Anthropic Donates Agent Skills to the Agentic AI Foundation: What the Open Standard Used by 70+ Platforms Means for Your Organization (October 2026)

What Happened on 22 October 2026

Anthropic announced that it is donating the Agent Skills standard to the Agentic AI Foundation (AAIF), a foundation under the Linux Foundation that Anthropic co-founded with Block and OpenAI at the end of 2025. The donation covers the Agent Skills name and logo, the GitHub organization that holds the specification, and the agentskills.io domain. The announcement coincided with AGNTCon and MCPCon North America in San Jose, where the Linux Foundation accepted four new projects on the same day.

If you already use skills in Claude, Claude Code or another tool, there is nothing to do. The file format does not change, and existing skills keep working. What changes is who owns the standard and who decides its direction from here. This article covers that part. For how to write a skill in detail, see our complete guide to Claude Skills and Agent Skills from April. The facts here were checked against Anthropic's announcement, the Linux Foundation press release and agentskills.io on 23 October 2026.

Who the Agentic AI Foundation Is, and What It Hosts

AAIF is a foundation within the Linux Foundation, in the same way that Kubernetes sits under the Cloud Native Computing Foundation. Its job is to be the neutral place where companies that compete in the market can agree on shared standards without any one of them holding the keys. According to reporting at launch, there were eight Platinum members: Amazon Web Services, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI. The foundation now hosts 10 projects.

Source: Linux Foundation press release, 22 October 2026, and launch coverage from December 2025

Agent Skills in One Minute

A skill is a folder with a SKILL.md file. The file starts with a short header, at minimum a name and a description, followed by plain-language instructions on how to do the job. You can add scripts, document templates and reference material in subfolders. An agent loads only the name and description at startup, and loads the full content when a task matches. That is why an agent can keep dozens of skills on hand without filling its context window. Here is a condensed example for an accounting team.

month-end-close/
├── SKILL.md
├── references/close-checklist.md
└── scripts/check-unposted.py

--- SKILL.md ---
---
name: month-end-close
description: Monthly accounting close checklist. Use when the user mentions month-end, period close or closing the books.
license: Proprietary. LICENSE.txt has complete terms
metadata:
  author: finance-team
  version: "2.3"
---
1. Run scripts/check-unposted.py to find documents not yet posted.
2. Walk through references/close-checklist.md item by item and report what fails.
3. Never report the period as closed until every item passes.

You can validate a skill with skills-ref validate ./my-skill from the reference library in the project's GitHub. To see how skills differ from plugins and MCP, read Claude Code vs Plugin vs MCP vs Skills.

What Changes and What Does Not

David Soria Parra, co-creator of MCP at Anthropic, said in the press release that the goal is for "anyone building with skills" to be able to "shape what's next". It is the same path MCP took when Anthropic donated it to the same foundation ten months earlier.

What This Means for Thai and Regional Organizations

This looks like developer news, but the real effect lands on the business teams who write "how we do things here" into skills. Three things to take away.

1. Your team's knowledge moves with you, not with the tool

A month-end close checklist, a purchase request review, a template for official correspondence: a skill your team writes for Claude today runs in ChatGPT, Cursor or Copilot in the same format. The hours spent writing skills are an investment in the organization's knowledge, not in a single vendor. But do not read this as "any tool will do". An open standard makes the work portable. The quality of the result still depends on the model and the vendor you choose. Choosing well at the start still matters most.

2. A million skills is a new supply chain to screen

3. Give every skill an owner, as you would a form

A good skill is an SOP (Standard Operating Procedure) an agent can read, so treat it like a controlled document. Record the author and version in the metadata field, keep everything in one git repository, have the department head approve before publishing, and use Claude's share-with-specific-people feature to limit who sees which skill. Start this at skill number ten, and you will not be cleaning up at skill number one hundred.

A View From the ERP Side

At Grand Linux we used skills on our own work first, such as the pre-publication article check, the quotation workflow and the system acceptance checklist, and then offered the same approach to customers running ERP (Enterprise Resource Planning): a month-end close checklist, purchase document review, budget status summaries. These skills work alongside connecting Claude to Saeree ERP over MCP, with role-based permissions and an audit trail. Saeree ERP has no built-in AI. The ERP is the source of truth, the skill is the procedure, MCP is the data pipe, and Claude is the reader and preparer.

Today's news means that the "procedure" layer is not tied to any one vendor, which is good for customers directly. Details are on our Claude Solutions page.

Conclusion

Anthropic donated Agent Skills to the Agentic AI Foundation on 22 October 2026, following the path MCP took earlier. The file format does not change and existing skills keep working, but decisions about the specification move from a single company to a neutral foundation. For organizations in Thailand and the region, that is one more reason to start writing team procedures as skills, with an owner, a version, and the same screening before installation you would give any software.

An open standard does not tell you whom to choose. It only guarantees that what your team writes today is not wasted when the tool changes.

- Saeree ERP team, on Agent Skills joining the Agentic AI Foundation, 22 October 2026

References

• Anthropic — Donating Agent Skills to the Agentic AI Foundation (22 October 2026)
• Linux Foundation — Agentic AI Foundation Announces Four New Projects (22 October 2026)
• agentskills.io — Agent Skills Specification
• Simon Willison — Donating Agent Skills to the Agentic AI Foundation (22 October 2026)
• Phoronix — Linux Foundation's Newest Endeavor: The Agentic AI Foundation (December 2025)