Featured Agent Skills Skills Open Standard Claude Code MCP

Agent Skills Joins the Agentic AI Foundation: What Changes

On October 22, 2026, Anthropic gave Agent Skills to the Linux Foundation's AAIF. What moved, what stays the same, and how to keep a SKILL.md portable.

• 7 min read • By Claude World

On October 22, 2026, Anthropic donated Agent Skills, the open standard behind the SKILL.md format, to the Agentic AI Foundation (AAIF): the Linux Foundation fund it co-founded with Block and OpenAI, and the home it gave the Model Context Protocol (MCP) in 2025. The format did not change. A skill is still a folder with a SKILL.md file, and Claude Code loads the skills you already have exactly as before. What changed is who owns the standard’s name, GitHub organization, and domain, and how the spec will evolve from here.

This article covers what was transferred, what the AAIF is, what it means for Claude Code users and skill authors, and how to write a skill that works beyond Claude Code.

What Anthropic Announced

Anthropic’s announcement calls Agent Skills “the open standard that gives AI agents specialized knowledge and workflows.” The donation covers “the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain.”

The people maintaining it are named too: “Engineers from Anthropic, Google, OpenAI, and other companies will maintain the standard as individuals, with David Soria Parra and Mahesh Murag as Lead Maintainers.”

The post also recaps how far the format has spread. Anthropic developed skills while working on Claude Code and launched them in October 2025. Within weeks, developers had published thousands of skills on GitHub and written loaders for Codex, Cursor, and OpenCode. Anthropic then published Agent Skills as an open standard in December 2025, “under its own GitHub organization and domain, so any agent could use it.” The adoption figures, as Anthropic words them:

  • “More than 70 agent platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code, have integrated Agent Skills since we published the standard in December 2025.”
  • “So far, 1 million skills have been published and nearly 280 million installed.”

What the AAIF Is

The AAIF was announced on December 9, 2025, when Anthropic donated MCP to it. That post describes the foundation as “a directed fund under the Linux Foundation, co-founded by Anthropic, Block and OpenAI, with support from Google, Microsoft, Amazon Web Services (AWS), Cloudflare, and Bloomberg.” MCP joined goose from Block and AGENTS.md from OpenAI as founding projects, and Anthropic said MCP’s governance model would remain unchanged.

Agent Skills follows the same path. In the new post’s words: “As with MCP, the project will continue to prioritize community input and transparent decision-making.” Membership is not a requirement: “You can continue to use the standard or contribute to it without joining the AAIF.”

What Changes and What Doesn’t

For Claude Code users

Nothing to do. Claude Code’s skills documentation still says “Claude Code skills follow the Agent Skills open standard,” and skills still load from ~/.claude/skills/, .claude/skills/, plugins, and the other documented locations. No Claude Code changelog entry mentions the donation or a spec change; the skill-related entries in v2.1.308 through v2.1.318 concern Claude Code’s own handling of skills, such as the fix for paths: skills not being found after a Bash cd deeper into a git repository (v2.1.312).

For skill authors and platform builders

The file format is the same. The specification on agentskills.io defines the same frontmatter as before the announcement; its source file in the project’s repository was last edited in August 2026. The addresses are the same too, since the domain and the GitHub organization moved with the standard.

What changes is governance. Anthropic has “proposed starting with a roadmap covering versioned releases, a formal proposal process, and security guidance for people who write and install skills.” These are proposals, not yet processes: as of November 4, the spec repository has no tagged releases.

Why Neutral Governance Matters

Agent Skills is no longer one product’s feature. By Anthropic’s count, more than 70 platforms have integrated it, and loaders for other agents appeared within weeks of launch. Until October 22, the standard’s name, domain, and GitHub organization still belonged to one company.

The post’s answer is a neutral owner. Anthropic says it is “committed to keeping Agent Skills open, community-driven, and vendor-neutral,” and that the AAIF “will provide a neutral home to the standard going forward, ensuring the community has a say in where Agent Skills go next.” It backs that with three specifics: maintainers from several companies who serve as individuals, a priority on community input and transparent decision-making, and participation without foundation membership. The proposed roadmap covers both sides of the format: versioned releases and a formal proposal process for the spec itself, and security guidance for “people who write and install skills.”

How to Write a Skill That Stays Portable

The specification defines six frontmatter fields:

FieldRequiredRule
nameYes1–64 characters: lowercase letters, numbers, and hyphens; no leading, trailing, or consecutive hyphens; must match the folder name
descriptionYes1–1,024 characters: what the skill does and when to use it
licenseNoA license name or a bundled license file
compatibilityNoUp to 500 characters of environment requirements; most skills don’t need it
metadataNoA map of string keys to string values
allowed-toolsNoSpace-separated pre-approved tools; experimental, and support varies by agent

Here is a minimal skill built only from the spec:

release-notes/
├── SKILL.md
└── references/
    └── STYLE.md
---
name: release-notes
description: Drafts user-facing release notes from merged pull requests and commit messages. Use when the user asks for release notes, a changelog entry, or a summary of what shipped in a version.
---

## Steps

1. Collect the merged changes for the requested version.
2. Group them under Added, Changed, Fixed, and Removed.
3. Write one plain-language line per change, following [the style guide](references/STYLE.md).

## Edge cases

- If you find no changes, say so instead of inventing entries.

The spec recommends keeping SKILL.md under 500 lines, moving detail into files such as references/, and linking them with relative paths one level deep. Its skills-ref reference library checks the frontmatter and naming rules:

skills-ref validate ./release-notes

Keep Claude Code extensions out of shared skills

Claude Code accepts all six spec fields, so a spec-only skill loads unchanged. It also reads fields of its own, which are not part of the standard: when_to_use, argument-hint, arguments, disable-model-invocation, user-invocable, disallowed-tools, model, effort, context, agent, background, hooks, paths, and shell. Body features such as !`command` dynamic context injection are Claude Code–only as well.

Claude’s own upload paths are strict. According to the docs, claude.ai skill uploads, the Skills API, and packaging with package_skill.py from anthropics/skills fail with a hard error on any field outside the spec’s six, and dynamic context injection doesn’t work in claude.ai chat or through the API. A few spec fields also behave differently in Claude Code:

  • name is optional in Claude Code, which falls back to the folder name, but the spec requires it. Always set it, and match the folder.
  • allowed-tools also accepts commas or a YAML list in Claude Code; the spec defines a space-separated string. In Claude Code the grant lasts only for the turn that invokes the skill.
  • license and compatibility are accepted by Claude Code, which doesn’t act on them.

To get Claude Code’s invocation control without adding a non-spec field, use settings instead. The docs offer "user-invocable-only" in skillOverrides as the alternative to disable-model-invocation: true when you don’t want to edit the file (it doesn’t apply to plugin skills):

{
  "skillOverrides": {
    "release-notes": "user-invocable-only"
  }
}

For how Claude Code loads skill names first and full instructions on demand, see Skills & Knowledge Loading. To build your own, see Custom Agents & Skills and our Claude Code Skills guide; for designing skills that also serve other agent platforms, see Agent Skill Design.

How to Follow or Contribute

  • Read the spec at agentskills.io.
  • Join the discussion on GitHub at agentskills/agentskills. Its contributing guide, last updated before the donation, sends proposals and questions to GitHub Discussions, keeps Issues for concrete bugs, and does not accept skill submissions.
  • Join the project’s Discord, which the post names as the place “to give feedback on proposed specification changes.”

None of this requires AAIF membership. For the rest of this month’s Claude Code changes, including those skill-related entries, see our November 2026 Claude Code roundup.

Sources