Anthropic hands Agent Skills to the Linux Foundation: why portable AI agent skills matter for small business

On October 22, 2026, Anthropic donated Agent Skills, an open format for packaging the instructions an AI agent follows to do a specific job, to the Agentic AI Foundation, which is part of the Linux Foundation. For small businesses, the practical news is portability: the written-down know-how that teaches an agent how your business does a task is becoming less tied to any single AI company. The catch is security, because a skill from an unknown source can carry harmful instructions.

The announcement came from both Anthropic and the Linux Foundation, timed to the AGNTCon + MCPCon North America conference in San Jose.

What are AI agent skills?

An AI agent skill is a folder of written instructions, plus any scripts or reference files, that teaches an agent how to do one specific task. Anthropic describes a skill as a folder containing a SKILL.md file “that teaches an agent a specific task or domain,” in its October 22 announcement. The Agentic AI Foundation’s own post calls it a standard for packaging reusable instructions that teach agents to perform specific tasks.

Anthropic’s earlier engineering write-up compares building a skill to putting together an onboarding guide for a new hire. That is a useful way to picture it. A skill is where a business writes down how a job is done properly:

• A dental clinic’s steps for confirming tomorrow’s appointments and flagging patients who have not replied.
• An HVAC company’s checklist for turning a service call into a written quote request for the office.
• A property manager’s rules for sorting maintenance emails by urgency.
• An online store’s process for answering “where is my order?” messages.

What changed on October 22, 2026?

Ownership of the standard moved to a neutral foundation. Anthropic said it is donating the Agent Skills name and logo, the GitHub organization and the agentskills.io domain to the Agentic AI Foundation, and that people can keep using the standard without joining the foundation, according to its announcement. Search Engine Watch reported the same transfer of branding, GitHub organization and website.

Agent Skills was one of four new projects the foundation announced that day. The Linux Foundation’s release lists Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face and OpenSandbox from Alibaba, bringing the foundation to 10 hosted projects in its first year. Anthropic notes that it previously donated the Model Context Protocol (MCP) to the same foundation.

How widely are AI agent skills used?

Widely, at least among AI tools. Anthropic says more than 70 agent platforms, including ChatGPT, GitHub Copilot and VS Code, have added support for Agent Skills, and that about 1 million skills have been published and nearly 280 million installed, per its announcement. The Agentic AI Foundation gives the same figures. Both come from parties involved in the standard, so treat them as their own counts rather than independent research.

The foundation’s post gives examples of companies building internal libraries of skills so that staff can share proven ways of working, and notes that a common format lets a company “carry its guidance across compatible tools.”

Why does portability matter to a small business?

Because the most valuable part of an AI agent setup is often not the AI model. It is the description of how your business works: the steps, the exceptions, the tone and the rules about when to ask a person. If that know-how is written in a common format, it is more likely to survive a change of tools.

There are limits. The Agentic AI Foundation notes that the agent still needs the right access and capabilities to do the task. Search Engine Watch points out that the specification’s field for pre-approved tools is experimental and supported differently across products, so a procedure that runs in one product can need different permissions in another. The Agent Skills specification itself marks that field as experimental.

So portability is a strong reason to keep your workflows documented and in your own hands. It is not a promise that everything moves over with a single click.

Are AI agent skills safe to install?

Not automatically. A skill is instructions plus, sometimes, code, and both can be harmful. Anthropic’s own guidance warns that malicious skills “may introduce vulnerabilities” or direct an agent “to exfiltrate data and take unintended actions,” and recommends installing skills only from trusted sources and reading every bundled file before using one from a less trusted source, according to its engineering write-up. Search Engine Watch summarized the same advice.

The risk is not theoretical. In February 2026, security firm Snyk said it scanned 3,984 skills from public skill directories and found 76 containing confirmed malicious payloads for credential theft, backdoors or data theft, with 13.4% of the skills it scanned having at least one critical security issue, according to its ToxicSkills research. Those figures come from a single vendor’s study, so read them as one data point, but they match the warning in Anthropic’s guidance. Anthropic also says the foundation’s proposed roadmap includes security guidance for people who write and install skills.

This connects to a risk we covered in our post on prompt injection through email: an agent follows instructions, so whoever writes the instructions matters.

What should small businesses do now?

Three practical steps, whichever AI tools you use:

1. Write your workflows down. Document how your important repeat tasks are done, including the exceptions and the points where a person must decide. That document is valuable with or without AI.
2. Ask who owns the instructions. If a provider builds AI agents for you, ask whether the instructions and workflow documents behind them are yours, and whether you can export them if you leave. We include that question in our guide on how to choose an AI agent provider.
3. Treat new skills like new software. Nobody on your team should add a downloaded skill to an agent that can reach your email, files or accounts without someone reviewing it first. The same thinking applies to any AI running without approval.

Frequently asked questions

Do I need to know what Agent Skills are to use AI agents?

No. Most business owners will never open a skill file. What matters is that someone has written down clearly how each task should be done, that the instructions are reviewed, and that you can get a copy of them.

Does this mean I can switch AI tools without starting over?

It makes switching easier for the written instructions, but not effortless. The new tool still needs access to your systems, and some features work differently from product to product. Plan for testing whenever you move a workflow.

Is a free skill from a public directory safe to use?

Only after it has been checked. Anthropic recommends installing skills from trusted sources and auditing anything else before use, paying attention to scripts and to instructions that connect to outside websites.

Want to know which of your workflows are worth writing down first?

Before any agent is built, it helps to know which of your repeat tasks are well understood, low risk and worth documenting. An Agent Opportunity Audit maps those workflows with you, so the know-how stays with your business.