On 22 October 2026, Anthropic donated its Agent Skills standard to the Agentic AI Foundation, a group that operates under the Linux Foundation. It might sound like a dry governance announcement, but it says a lot about where AI agents are heading and who gets to shape the rules they follow.
What Agent Skills are
An agent skill is a packaged set of instructions that teaches an AI agent how to do a particular task well. Think of it as a recipe card. A skill might explain how to create a properly formatted Excel file, how to follow a company's brand guidelines, how to write a weekly report in a team's preferred structure, or how to review a contract against a checklist.
Technically, a skill is a folder containing a file called SKILL.md. That file starts with a short header, called frontmatter, with two required fields, a name and a description, plus four optional fields. Below that header are the instructions themselves. The folder can also contain supporting files such as templates, examples or scripts.
The clever part is how skills load. An agent sees only the short description of each available skill. When a task matches a description, the agent reads the full instructions. This means an agent can have hundreds of skills available without being overloaded by instructions it does not need for most tasks.
How popular the format has become
Anthropic opened the Agent Skills format in December 2025. According to figures Anthropic shared at the time of the donation, about one million skills have been published and nearly 280 million skills installed since then. Reports describe more than 70 agent tools that can now read the format, including ChatGPT, Cursor, GitHub Copilot and VS Code.
That spread is the real story. A format that started inside one company's products has become something many competing tools understand.
What the Agentic AI Foundation is
The Agentic AI Foundation, often shortened to AAIF, is a neutral home for open standards related to AI agents. It sits under the Linux Foundation, the non-profit organisation that also hosts projects such as the Linux kernel and Kubernetes.
The AAIF was set up in December 2025, and one of its first projects was the Model Context Protocol, or MCP, another standard Anthropic created. MCP lets AI assistants connect to outside tools and data in a standard way. Agent Skills now joins MCP under the same roof.
On the same day as the Agent Skills donation, other companies contributed projects too, including OpenAI's Open Responses, Hugging Face's OpenEnv and Alibaba's OpenSandbox. Reports put the foundation at 10 projects and 305 members after these additions, up from three projects at launch.
Why give it away
There are several reasons a company would hand a successful standard to a neutral foundation.
Trust. Competitors are more willing to build on a standard when no single company controls it. If Agent Skills remained Anthropic's alone, rivals might worry about future changes that favour Anthropic's products.
Wider adoption. A neutral standard is easier for large enterprises and governments to adopt, because it reduces dependence on one vendor.
Shared maintenance. Once many companies rely on a format, keeping it stable, secure and well documented becomes a shared job.
Precedent. The same approach worked for MCP, which became widely adopted across the industry after being placed under neutral governance.
What changes for developers
For most people building or using skills today, nothing changes immediately. Existing skills keep working. According to reports on the donation, the code is licensed under Apache 2.0 and the documentation under Creative Commons Attribution 4.0, so anyone can use or contribute to the standard without joining the foundation.
Some things are still being worked out. The specification does not yet have formal version numbers or a defined process for proposing changes and handling security issues. Expect the foundation to set these up over time, probably following the model used for MCP.
Portability is worth noting. Some tools, including Claude Code, add their own optional fields to skills. Those extra fields may not work on other platforms. If you want a skill to work everywhere, stick to the fields defined in the standard and keep a separate version for any platform-specific extras.
The security side
Skills are powerful because agents follow their instructions and can run any scripts they contain. That also makes them a security risk. A malicious skill could instruct an agent to leak data or run harmful code. Reports have described campaigns that planted hundreds of malicious skills in a single marketplace.
The practical advice is straightforward:
- Install skills only from sources you trust.
- Read a skill's instructions and scripts before installing it.
- Pin versions so a skill cannot change silently after you approve it.
- Do not treat optional permission fields as a real security boundary.
Neutral governance should eventually bring clearer security processes, but for now, caution is your responsibility.
Why this matters beyond Anthropic
The bigger picture is that the AI industry is converging on shared building blocks for agents. MCP standardises how agents connect to tools. Agent Skills standardises how agents learn procedures. Other projects in the same foundation cover how agents run code safely and how they exchange responses.
For businesses, including Indian IT firms building agent solutions for global clients, common standards reduce lock-in. A skill written once can work across several agent platforms, and a company can switch AI providers without rewriting all its procedures.
The bottom line
Anthropic giving Agent Skills to the Linux Foundation is a bet that open, shared standards will make AI agents more useful for everyone, including Anthropic. For developers, it means the skills you write today are more likely to work across many tools tomorrow. Keep your skills portable, treat third-party skills with healthy suspicion, and watch the Agentic AI Foundation as it sets the rules that agents across the industry will follow.
