Anthropic donates Agent Skills, its format for teaching tasks to AI agents, to the AAIF foundation
IA en un minuto newsroom · Editor: Jon Elgezabal
In 30 seconds
Anthropic has handed Agent Skills, its open standard of instructions for AI agents, to the Agentic AI Foundation, which is tied to the Linux Foundation. Engineers from several companies, among them Google and OpenAI, will look after it personally, and it can still be used without joining the foundation. The transfer covers brand, logo, domain and GitHub organization, just as happened before with MCP.
Agent Skills, Anthropic's format for teaching AI agents how to do tasks, is no longer in the hands of a single company: on Thursday, October 22, Anthropic announced it is handing it over to the Agentic AI Foundation (AAIF), part of the Linux Foundation.
Each skill is a folder holding a SKILL.md file that shows an agent how to carry out one particular task, plus whatever scripts or supporting documents it requires. Anthropic came up with the format while building Claude Code, launched it in October 2025 and released it as an open standard that December. The company says more than 70 platforms have adopted it since then, among them VS Code, ChatGPT, GitHub Copilot and Cursor, and that one million skills have been published, with close to 280 million installs.
The handover covers the name, the logo, the GitHub organization and the agentskills.io domain. The project will be looked after by engineers from Anthropic, OpenAI, Google and other firms acting as individuals, led by David Soria Parra and Mahesh Murag, and nobody will have to join the AAIF to use it or contribute. According to the AAIF blog, a technical committee with members from several organizations will steer it, and the specification is open source under the Apache 2.0 and CC BY 4.0 licenses.
It follows the path of MCP (the protocol that connects AI to other systems), which Anthropic had already given to the same foundation, set up in 2025 by Anthropic, Block and OpenAI. As a first step, Anthropic proposes a roadmap with numbered versions, a formal process for suggesting changes and security guidelines for those who write and install skills.
Why it matters · analysis and opinion
When products from rival companies use the same format, the question that holds many businesses back is who controls it. The handover settles that: the rules will be set by a committee with people from several organizations, the same path MCP followed, and Anthropic no longer has the final say. For anyone building skills inside a team, the practical change is that the work is not tied to a single assistant and can move to another without being redone, as long as the tool is compatible. What is missing sits in the proposed roadmap: numbered versions, a formal process for changes and security guidelines for those who install other people's skills. How that gets resolved will decide whether companies dare to share them outside their own walls.
Official source: Anthropic · Written with the help of AI: how we make the news
What does it mean for your business? · analysis and opinion
This affects small businesses, freelancers and teams in Spain and Latin America that already work with tools compatible with this format, such as ChatGPT, Cursor, GitHub Copilot or VS Code, and that write or install skills for tasks they repeat often. For anyone who only uses a chatbot for one-off questions, the donation changes nothing day to day, and the standard can still be used without joining the AAIF.
The sensible move is to store internal procedures (for example, how to review a report against the company's own criteria) as a folder with its SKILL.md and whatever templates it needs, so they can be carried from one compatible tool to another without rebuilding them. Before handing a skill to a new agent, it is worth checking which programs it requires, whether it needs an internet connection and what access that agent has, because support for some experimental fields varies between implementations.
The main risk lies in installing third-party skills without reviewing them. Anthropic has proposed security guidance for those who create and install them, but for now it is only a proposal, not approved rules. Our advice is to read each SKILL.md and its scripts before activating them, give the agent only the permissions it really needs and stick to skills from known sources.
