Anthropic has given away control of Agent Skills, its format for teaching AI agents specific tasks. In a post dated October 22, the company said it is donating the standard to the Agentic AI Foundation (AAIF), which operates under the Linux Foundation. It's one of four agent projects handed to the foundation this week, according to The Register.
That sounds like plumbing, and it is. But plumbing decides how the tools built on top of it behave, so it's worth understanding what changed.
What Agent Skills is
Anthropic describes the idea in one line: "A skill is a folder containing a SKILL.md file that teaches an agent a specific task or domain." The folder can also hold scripts or reference material the agent needs. Anthropic says it built skills while developing its Claude Code tool, to supply the step-by-step procedural knowledge general-purpose agents were missing.
A concrete way to picture it: rather than explaining a company's release checklist to an AI agent every time, a developer writes the checklist once as a skill, and any agent that understands the format can load it when the task comes up. The value of a shared format is that the same skill should work across different tools.
According to Anthropic, that portability is already common. The company says more than 70 agent platforms have integrated Agent Skills, naming ChatGPT, Cursor, GitHub Copilot and VS Code among them. Treat that figure as Anthropic's own claim.
What was handed over
Anthropic lists four things it is donating: the standard itself, the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain. Ownership of the name and the website matters as much as the specification, because whoever holds them controls what counts as "official".
The day-to-day work is meant to be shared. Anthropic 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. Its proposed roadmap covers versioned releases, a formal proposal process, and security guidance for writing and installing skills. Developers don't need to join anything: "You can continue to use the standard or contribute to it without joining the AAIF."
Anthropic's stated aim is "keeping Agent Skills open, community-driven, and vendor-neutral." The post doesn't state a license for the standard.
The bigger picture, and the skeptical view
The Register puts the donation in context. AAIF was founded in December 2025 under the Linux Foundation, launched with three projects, and now oversees 10. This week's additions came from four companies at once: Alibaba donated OpenSandbox, for sandboxing what agents can do; Anthropic gave Agent Skills; Hugging Face gave OpenEnv, for building reinforcement-learning environments; and OpenAI gave Open Responses, an open API specification meant to work across model providers. The Register says the foundation has over 300 members.
The Register is also openly skeptical. Its headline frames the moves as "vendor-affiliated AI agent tech" being "adopted by friendly, funded foundation," and it describes the industry's embrace of AI agents as "enthusiastic (or credulous)." That's the fair question to ask. Moving a project to a neutral foundation can genuinely spread control, but some of the companies that created these standards also co-founded the foundation: Anthropic says AAIF was co-founded by Anthropic, Block and OpenAI.
For developers, very little changes on day one. Nothing in Anthropic's post suggests existing skills stop working, and the standard remains open to use. The difference is governance over time: changes to the specification are meant to go through a public proposal process rather than one company's roadmap.
For anyone building with AI agents, that's the part to watch. Whether vendor-neutral governance works in practice will show in the next few versioned releases, and neither Anthropic nor AAIF has put dates on those yet.
Reporting notes