Anthropic hands Agent Skills to the foundation behind MCP

Anthropic is giving Agent Skills, the SKILL.md format that started inside Claude Code, to the Agentic AI Foundation. Its October 22 announcement also shows how far the format has spread beyond Claude: 1 million skills published, nearly 280 million installs, and more than 70 agent platforms that load them.

The transfer covers the Agent Skills name and logo, the GitHub organization and the agentskills.io domain, Anthropic said in its post. The foundation sits under the Linux Foundation. Anthropic co-founded it with Block and OpenAI last year and has already given it the Model Context Protocol. Anthropic’s ClaudeDevs account spelled out what this means for anyone already using skills: existing skills keep working, the licenses stay the same, and you don’t have to join the AAIF to use the standard or contribute to it.

The adoption numbers grew quickly. Anthropic launched Agent Skills on October 16, 2025, and within weeks developers had published thousands of skills on GitHub and written loaders for Codex, Cursor and OpenCode. On December 18, 2025, Anthropic published the format as an open standard with its own GitHub organization and domain. Mahesh Murag, a product manager at Anthropic, said at the time that the skills repository had “already crossed 20k stars on GitHub, with tens of thousands of community-created and shared skills,” per VentureBeat. Ten months on, that count of community skills has grown to 1 million. On average, each published skill has been installed about 280 times, although Anthropic gives no figure for how those installs are distributed.

The list of platforms grew too. In December, Murag said Microsoft had adopted skills in VS Code and GitHub, along with coding agents including Cursor, Goose, Amp and OpenCode, and that Anthropic was “in active conversations with others across the ecosystem.” The list Anthropic gave today includes ChatGPT next to Cursor, GitHub Copilot and VS Code.

The format is simple, which helps explain why other vendors could pick it up. A skill is a folder with a SKILL.md file that teaches an agent one task or domain, plus any scripts or reference files the task needs. The file opens with YAML frontmatter that must include a name and a description. At startup, the agent loads only those two fields for every installed skill into its system prompt. It reads the full body of SKILL.md only when a task looks relevant, and it opens the bundled files only when it needs them. Anthropic calls this layering progressive disclosure. VentureBeat reported that a skill summarized this way costs just a few dozen tokens of context. Skills can also include executable code for steps where a script is more reliable than generated text. Anthropic says it built the format because Claude Code was good at general knowledge work but often lacked the procedural knowledge that real business tasks require.

The two announcements describe governance a little differently. Anthropic’s blog post says 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 ClaudeDevs post describes a Technical Steering Committee and names more companies: Anthropic, Earendil (Pi), Google, Microsoft, OpenAI and OpenCode. In both versions, the first job is a roadmap Anthropic has proposed: versioned releases, a formal proposal process, and security guidance for people who write and install skills. That guidance is only being proposed now, after nearly 280 million installs of folders that can carry executable scripts.

The foundation took in more than one standard that day. On the same Thursday, OpenAI donated its Open Responses API spec to the AAIF. The foundation launched in December 2025 with MCP, goose and AGENTS.md, so Anthropic’s tool protocol and its skill format now share a home with OpenAI’s API spec. Anthropic says the project will follow the MCP approach, with priority on community input and transparent decision-making.

The specification is at agentskills.io, discussion happens on GitHub, and feedback on proposed specification changes goes through the project’s Discord.