The Agentic AI Foundation’s announcement graphic for its four new projects, including Anthropic’s Agent Skills. Image: Agentic AI Foundation / The Linux Foundation
Anthropic is giving away Agent Skills, the format it built to teach Claude new tasks, which has since spread to ChatGPT, Cursor, GitHub Copilot and more than 70 other agent platforms. The company said on Thursday it is donating the open standard to the Agentic AI Foundation (AAIF), the Linux Foundation group it co-founded with Block and OpenAI last year, and the home Anthropic already gave the Model Context Protocol (MCP).
The foundation announced Agent Skills alongside three new projects from OpenAI, Hugging Face and Alibaba at its AGNTCon + MCPCon North America conference in San Jose, California, which runs on October 22 and 23. That takes the AAIF to 10 projects in its first year, up from three when it launched in December 2025.
What Agent Skills are, and who uses them #
A skill is a folder holding a SKILL.md file that teaches an agent one task or area of know-how, plus any scripts or reference material it needs. Agents load a skill only when a job needs it. That is why Anthropic tells heavy users to move long instructions into skills to make their Claude Code limits last, and skills sit alongside hooks, plugins and MCP servers as one of the main ways to extend Claude Code.
Anthropic says it came up with skills while working on Claude Code, which handled general knowledge work well but often lacked the procedural knowledge that real business tasks need. Skills launched in October 2025. Within weeks, developers had published thousands on GitHub and written s for OpenAI’s Codex, Cursor and OpenCode, and in December 2025 Anthropic published the format as an open standard with its own GitHub organisation and website, agentskills.io.
By Anthropic’s count, 1 million skills have now been published and nearly 280 million installed. Those figures come from the company and haven’t been independently checked.
What Anthropic is handing over #
The donation covers the Agent Skills name and logo, the GitHub organisation and the agentskills.io domain. Developers can keep using the standard, or contribute to it, without joining the AAIF.
Engineers from Anthropic, Google, OpenAI and other companies will maintain it as individuals, led by Anthropic’s David Soria Parra and Mahesh Murag. Anthropic has proposed a first roadmap with versioned releases, a formal process for proposing changes and security guidance for people who write and install skills. Feedback on changes to the specification goes through the project’s Discord.
It follows the path MCP took last December. MCP, the protocol that connects agents to outside tools and data, now underpins things like the gateway Walmart uses to let approved AI agents shop its store. Soria Parra, who co-created MCP, said in the foundation’s release:
We launched Agent Skills a year ago to give developers and enterprises alike an open standard to make their agents more capable. Today, we’re donating Agent Skills to the AAIF to ensure that anyone building with skills can shape what’s next.
David Soria Parra, co-creator and Lead Maintainer of MCP, Anthropic
Developer Simon Willison pointed out on his blog that he had called it: when Anthropic first declared skills an open standard last December, he wrote that he “wouldn’t be surprised to see this end up in the AAIF”.
‘No single company controlling the outcome’ #
Each of the other three new projects comes from a different company. Open Responses, from OpenAI, is an open specification based on OpenAI’s Responses API, so developers can write a request once and run it across model providers. OpenEnv, from Hugging Face, is a framework and hub for building and sharing the reinforcement learning environments used to train and test agents. OpenSandbox, from Alibaba, is a platform for running coding agents and AI-written code in isolated sandboxes, with Docker and Kubernetes runtimes.
They join MCP, Block’s goose agent, OpenAI’s AGENTS.md, the Agent2Agent (A2A) protocol, Agent Router and agentgateway. Manik Surtani, the AAIF’s CTO and co-founder, said:
Agent Skills, Open Responses, OpenEnv, and OpenSandbox each cover a different part of that stack, from how agents gain capabilities to where they run, and four different companies chose to contribute them to neutral governance.
Manik Surtani, CTO and co-founder, Agentic AI Foundation
The foundation’s job, he added, is giving the community one place to build shared standards “with no single company controlling the outcome”.
Why it matters #
A skill written once can now move between Claude, ChatGPT, Copilot, Cursor and dozens of other tools under rules no single company controls, the same deal MCP got last year. That makes building a library of skills a safer bet for developers and companies, and the promised security guidance matters too: installing a skill means letting an agent follow someone else’s instructions and run their scripts.