Anthropic donates Agent Skills to the Agentic AI Foundation: what changes and what does not

Anthropic has donated the Agent Skills standard to the Agentic AI Foundation (AAIF), a project of the Linux Foundation, in an announcement dated October 22, 2026. The gift covers the Agent Skills name and logo, its GitHub organization and the agentskills.io domain. Anthropic says more than 70 agent platforms already support the format.

For people who build with agents, the headline is simple: a format Anthropic invented is now held by a neutral foundation. The details show what that does and does not change.

What was donated

Anthropic co-founded AAIF with Block and OpenAI last year, and it had already donated the Model Context Protocol (MCP) to the same foundation. Agent Skills now follows. The Linux Foundation's release describes Agent Skills as an open format that packages instructions, scripts and resources so agents can load specialized capabilities on demand and reuse them across platforms.

A skill is a folder with a SKILL.md file holding a name, a description and instructions, plus optional scripts, references and templates. Agents load skills in three stages: discovery of names and descriptions at startup, activation of the full instructions when a task matches, and execution. The agentskills.io documentation calls this progressive disclosure, and it means an agent can keep many skills on hand with a small context footprint.

The numbers, and who reports them

Anthropic's post gives a timeline and adoption figures.

• October 2025: Skills launched with Claude Code.
• December 2025: Agent Skills was published as an open standard.
• October 22, 2026: the donation to AAIF.
• More than 70 agent platforms have integrated it, including ChatGPT, Cursor, GitHub Copilot and VS Code.
• About 1 million skills published and nearly 280 million installed.

Every one of those figures is Anthropic's own. We found no independent count of skills or installs, so treat the scale as reported, not verified.

Who runs it now

Anthropic says engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals, not as company delegates. The Lead Maintainers are David Soria Parra and Mahesh Murag. Soria Parra is also a co-creator and lead maintainer of MCP, so the same person is central to both projects.

The announced roadmap is modest: versioned releases, a formal proposal process, and security guidance for people who write and install skills. Anyone can use or contribute to the standard without joining AAIF, according to Anthropic.

That individual-maintainer structure is worth noting. It spreads influence across companies on paper, but a small group of maintainers who work at the largest labs still shapes the spec. That is how many open-source projects work, and it is not a criticism, only a fact to keep in view.

The same-day AAIF announcement

The Linux Foundation announced four new AAIF projects at AGNTCon and MCPCon North America in San Jose, October 22 to 23: Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face and OpenSandbox from Alibaba. That brings AAIF to 10 hosted projects, up from three launch projects in December 2025. The existing ones include MCP, Agent2Agent (A2A), AGENTS.md and goose.

Alibaba's Xiaobin Xu said of OpenSandbox: "we don't want its future to be defined by a single company." The line captures the logic of the whole foundation: shared plumbing for agents that no single lab controls.

For readers new to the stack, our explainer on the agent harness covers where skills sit among tools, memory and controls.

What the announcement does not say

Two gaps stand out. First, Anthropic's page does not state the license for the standard. Second, it does not list AAIF members beyond the co-founders. Neither is a red flag by itself, but both matter to a company deciding whether to build on the format.

Security is the live issue. A skill is instructions and code that an agent runs with your permissions, so installing a stranger's skill resembles installing a stranger's script. One secondary report says Snyk found 76 malicious skills in a sample from a skills marketplace, which prompted Anthropic to propose security guidance. We have not verified that count, and Anthropic's own post only promises guidance.

What it means for builders

If you ship an agent product, the donation lowers the risk of building on skills, since the name and the repository no longer belong to one lab. If you write skills, expect a versioned spec and a proposal process, which means breaking changes should become rarer and better announced.

How skills differ from MCP

Skills and MCP are often mentioned together, but they do different jobs. A skill is a folder of instructions and optional code that an agent reads when a task matches its description. MCP, which Anthropic also donated to AAIF, is a protocol that connects an agent to live external systems. A skill can tell an agent how your team writes a report; an MCP server can let the agent fetch the data for it. Both now sit in one foundation, with the same lead figure, David Soria Parra, central to MCP and Agent Skills alike. That overlap simplifies coordination, and it also concentrates decisions in a small circle, which is one more reason to watch the proposal process once it exists.

Our position

Giving away a standard that 70 platforms already use costs Anthropic little in lock-in terms and buys goodwill and stability. That does not make it a bad deal for everyone else. The test is whether the first versioned release arrives on a published schedule and whether security guidance has teeth. We will report both when they land.