- Anthropic is donating the Agent Skills standard to the Agentic AI Foundation. It says more than 70 platforms support it and nearly 280 million skills are installed.
- A skill is a folder of instructions, scripts and reference files that an agent loads. Anthropic's proposed roadmap includes security guidance for people who write and install skills.
- Leaders should ask which skills their agents use, who wrote them and who approves new ones.
The folder that teaches an agent your process
Every company has know-how that lives in people's heads and in scattered documents. How a refund is approved. How a contract is reviewed. How a report must be formatted. AI agents are good at general work, but they do not know your procedures.
Agent Skills is the standard built to close that gap. On October 22, Anthropic handed control of it to the Agentic AI Foundation (AAIF), a body that operates within the Linux Foundation. The idea is simple. The procedure becomes a folder, and any agent can read it.
What Anthropic announced
Anthropic says it launched skills in October 2025 and published them as an open standard in December 2025. By its count, more than 70 agent products now support the standard. ChatGPT, Cursor, GitHub Copilot and VS Code are among them. On usage, Anthropic reports that skills published so far total 1 million, and that installs are close to 280 million.
Anthropic is donating the Agent Skills name and logo, its GitHub organization and the agentskills.io domain. It co-founded the AAIF with Block and OpenAI last year, and earlier donated the Model Context Protocol (MCP) to it. The maintainers will serve as individuals, not as company delegates. They come from Anthropic, Google, OpenAI and other firms. David Soria Parra and Mahesh Murag are the Lead Maintainers.
These figures and the framing come from Anthropic, the donor. The announcement gives no independent count.
How a skill works
A skill is a folder. Inside is a file named SKILL.md that describes a task or domain. The folder can also hold scripts and reference material the agent needs. The idea came out of Anthropic's work on Claude Code. In its account, the tool was strong at broad knowledge work but lacked the step-by-step know-how that real business tasks demand.
Think of a new hire's onboarding binder. Now imagine the binder can also include small programs. That is why skills are more than documents. A folder that can carry scripts changes what an agent does, not just what it knows.
Why the handover matters
The thesis here is that skills turn company procedure into a portable asset, and a portable asset needs an owner. A standard controlled by one vendor is a risk if your workflows depend on it. Moving it to a foundation is meant to answer that concern, in Anthropic's account. Anthropic commits to keeping the standard open, shaped by its community and not tied to one vendor. It adds that teams can adopt it or help shape it without becoming AAIF members.
Neutrality is Anthropic's stated intent, and it has yet to be tested. Anthropic engineers remain among the maintainers, and the starting roadmap is Anthropic's own proposal.
The comparison to MCP is fair, since Anthropic took the same path with that protocol. It is one more data point, though, not proof that neutral governance works.
What a foundation does not settle
Governance is not the same as vetting. Anthropic's proposed roadmap covers numbered releases, a formal process for submitting changes, and security guidance for skill authors and the people who install their work. The announcement describes no security incidents.
The exposure is easy to see. By Anthropic's count, skills are already widely installed. The proposed roadmap includes security guidance for writing and installing them. The announcement does not say when that guidance will arrive.
Picture the operations manager who turns a team's process into a skill, and the employee who installs a colleague's skill from a public list. Each trusts a folder. Neither may know what its scripts do. The foundation will shape the standard. It will not check what your people install.
Questions to put to your team
First, ask for an inventory. Which skills do your agents and coding tools load today, and who wrote them? Second, ask who approves a new skill before it reaches a shared workspace, and whether someone reads its scripts.
Third, decide which internal procedures you want captured as skills, and who owns them when staff leave. Fourth, assign someone to follow the AAIF security guidance as it appears and compare it with your own rules.
The standard now sits with a foundation that Anthropic calls neutral. The trust decision still sits with you.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: Anthropic.
