adtestbench Higgsfield

AI news · Anthropic · open standards

Anthropic hands Agent Skills to the Agentic AI Foundation

The standard for reusable agent instructions joins MCP at the Linux Foundation group, which took in projects from OpenAI, Hugging Face and Alibaba the same day.

By Alexandre S. , 06:30 UTC

Anthropic has given Agent Skills, its open standard for teaching AI agents a particular job, to the Agentic AI Foundation (AAIF), a group run under the Linux Foundation, Anthropic said on Thursday 22 October 2026. The foundation already looks after MCP, the connection standard Anthropic donated in 2025.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a domed building with columns standing on the horizon.
Drawn by adtestbench from “Donating Agent Skills to the Agentic AI Foundation”,

In practice a skill is a folder holding a written how-to (the SKILL.md file) and any scripts or documents the agent needs for that job. More than 70 agent platforms have adopted the format since it went public in December 2025, VS Code, Cursor, ChatGPT and GitHub Copilot among them, and Anthropic counts 1 million skills published and close to 280 million installs. The agentskills.io domain, the GitHub organisation, the name and the logo all pass to the foundation, while individual engineers from OpenAI, Google, Anthropic and others stay on as maintainers.

Three more projects joined the same day, SiliconANGLE reports: Alibaba’s OpenSandbox and Hugging Face’s OpenEnv, both for the simulated set-ups labs train models in, and OpenAI’s Open Responses, a shared template for model APIs. The AAIF now runs ten projects, against three at its launch last December.

A workflow written once as a skill, such as a weekly ad report or a variant brief, already runs in ChatGPT, Copilot and Claude, and the donation takes the name, domain and spec out of Anthropic’s hands, so no single AI vendor controls where it goes next.