Agent Skills Moves From Anthropic to the Linux Foundation

On October 22, 2026, Anthropic donated Agent Skills to the Agentic AI Foundation (AAIF), a group under the Linux Foundation. If your roadmap has an item that says "ship this as a skill", the owner of the standard just changed. The announcement describes a change of owner, not a change of format. Agent Skills at the Linux Foundation is a governance story, so the useful question is what governance does for your plans.

What happened: Agent Skills gets a neutral home

Anthropic gave AAIF four things:

• The Agent Skills open standard
• The Agent Skills name and logo
• The agentskills.io GitHub organization
• The agentskills.io domain

AAIF was co-founded in 2025 by Anthropic, Block and OpenAI, and it already hosts MCP, which Anthropic donated earlier. Engineers from Anthropic, Google, OpenAI and others will maintain the standard as individuals. David Soria Parra and Mahesh Murag are the lead maintainers. Anyone can use or contribute to the standard without joining AAIF.

Anthropic's own figures: about 1 million published skills, nearly 280 million installs, and more than 70 platforms that support skills, including ChatGPT, Cursor, GitHub Copilot and VS Code. Those are the vendor's numbers, not audited ones, so don't paste them into a business case as active users.

If you missed skills the first time: a skill is a folder with a SKILL.md file that teaches an agent a task, plus optional scripts and reference material. Simon Willison noted he had said in December 2025 that he wouldn't be surprised to see it end up at AAIF.

Why Agent Skills at the Linux Foundation matters for your roadmap

What gets better. A skill you write for one agent now sits on a standard no single vendor controls. That lowers the risk of building on it. Anthropic's proposed roadmap adds versioned releases, a formal proposal process, and security guidance for people who write and install skills.

What doesn't change. Notice the word "proposed". Neutral governance is not a stability guarantee, and it doesn't make 70+ platforms behave identically. The spec itself marks the allowed-tools field as experimental and says support varies between agent implementations. Don't schedule work that depends on roadmap items that haven't shipped.

Security is still on you. Skills can bundle scripts, so installing a third-party skill means running someone else's code. The security guidance is still on the roadmap, so your own review process is the only guardrail today.

Keep your skills portable with spec-only fields

The specification requires two frontmatter fields, name and description. The optional ones are license, compatibility, metadata and allowed-tools. Stick to those and your skill has the best chance of loading everywhere.

---
name: release-notes-drafter
description: Drafts customer-facing release notes from merged pull requests and a changelog. Use when the user asks for release notes, a changelog summary, or a "what's new" post.
license: Apache-2.0
metadata:
  owner: product-ops
  version: "1.0"
---

# Release notes drafter

1. Read the merged pull request titles and the CHANGELOG.
2. Group changes by customer-facing impact.
3. Draft notes in plain language, one short paragraph per group.

Save it as release-notes-drafter/SKILL.md. The name must be lowercase letters, numbers and hyphens, at most 64 characters, and it must match the parent directory name. The description is what an agent reads to decide whether to load the skill, so say what it does and when to use it. Put your own tracking data, such as an owner or version, under metadata rather than inventing new top-level keys.

The spec points to the skills-ref reference library for checking a skill. Check the agentskills repository for how to install it:

skills-ref validate ./release-notes-drafter

This checks that the frontmatter is valid and follows the naming rules.

What to do on Monday morning

1. Inventory your skills. List which exist, who owns each, and which clients load them.
2. Find non-spec frontmatter. Any field outside the six above may fail on other clients. Keep it only where you knowingly target one product.
3. Gate third-party skills. Treat anything with a scripts/ folder like a dependency: someone reads it before it ships.
4. Record the spec revision. Versioned releases are only proposed, so note which version of the spec you validated against.
5. Watch the proposals. Spec discussion happens in the agentskills repo, and Anthropic points feedback on proposed changes to the project's Discord.

When this changes nothing for you

If your skills live in one team's repo and run on one client, this is a quiet quarter. Don't invent migration work for a standard that hasn't broken anything.

For anyone putting agents into production, a vendor-neutral spec means the skills you write today are less likely to be stranded by one company's roadmap, as long as you review what you install and keep to the fields the spec defines.