation, allowing the standard to be inspected, implemented, and modified under those terms. [1][2]

The public specification keeps the core structure deliberately small. In addition to its main instruction file, a skill can carry executable code and supporting context. That portability makes it easier to reuse a workflow across compatible products, but it also expands the trust boundary: a folder may contain instructions that change agent behavior and scripts capable of taking actions. Moving the project to a foundation addresses governance; it does not replace content review, execution isolation, least-privilege permissions, or controls over tools and data.

The move strengthens the possibility that Agent Skills can become an interoperable layer instead of fragmenting into incompatible vendor formats. It also formally distributes influence over the roadmap. The meaningful tests will be practical: transparent governance rules, verifiable compatibility across implementations, and security processes for third-party skills. For now, what is confirmed is the announced transfer of the standard’s assets and the continued availability of its open repository. Future adoption and the degree of genuine neutrality will depend on how the AAIF operates and who participates in its decisions.

Sources

[1] Read More — Donating Agent Skills to the Agentic AI Foundation — Anthropic
[2] Read More — Agent Skills specification and documentation — GitHub
[3] Read More — Agent Skills Overview — Agent Skills