Agent Skills Donated to the Agentic AI Foundation: An Open Standard for AI Agents and What It Means for Vendor Lock-In

In brief: On 22 October 2026 Anthropic announced it was donating the Agent Skills standard (along with its name, logo, GitHub organisation and the agentskills.io domain) to the Agentic AI Foundation (AAIF) under the Linux Foundation. According to Anthropic, since being published in December 2025 the standard has been integrated by over 70 agent platforms (including ChatGPT, Cursor, GitHub Copilot and VS Code); about 1 million skills have been published and nearly 280 million installed. A skill is a folder containing a SKILL.md file that teaches an AI agent how to do a given task. What the donation means: the business know-how you teach agents can be used across different platforms without being locked into one company's format. This article is based on Anthropic's announcement (23 October 2026).

Last updated: 23 October 2026

What is Agent Skills?

In Anthropic's description, Agent Skills is an open standard that gives AI agents specialised knowledge and workflows. Each skill is a folder with a SKILL.md file describing a task or domain, plus scripts and reference material if needed. Anthropic developed it while building Claude Code, having seen that agents lacked the procedural knowledge business tasks require.

A simple example: a skill for “our company's invoice checking process” defines which fields to check, which errors are common and the format of the output. When it needs to, the agent loads the skill and does the job accordingly.

What was donated, and to whom?

• Donated: The standard, the Agent Skills name and logo, the GitHub organisation and the agentskills.io domain.
• Recipient: The Agentic AI Foundation (AAIF) under the Linux Foundation. Anthropic says it co-founded the foundation last year with Block and OpenAI, and also donated the Model Context Protocol (MCP) to it.
• Why: Anthropic says it wants the standard to remain “open, community-driven and vendor-neutral.” You do not need to join the foundation to use or contribute to the standard.
• Governance: Engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals; the proposed roadmap includes versioned releases, a formal proposal process and security guidance for skill authors and installers.

SD Times reported that the foundation announced four new projects the same day, contributed by “major AI players” and covering how agents gain new capabilities. I could not verify the project names, so I do not list them here.

Why does it matter for businesses?

1. Portability: your knowledge isn't trapped in one agent

If the effort you put into teaching an agent your internal processes is kept in a common format, it can be reused across different agent platforms. That reduces vendor dependence: even if the tool you choose today changes tomorrow, you can carry your skills over. (Together with MCP, this makes two shared layers: tool connection and business know-how.)

2. Writing skills builds an asset

Well-written skills are the executable form of your company's institutional knowledge: the sales quote template, the support reply rules, the code review checklist. Treat them like a product: give them an owner, a version and a review process.

3. A new security surface

Because skills contain instructions and scripts, they are a kind of code and authority handed to an AI agent. That is why the standard's roadmap including “security guidance for skill authors and installers” matters. This week's LiteLLM supply chain attack showed that every component in the AI toolchain is an attack surface. Installing a skill whose source you don't know is like installing a package you don't know.

4. AI agents are becoming enterprise tools

Standardisation is a sign that agents are moving from “experiment” to “business process.” To manage this shift, also look at access and authority: 8 controls when giving AI agents access to company data.

How to prepare: 7 steps

1. List recurring processes: Choose 5–10 processes you want to teach agents (report preparation, code review, support replies, etc.).
2. Write them as skills: For each process, prepare a short document with purpose, inputs, steps, a checklist and an example output.
3. Put them under version control: Skills should be versioned, reviewed and owned like code.
4. Verify the source: Before installing a skill from outside, examine its source and contents (including scripts).
5. Run with least privilege: Limit which tools and data a skill can reach.
6. Test on more than one platform: Verify that the same skill behaves consistently in two agents; see portability with real evidence.
7. Apply privacy rules: Don't embed customer data or secrets in a skill. See our AI use policy guide.

Caution: being a standard doesn't solve everything

• The figures (70+ platforms, ~1 million skills, ~280 million installs) are Anthropic's own statements with no independent verification.
• Having the same standard does not mean every platform works equally well; test with your own data.
• The foundation's governance is still new; the roadmap is at the “proposed” stage.

AI agent projects with Aktaş Digital

We plan projects that transfer your business processes to AI agents securely and portably and define authority and data boundaries. To get a scope and cost breakdown, you can use our quote form. Related services: AI solutions.

Frequently Asked Questions

Are Agent Skills and MCP the same thing?

No. In Anthropic's description, MCP is the protocol for agents to connect to tools and data; Agent Skills are knowledge packages that teach agents how to do specific tasks. Both were donated to the AAIF.

Do I need to know how to code to write a skill?

A basic skill is a folder and a text file called SKILL.md; scripts are optional. But for authority and security matters it is good to get technical support.

Does this standard end vendor lock-in?

It reduces it but doesn't end it. Skills written in the same format can be used on several platforms; but models, prices and integrations still differ by platform.

Is downloading ready-made skills from outside safe?

Like any outside component, it carries risk. Verify the source, inspect the contents and run with least privilege.

Conclusion

Agent Skills moving to a neutral foundation opens an era of a common format for business know-how for AI agents. The opportunity for businesses is portability; the risk is unsafe skills. The first thing you can do today: pick three recurring processes you want agents to learn and write each as a one-page skill.