Anthropic Donates Agent Skills Standard to Open AI Foundation

Anthropic announced it is donating Agent Skills, an open standard for packaging specialized workflows that AI agents can load on demand, to the Agentic AI Foundation (AAIF). The foundation, which operates as a directed fund under the Linux Foundation, now holds the Agent Skills name, logo, GitHub organization and the agentskills.io domain. The transfer ensures no single company can unilaterally control where the standard goes next.

The numbers behind the announcement are striking. Since Anthropic published Agent Skills as an open standard in December 2025, more than 70 agent platforms have integrated it — including ChatGPT, Cursor, GitHub Copilot and VS Code. One million skills have been published to date, and nearly 280 million have been installed.

What Agent Skills Actually Does

The standard grew out of a practical problem Anthropic encountered while building Claude Code, which launched in October 2025. The model was capable at general knowledge tasks but struggled with the kind of step-by-step procedural knowledge that real business workflows demand. The fix was structurally simple: a skill is a folder containing a SKILL.md file that instructs an agent on how to perform a specific task, plus any scripts or reference material it needs. Within weeks of Claude Code’s launch, developers had published thousands of skills on GitHub and built loaders for tools like Codex, Cursor and OpenCode. Anthropic then formalized the format as a standalone open standard.

Going forward, engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals rather than as official company representatives. David Soria Parra and Mahesh Murag are named lead maintainers. Anthropic has also proposed a roadmap that includes versioned releases, a formal proposal process for specification changes, and security guidance aimed at both skill authors and the developers who install them.

A Layered Stack, Not a Standards War

It’s worth being precise about where Agent Skills fits in the broader AI interoperability picture, because this space has gotten crowded fast. Google’s Agent-to-Agent (A2A) protocol, announced in April 2025, handles how AI agents built by different vendors discover each other and delegate tasks. Anthropic’s Model Context Protocol (MCP) — which the company also donated to the AAIF in December 2025 — governs how agents access external tools. Agent Skills operates at a different layer still: it defines what domain-specific knowledge an agent can load to perform a task. These three standards are less rivals than complementary layers of the same stack, and most production systems in 2026 run more than one of them simultaneously.

The AAIF itself launched with more than 150 member organizations — reportedly the fastest-growing foundation in Linux Foundation history — and has backing from Google, Microsoft, AWS, Bloomberg and Cloudflare alongside co-founders Anthropic, OpenAI and Block. That coalition gives the governance structure genuine credibility, though adoption of a standard and control of a standard are different things. Holger Mueller of Constellation Research offered a note of caution, observing that Anthropic’s prior success with MCP doesn’t guarantee the same outcome here, and that history doesn’t always repeat itself when a single vendor tries to set an industry standard.

What This Means If You’re a Student or Early-Career Developer

For students already working inside tools like Cursor, GitHub Copilot or VS Code, Agent Skills has an immediate practical implication: any workflow you package as a skill is now portable across all 70-plus platforms that support the standard. Instead of re-configuring AI behavior separately for each tool, you build once and carry it wherever you work. The format itself is low-barrier — a SKILL.md file plus optional scripts and reference materials is all it takes to get started.

The governance donation also opens a concrete contribution pathway. Because the AAIF uses the same governance model as projects like Kubernetes and PyTorch, contributing to Agent Skills carries the same open-source credibility on a resume or portfolio. The announced roadmap — with a formal proposal process and a community Discord for feedback on specification changes — creates entry points that don’t require insider access or deep institutional affiliation. That’s a meaningful opportunity for contributors coming from academic programs or bootcamp backgrounds who want a verifiable open-source track record.

The standard is free to use and contribute to without joining the AAIF. The specification is available at agentskills.io, and community discussion happens on GitHub.

Source: Anthropic