Agent Skills leaves Anthropic for the Agentic AI Foundation

Felix Laurent 22 Oct 2026 3 min read

ai agents

Anthropic is donating Agent Skills to the Agentic AI Foundation (AAIF), the Linux Foundation fund that already hosts MCP. Agent Skills became an open standard in December 2025, less than a year ago. More than 70 agent platforms support it, including ChatGPT, Cursor, GitHub Copilot and VS Code. The move turns a format born inside Claude Code into shared industry plumbing.

What Anthropic announced

Anthropic published the news on October 22, 2026. It came during AGNTCon + MCPCon North America in San Jose. The company is handing the AAIF the Agent Skills name and logo, the GitHub organization and the agentskills.io domain.

A skill is a folder with a SKILL.md file inside. That file teaches an agent one task or one domain. A skill can also carry scripts and reference files that the agent loads only when it needs them.

Anthropic says 1 million skills have been published so far. It puts installs at nearly 280 million. Those are the company's figures, repeated in the AAIF press release.

What changes for skill authors and users

The practical answer is very little. Anthropic and the AAIF list four points that matter to anyone who writes or installs skills:

Existing skills keep working, and the licenses do not change: Apache 2.0 for code, CC BY 4.0 for docs
Anyone can use the standard or contribute without joining the AAIF
A technical steering committee will govern the spec, with engineers from Anthropic, Earendil (Pi), Google, Microsoft, OpenAI and OpenCode
The maintainers serve as individuals, and the lead maintainers are David Soria Parra and Mahesh Murag

The spec stays at agentskills.io, and the discussion stays on GitHub. The change is about who decides, not about how a skill is built.

Why Anthropic is giving it away

Agent Skills began as an internal fix. Anthropic says Claude Code handled general knowledge work well. It lacked the procedural know-how of real business tasks, and skills filled that gap.

The format spread fast after its October 2025 launch. Developers published thousands of skills on GitHub within weeks. They also wrote loaders for Codex, Cursor and OpenCode. In December 2025 Anthropic published the format as an open standard under its own GitHub organization.

That ownership was the weak point. A rival lab has less reason to build on a spec that one competitor controls. A neutral home removes that objection. MCP followed the same path in December 2025.

There is also a cost argument. A format with 70+ integrations needs versioning, review and security work. A foundation spreads that work across the companies that depend on it.

The three projects that joined with it

Agent Skills was one of four projects the AAIF added on October 22. Each one covers a different layer of the agent stack.

Project	Contributed by	What it does
Agent Skills	Anthropic	Packages instructions, scripts and resources that agents load on demand
Open Responses	OpenAI	A shared API schema for talking to models across providers
OpenEnv	Hugging Face	A framework and hub for reinforcement learning environments
OpenSandbox	Alibaba	A sandbox platform for running agent code on Docker and Kubernetes

With these four, the AAIF hosts 10 projects one year after launch. It started in December 2025 with three: MCP, goose and AGENTS.md. Competitors Anthropic and OpenAI now both route their agent formats through the same body.

What is still missing

The committee has not shipped anything yet. Its first task is a roadmap, and Anthropic lists three priorities for it:

Versioned releases of the spec
A formal proposal process for changes
Security guidance for people who write and install skills

The third point matters most. A skill can bundle scripts, and an agent may run them on the user's machine. With nearly 280 million installs, the standard still has no shared guidance on how to vet a skill. Our Claude skills guide reads the code of 20 popular skill repos for exactly this reason.

The other open question is control. David Soria Parra, one of the two lead maintainers, also leads MCP at Anthropic. The AAIF model only works if outside maintainers can change the spec's direction. The first versioned release will show whether they can.