The second donation in a pattern
Anthropic has donated the Agent Skills open standard to the Agentic AI Foundation, the Linux Foundation-hosted body it co-founded with Block and OpenAI, transferring the name and logo, the GitHub organization, and the agentskills.io domain out of its own control. This follows the same move Anthropic made with Model Context Protocol, which it donated to the same foundation after MCP had already become, in the words of Anthropic's chief product officer Mike Krieger, the industry standard for connecting AI systems to data and tools.
The timeline matters for understanding the strategy here. Agent Skills launched in October 2025, and Anthropic published the format as an open standard under its own GitHub organization and domain in December 2025. Only after a full year of adoption, once the standard had proven itself across the ecosystem, did Anthropic hand over control. That sequencing, build it, prove it works, then open it, looks like a deliberate playbook rather than a one-off decision.
The adoption numbers behind the decision
Anthropic's own figures put more than 70 agent platforms on Agent Skills, including ChatGPT, Cursor, GitHub Copilot, and VS Code, with more than 1 million skills published and nearly 280 million installs. Those numbers come from Anthropic itself, with no disclosed methodology for how installs are counted or whether the figure deduplicates across platforms, so enterprise buyers should treat them as directional rather than audited.
Even with that caveat, the breadth of platform adoption is the more meaningful signal than any single install count. A standard that competing agent platforms, including products built by direct competitors to Anthropic, have chosen to support is a stronger proof point than raw usage figures, because it means rival vendors found more value in compatibility than in building a competing format.
What actually changes in governance
Under the new structure, maintainers are engineers from Anthropic, Google, OpenAI, and other companies acting as individuals rather than as company representatives, with lead maintainers David Soria Parra and Mahesh Murag carrying over from Anthropic's team. Contributing to the standard does not require joining the Agentic AI Foundation, which keeps the barrier to participation low for smaller players who want influence without the overhead of a formal membership commitment.
The practical reason to move the name, repository, and domain out of a single company's account is to give competing vendors genuine say in the standard's direction going forward, rather than leaving them dependent on one company's roadmap and goodwill. Anthropic's stated roadmap commitments, versioned releases, a formal proposal process, and security guidance, are described in reporting as a proposal rather than a dated commitment, which means the real test of this governance model is still ahead rather than already proven.
The security gap enterprises should not ignore
A skill, by design, can include executable scripts that run with the user's own permissions, which means the open, vendor-neutral framing does not automatically translate into a vetted or secure one. With a million skills published and no disclosed curation mechanism, the standard almost certainly includes duplicate, abandoned, or simply untested skills sitting alongside genuinely useful ones, and nothing in the governance change addresses that quality gap directly.
Enterprise technology leaders adopting Agent Skills inside their own agent platforms should review third-party skill scripts before deployment and pin versions rather than auto-updating to whatever a public repository serves next, treating this the same way a mature organization already treats any other open-source dependency with a large, loosely curated package ecosystem. The right benchmark for judging whether this governance change actually delivers is the foundation's first real versioned release and the substance of its promised security guidance, not the announcement itself.
Why this matters beyond Agent Skills specifically
This donation reinforces a broader pattern worth tracking: the leading AI labs are increasingly choosing to open-source the connective infrastructure around their models, tool-calling protocols, agent conventions, skill formats, while keeping the models themselves proprietary and commercial. That split lets Anthropic benefit from ecosystem-wide adoption and goodwill on the infrastructure layer while still competing hard on the model layer where its actual revenue comes from.
For enterprise architects deciding which agent standards to build around, that pattern is itself useful signal. A standard with multi-vendor platinum backing and a track record of being handed to neutral governance once adopted is a safer long-term bet than a competing proprietary format controlled by a single vendor, simply because the governance structure reduces the risk of the standard being changed unilaterally in ways that disadvantage any one platform relying on it.
