Agent Skills moves to the Agentic AI Foundation: what it means for teams

On October 22, 2026, Anthropic donated the Agent Skills standard to the Agentic AI Foundation (AAIF), a Linux Foundation project that Anthropic co-founded with Block and OpenAI. AAIF announced it the same day alongside three other new projects, Open Responses, OpenEnv and OpenSandbox, bringing the foundation to 10 hosted projects in its first year.

SkillRepo is built on this standard, so we have a stake in it. Here is the short version for a team that runs skills in more than one agent. Nothing in your repos changes because of the donation. The format is one a team can standardize on: the same folder already runs in seven agents, and it is now maintained under a written proposal process. What remains is your team's work: which version, in which repo, approved by whom.

What Anthropic announced

According to Anthropic's announcement, the donation covers the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain. Anyone can keep using the standard or contributing to it without joining AAIF. Engineers from Anthropic, Google, OpenAI and other companies will maintain it as individuals, with David Soria Parra and Mahesh Murag as lead maintainers.

What did not change, and what to watch

The format itself does not change. A skill is still a folder with a SKILL.md file and optional supporting files, and that same folder runs today in Claude Code, Cursor, Codex CLI, Gemini CLI, Windsurf, Cline and VS Code + Copilot. There is no versioned release of the specification yet, so the folder you have is the one every agent reads.

What Anthropic did propose is a roadmap: versioned releases, a formal proposal process, and security guidance for people who write and install skills. None of it has a date, and none of it changes the files in your repos.

What changes, and what does not

A format maintained by engineers from several companies, under a written proposal process, is one more people can plan around. That is good for everyone who writes skills, and it is good for us.

What the standard describes is the folder. Portable is not the same as current, and it is not the same as approved. The donation does not change where skills run; it changes who holds the format, and that makes it a good moment to treat the skill folder as the unit your team approves and distributes to every agent. Once you do, the questions that matter are different ones. Which version is each repo running? Is the copy in one person's dev environment older than the copy in another's? Who approved the version your agent loaded this morning? If a version turns out to be wrong, how do you get it out of service?

Nothing in the announcement or the proposed roadmap is about those questions. A standard defines the file. Getting the right version of that file into every agent, on every machine, and knowing who decided it was the right one, is a different layer.

For teams, git is the right place to keep the source. What git does not do is tell you that a colleague's copy is three commits behind. You fix a skill on Monday, and on Thursday half the team's agents are still running the old one. More on SkillRepo and git.

Where SkillRepo fits beside the Agent Skills standard

Three layers:

• The standard, now at AAIF, defines what a skill file is.
• Your agents read that file from a folder they already watch.
• SkillRepo sits between them. It gets the version your team approved into those folders on every machine, keeps it current, and shows you when a copy has drifted. It also tells you what you are about to run before you run it: who published it, under what license, with what safety grade, and who on your team approved it.

SkillRepo is the open distribution and governance layer for AI agent skills. We build on the Agent Skills specification like anyone else, and nothing in the announcement changes how we work with it. SkillRepo is independent of AAIF and the Agent Skills project.

Distribute. Keep your skills in the GitHub repository they already live in; SkillRepo syncs every push. On each machine, a one-time skillrepo init writes your library as plain files into the folders your agents read. From then on it stays current, automatically at session start in most agents. You stop pulling. Edits made locally are never overwritten, but you are now aware of them. Nothing is held here: your source stays in your repository and what lands on disk is the standard's own folder, readable without us.

Govern. For a team, the same delivery adds control: pin the versions you approve, hold a publisher's update in a review queue until a lead approves it, and scope what each repo receives. See which repos have drifted, and keep an audit log. Recall a version and each repo falls back to the last good one on its next sync. How teams use it.

Trust. A skill you pull from the catalog carries a two-layer A to F grade, its license, its provenance, and attribution to a verified GitHub identity. The grade is a signal, not a certification: read what a skill does and review it before you run it. Your own skills stay private by default, and the content we process is never used to train models and never sold.

Nothing about your existing skills needs to change. Run npx skillrepo init in one working repo and look at what landed on disk; both the Developer and Team plans start with a free trial and no card.