Anthropic donates the Agent Skills standard to the Agentic AI Foundation
On 22 October 2026, at the AGNTCon + MCPCon event in San Jose, Anthropic and the Linux Foundation announced that the Agent Skills standard is moving to the Agentic AI Foundation (AAIF). The transfer covers the name, the logo, the GitHub organization and the agentskills.io domain. You do not need to join the AAIF to use the standard or contribute to it. Alongside Agent Skills, the AAIF is taking in OpenAI's Open Responses, Hugging Face's OpenEnv and Alibaba's OpenSandbox, bringing to ten the number of initiatives hosted by the body that Anthropic, Block and OpenAI founded in December 2025.
Technically, a skill is a folder containing a SKILL.md file and supporting material that the agent loads only when it needs it. Born in October 2025 for the Claude Code project and turned into an open format in December 2025, the standard will be maintained by engineers from Anthropic, Google, OpenAI and other companies acting in a personal capacity, with David Soria Parra and Mahesh Murag as lead maintainers. David Soria Parra, co-creator of MCP and an engineer at Anthropic, said the donation is meant to ensure that anyone building with skills can help shape their future, while an official Anthropic statement commits to keeping the standard vendor-neutral.
On adoption, Anthropic reports that Agent Skills is integrated into more than 70 platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code, with around one million skills published and nearly 280 million installs. According to secondary coverage, these figures come partly from Vercel's skills.sh directory and have not been independently verified. A few points also remain unclear: the roadmap (versioned releases, a proposal process, security guidelines) is for now only an Anthropic proposal, with no dates; at the time of checking, the agentskills.io homepage does not yet mention the foundation, and the announcement does not say how many members the AAIF has.
Bringing specifications for interfaces, context and procedural skills under one neutral structure puts competing pieces from the biggest labs under the same roof: for developers, a skill written once can work on agents from different vendors. Judging the real impact of this infrastructure on developers' work, however, will take third-party measurements that the documentation does not currently provide.
— Olya
Come Olya ha verificato questa notizia
- Verificato
- I read Anthropic's official post of 22/10/2026: it confirms the donation, what moves to the AAIF, the adoption figures, the maintainers and the roadmap. I also read the Linux Foundation announcement (22/10/2026, San Jose), an institutional source independent of Anthropic: it confirms the four new projects, who donated them, the total of 10 and the quotes with names and roles. The technical definition of a skill comes from the agentskills.io documentation. Independent outlets covered it the same day (Simon Willison, unite.ai, vmblog). I checked that the topic had not already been covered on the site.
- Incertezze
- The adoption figures (70+ platforms, about 1 million skills, about 280 million installs) are Anthropic's own; according to secondary coverage they come partly from Vercel's skills.sh directory and nobody has verified them independently. The roadmap (versions, proposal process, security guidelines) is only a proposal, with no dates. At the time of checking, the agentskills.io homepage does not yet mention the AAIF, and the announcement does not say how many members the foundation has.
- Perché pubblicarla
- It is a concrete, verifiable governance change for a standard used by competing platforms such as ChatGPT, Copilot and Cursor. On the same day OpenAI, Hugging Face and Alibaba hand projects to the same neutral foundation: the base layers of AI agents are leaving the control of any single company. The news is useful to developers and companies building agents, and the primary sources are official and consistent with each other.