Anthropic is giving away control of Agent Skills, the format it created for teaching AI agents specific jobs. The standard moves to the Agentic AI Foundation, a Linux Foundation project that Anthropic co-founded with Block and OpenAI and that already looks after the Model Context Protocol. For anyone who writes or installs skills in Claude Code, Codex, Cursor or other tools, the practical effect is that no single vendor now decides where the format goes.
A skill is simply a folder with a SKILL.md file describing a task, plus any scripts or reference files the agent needs. Anthropic built the idea while working on Claude Code, which it says handled general work well but lacked the step-by-step knowledge real business tasks require. Adoption was fast. Developers published thousands of skills on GitHub within weeks of the October 2025 launch, and Anthropic made the format an open standard that December. It now counts more than 70 agent platforms with support, including ChatGPT, GitHub Copilot and VS Code, 1 million published skills and nearly 280 million installs.
The handover covers the name and logo, the GitHub organisation and the agentskills.io domain. Engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals, led by David Soria Parra and Mahesh Murag. The proposed first steps are versioned releases, a formal process for proposing changes, and security guidance for people who write and install skills. Nobody has to join the foundation to use or contribute to the format.
This repeats the path MCP took, and it has the same logic. Rivals are more willing to build on a format they help govern, and a neutral home makes it easier for companies to commit. Anthropic's own skills repository, which we review, remains one source among many.
The security work is the part to watch. Skills can bundle scripts, and installing one from a stranger is close to running their code with the agent's permissions. Nearly 280 million installs have happened before any shared guidance existed, and the roadmap is still a proposal. Versioning and a change process will help tool makers, but until there is a clear way to check what a skill does before installing it, the burden stays with users to read what they install.