Anthropic Donates Agent Skills to AAIF

Anthropic has handed Agent Skills, its open format for teaching AI agents specific tasks, to the Agentic AI Foundation (AAIF), the Linux Foundation body that already governs the Model Context Protocol. The company said in its announcement on Thursday that “more than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, have integrated Agent Skills since we published the standard in December 2025.”

Anthropic also says that “1 million skills have been published and nearly 280 million installed”. Those figures are the company’s own and have not been independently checked.

The donation was one of four new projects announced by the foundation at AGNTCon + MCPCon North America in San Jose. According to the AAIF release, OpenAI is contributing Open Responses, Hugging Face is contributing OpenEnv and Alibaba is contributing OpenSandbox. The additions take the foundation “to 10 hosted projects in its first year, up from its three launch projects in December 2025”.

What Anthropic is handing over

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

Anthropic is donating “the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain”. Engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals, with David Soria Parra and Mahesh Murag as Lead Maintainers. Developers can keep using or contributing to the standard without joining the AAIF, the company says.

Anthropic has proposed a starting roadmap covering versioned releases, a formal proposal process and security guidance for people who write and install skills. No timetable was given.

The foundation now spans four layers of the agent stack

The AAIF says the four projects cover how agents gain skills, how applications talk to models across providers, how agents are trained and evaluated, and where they run safely. Open Responses is an open specification based on the OpenAI Responses API, meant to let developers describe requests once and run them across model providers. OpenEnv is a framework and hub for reinforcement-learning environments. OpenSandbox is a sandbox platform for AI applications with Docker and Kubernetes runtimes.

They join A2A, Agent Router, agentgateway, AGENTS.md, goose and MCP. “Four different companies chose to contribute them to neutral governance,” said Manik Surtani, CTO and co-founder of the Agentic AI Foundation.

Alibaba’s Xiaobin Xu, Head of Development Infrastructure, said: “While OpenSandbox was created at Alibaba, we don’t want its future to be defined by a single company.”

Shared security guidance matters more than the logo

The most consequential item on the roadmap is the least glamorous one. A skill can carry scripts that an agent runs, and with nearly 280 million installs by Anthropic’s count, a poisoned or careless skill is a real supply-chain risk, much like a bad package in a software registry. Moving the format under neutral governance gives rival vendors a single place to agree on how skills should be reviewed, signed and installed, rather than each platform inventing its own checks.

The pairing with OpenSandbox points the same way. A common format for packaging agent know-how, alongside a common place to run untrusted agent code in isolation, gives developers two halves of a defence: control what agents load, and contain what they execute. TSN covered a similar push for contained agents in NVIDIA’s OpenShell blueprint.

Neutral homes are becoming the norm for agent standards

Anthropic followed the same path with MCP, which it also donated to the AAIF. The pattern suggests that leading labs now see more value in a widely adopted standard than in owning one outright, especially as competitors such as OpenAI contribute their own specifications to the same body. The test will be whether the formal proposal process lets smaller contributors shape changes, or whether the large companies that maintain the code continue to set the direction. Readers can follow related work on evaluating agents in Microsoft’s Scope platform and Anthropic’s latest agent product, Claude Duet.