Anthropic is giving away Agent Skills, the open standard that teaches AI agents how to handle specific tasks, to the Agentic AI Foundation. The move, announced on October 22, means no single company will steer where Agent Skills goes next. It arrived alongside three other donated projects from OpenAI, Hugging Face and Alibaba.
What Anthropic Is Donating With Agent Skills
A skill is simply a folder with a SKILL.md file inside. That file explains a task or a subject to an agent, and the folder can also hold scripts and reference material the agent loads only when it needs them.
Anthropic built the idea while working on Claude Code, then launched skills in October 2025. It published Agent Skills as an open standard in December 2025, so other agents could read the same format. If you have used Claude for work, you may already have run into skills without noticing.
The donation covers more than the spec itself. Anthropic says it is handing over:
- The Agent Skills name and logo
- The project’s GitHub organization
- The agentskills.io domain
You do not need to join the foundation to keep using the standard or to contribute to it. Engineers from Anthropic, Google, OpenAI and other companies will maintain it as individuals, with David Soria Parra and Mahesh Murag as lead maintainers.
Takeaway: the format behind Claude’s skills now belongs to a shared foundation, not to Anthropic alone.
How Big Agent Skills Has Grown
The numbers Anthropic shared show how fast the format spread beyond Claude. The company says more than 70 agent platforms have added support, including ChatGPT, Cursor, GitHub Copilot and VS Code.
| Metric | Figure (per Anthropic) |
|---|---|
| Agent platforms supporting the standard | More than 70 |
| Skills published so far | About 1 million |
| Skill installs | Nearly 280 million |
| Open standard published | December 2025 |
Anthropic has proposed a starting roadmap for the project under its new home. It includes versioned releases, a formal process for proposing changes, and security guidance for people who write and install skills.
Three More Projects Joined the Same Day
The foundation, which runs under the Linux Foundation, announced all four new projects at its AGNTCon + MCPCon North America event in San Jose. Each one covers a different layer of how agents are built and run.
Open Responses from OpenAI
Open Responses is an open spec for talking to language models from different providers through one shared interface. It is based on OpenAI’s Responses API and defines a common schema, streaming events and tooling. The goal is that you describe a request once and run it across model providers with little rewriting.
OpenEnv from Hugging Face
OpenEnv is a framework and community hub for reinforcement learning environments, the practice spaces where agents are trained. It packages each environment as a container that speaks both MCP and the Gym interface that training libraries already use.
OpenSandbox from Alibaba
OpenSandbox is a general sandbox platform for running agent code safely. It ships SDKs in several languages, unified sandbox APIs, and Docker and Kubernetes runtimes for coding agents, GUI agents, evaluations and training. Safe execution is a hot topic right now, as seen in recent launches like agent sandbox tools.
Why the Agentic AI Foundation Matters
Anthropic co-founded the foundation last year with Block and OpenAI, and it already donated the Model Context Protocol there. With these four additions, the group says it has grown from three projects to 10 in its first year.
The other projects it hosts are MCP, the Agent2Agent protocol, Agent Router, agentgateway, AGENTS.md and goose. Putting rival companies’ standards under one neutral roof makes it more likely that tools from different vendors keep working together.
What This Means for You
If you build or use AI agents, this is mostly good news. Here is what to keep in mind:
- Skills you write today should keep working across Claude, ChatGPT, Cursor and other supporting tools.
- Changes to the format will go through a public proposal process instead of one company’s roadmap.
- Security guidance for skills is on the roadmap, so check what you install until it lands.
- If you switch between model providers, Open Responses may cut down the code you have to rewrite.
For Claude Code users, nothing changes overnight. You can keep using skills alongside features like Claude Code Review and tips such as skipping permission prompts. For more agent tooling updates, see our coverage of agent tracing support and the latest AI news.
Sources
Primary source: Anthropic’s Agent Skills donation announcement. Additional details from the Linux Foundation press release and SiliconANGLE.
Conclusion
Anthropic’s decision to donate Agent Skills turns a Claude-born idea into shared industry plumbing. With more than 70 platforms already on board, the standard now gets neutral governance and a public roadmap. OpenAI, Hugging Face and Alibaba added their own projects the same day, which shows how much the big players want agents to work together. For you, that should mean skills and agent code that travel more easily between tools. The next thing to watch is the security guidance promised for people who write and install skills.
