Anthropic hands Agent Skills to a Linux Foundation body. A standard with 280 million installs is only now getting versions and security rules

On October 22, Anthropic handed the Agent Skills standard to the Agentic AI Foundation (AAIF), which sits under the Linux Foundation. The format, used by more than 70 agent platforms according to the company, now gets a neutral owner. For developers and companies, though, the bigger story is what the standard still lacks: security rules.

A skill is a folder of instructions and scripts, and people have installed hundreds of millions

A skill is a folder containing a SKILL.md file that teaches an agent a specific task or domain, optionally bundled with scripts and reference material. Anthropic introduced the format in October 2025 while working on Claude Code and published it as an open standard in December 2025. According to the company, a million skills have been published since then and installed nearly 280 million times. More than 70 platforms support the standard, including ChatGPT, Cursor, GitHub Copilot and VS Code.

Anthropic is handing over the Agent Skills name and logo, the GitHub organisation and the agentskills.io domain. The standard remains usable without joining the AAIF. Engineers from Anthropic, Google, OpenAI and other companies will maintain it as individuals. The lead maintainers are David Soria Parra and Mahesh Murag, both from Anthropic. Soria Parra helped create MCP, Murag helped create Agent Skills.

Anthropic co-founded the AAIF with Block and OpenAI last year and contributed the Model Context Protocol to it. Simon Willison noted on his blog that he predicted exactly this move last December.

Companies get a format no single vendor controls

For companies building internal agents, this is a relief similar to MCP. A skill written once for Claude Code should also work in Cursor or Copilot. And its future is no longer decided only by a company whose models compete with everyone else's. A corporate library of procedures is less likely to turn into lock-in with one vendor.

It is also a smart move for Anthropic. It invented the format, spread it, and is handing it over once competitors use it anyway. It keeps its influence through people, since both lead maintainers are its employees. For now, the neutral home looks mostly like a neutral address.

Security guidance arrives after 280 million installs

The roadmap Anthropic proposes starts with versioned releases, a formal proposal process and security guidance for people who write and install skills. The order is telling. A format that can slip arbitrary instructions and executable scripts into an agent spread to hundreds of millions of installs before it got versioning and rules for safe use.

In practice a skill is a prompt and code in a single package. Anyone who pulls one from GitHub without review lets foreign instructions run with the same permissions their agent has. The foundation will not solve this on its own. It only provides a place where the rules can be decided in public.

The first versioned spec and the shape of the security rules will decide it

The first concrete signal will be a versioned specification with a formal change process. It will show whether platforms load skills the same way or each keeps its own extensions. The second will be the security guidance: whether it brings package signing, provenance checks and script restrictions, or just a list of good advice.

The third signal is who maintains it. If leadership roles from Google, OpenAI or smaller platforms join the Anthropic people over the next releases, neutrality will be more than a line in a press release.

Lilith's verdict

Anthropic carried Agent Skills into a neutral foundation, and both lead maintainers brought their company badges along. Meanwhile people have installed 280 million folders that whisper instructions to agents, and the guide to vetting them is still just a roadmap item.