Skip to content

Anthropic

Anthropic hands Agent Skills to the foundation behind MCP

2 min read

Written by an AI model from the linked sources and checked against them by a second model. How we work

Anthropic on October 22 donated Agent Skills, its open standard for teaching agents a task through a folder with a SKILL.md file and any scripts or reference material the task needs, to the Agentic AI Foundation (AAIF), the Linux Foundation body that already holds MCP. If your team writes or installs skills, the handover asks nothing of it: Anthropic says existing skills keep working and the licenses don’t change, and the standard can still be used or contributed to without joining the AAIF.

Since Anthropic published the spec as an open standard in December 2025, more than 70 agent platforms have integrated it, including ChatGPT, Cursor, GitHub Copilot and VS Code, and 1 million skills have been published. Skills came out of Anthropic’s work on Claude Code, and within weeks of their October 2025 launch developers had put thousands on GitHub and written loaders for Codex, Cursor and OpenCode.

The foundation receives the Agent Skills name and logo, the GitHub organization and the agentskills.io domain. Engineers from Anthropic, Earendil (Pi), Google, Microsoft, OpenAI and OpenCode will govern the standard as a Technical Steering Committee, maintaining it as individuals, with David Soria Parra and Mahesh Murag as Lead Maintainers. Anthropic co-founded the AAIF with Block and OpenAI in December 2025, and it says the project will, as with MCP, prioritize community input and transparent decision-making.

The committee’s first job is a roadmap Anthropic has proposed: versioned releases, a formal proposal process, and security guidance for people who write and install skills. For a standard with nearly 280 million installs, that last item is still only a proposal. Anthropic’s own advice dates from the engineering post that introduced skills, which warned that malicious skills may introduce vulnerabilities or direct Claude to exfiltrate data and take unintended actions, and recommended installing skills only from trusted sources and auditing any from a less-trusted source before use.

Skills that pre-approve tools are where those 70-plus platforms can differ: the specification marks its allowed-tools field, a space-separated list of pre-approved tools, “Experimental. Support for this field may vary between agent implementations.”

The two standards a Claude plugin bundles now sit with one foundation: in September Anthropic called plugins, which package MCP and skills together, the main way third-party developers build extensions for Claude.

Feedback on proposed changes to the specification goes through the project’s Discord; the spec itself is at agentskills.io, with the discussion on GitHub.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.