Claude Code

Agent Skills: Anthropic Hands the Standard to the Agentic AI Foundation

2 min read AI-generated

A million skills published, nearly 280 million installed. From now on the standard will be maintained by individuals from Anthropic, Google and OpenAI.

Featured image for "Agent Skills: Anthropic Hands the Standard to the Agentic AI Foundation"

On 22 October Anthropic handed Agent Skills over to the Agentic AI Foundation. The open standard that teaches agents specialised knowledge and workflows now belongs to a foundation under the Linux Foundation rather than to Anthropic.

What actually changes hands

The donation covers the name and logo, the GitHub organisation and the agentskills.io domain. Anthropic’s numbers: since the standard was published in December 2025, more than 70 agent platforms have integrated Agent Skills, among them ChatGPT, Cursor, GitHub Copilot and VS Code. A million skills have been published and nearly 280 million installed.

Anthropic co-founded the Agentic AI Foundation last year with Block and OpenAI. The Model Context Protocol already lives there. You can keep using Agent Skills, or contribute to it, without joining the foundation.

A folder with one SKILL.md

The format is plain: a folder holding a SKILL.md file that explains a task or a domain to an agent, plus whatever scripts and reference material it needs.

Anthropic describes the origin as a by-product of Claude Code. General knowledge work went fine, but the procedural knowledge that real business tasks require was missing. Within weeks of the October 2025 launch, thousands of skills sat on GitHub and developers had written loaders for Codex, Cursor and OpenCode. Only then did Anthropic publish the standard under its own GitHub organisation and domain.

Maintained by people, not by companies

Engineers from Anthropic, Google, OpenAI and other companies will carry the standard forward as individuals, not as representatives of their employers. The lead maintainers are David Soria Parra and Mahesh Murag, the same names behind MCP.

For the start Anthropic proposes a roadmap with three items: versioned releases, a formal proposal process, and security guidance for the people who write and install skills. Feedback on proposed specification changes is meant to run through the project’s Discord.

The second standard Anthropic has given away

MCP was the blueprint, and it worked. I wrote about Walmart’s Agent Gateway today as well, and the shopping agent there speaks MCP rather than some Walmart-specific interface. A standard owned by one vendor is a bet on that vendor’s good behaviour. A standard owned by a foundation is infrastructure.

Watch the security line on that roadmap. A skill is executable text that an agent reads and follows, and 280 million installations are an attack surface nobody has yet said anything binding about.

Sources

Agent SkillsAnthropicMCP