Anthropic Donates Agent Skills to AAIF — What Changes for Developers Anthropic donated its Agent Skills standard — a SKILL.md-based format for packaging AI agent workflows that has reached 280 million installs, 70+ platform integrations and one million published skills — to the Agentic AI Foundation under the Linux Foundation on October 22. The transfer covers the Agent Skills name, logo, GitHub organization, specification and the agentskills.io domain, while Claude Code's agent implementation and Anthropic-created skills stay with Anthropic; existing SKILL.md files remain compatible with no migration required. On the same day OpenAI donated Open Responses, Hugging Face donated OpenEnv and Alibaba donated OpenSandbox, expanding the AAIF from six projects to ten. 280 million installs. 70+ agent platforms. And Anthropic just gave away the standard behind all of it. On October 22, Anthropic donated Agent Skills — the open standard for packaging specialized AI agent workflows — to the Agentic AI Foundation AAIF https://aaif.io under the Linux Foundation. The same day, OpenAI donated Open Responses, Hugging Face donated OpenEnv, and Alibaba donated OpenSandbox. In a single afternoon, the AAIF went from six projects to ten, and the infrastructure layer for agentic AI quietly moved into neutral governance. What Is Agent Skills and Why Do 280M Installs Exist Agent Skills is a simple but powerful standard: a folder containing a SKILL.md file with YAML frontmatter and Markdown instructions, plus optional scripts and reference files. It tells an AI agent what to do for a specific task — not through executable code, but through structured knowledge. When an agent starts up, it pre-loads skill metadata into its system prompt; when it activates a skill, it loads the full instructions. Anthropic launched it in October 2025, published it as an open standard in December 2025, and by October 2026 it had reached 70+ integrations https://www.anthropic.com/news/donating-agent-skills-to-the-agentic-ai-foundation : ChatGPT, Cursor, GitHub Copilot, VS Code, and dozens more. One million skills published. Those are infrastructure numbers, not toy-project numbers. What Anthropic Actually Handed Over The donation covers: the Agent Skills name and logo, the GitHub organization, the specification, and the agentskills.io domain. All of it moves to AAIF stewardship. What stays at Anthropic: Claude Code’s agent implementation and any Anthropic-created skills. The standard transfers; the product does not. This is a meaningful distinction — Anthropic is giving away governance of infrastructure, not a business unit. What You Need to Do Nothing urgent. Existing SKILL.md files remain fully compatible. No migration. No rewrites. The spec is backward compatible, and the AAIF has explicitly said developers can keep using or contributing to Agent Skills without joining the foundation. What does change: issues and proposals now go to the AAIF GitHub https://www.linuxfoundation.org/press/agentic-ai-foundation-announces-four-new-projects-agent-skills-open-responses-openenv-and-opensandbox , not Anthropic’s repos. The foundation’s roadmap includes versioned spec releases, a formal proposal process similar in spirit to Python’s PEP process , and security guidance for skill authors and consumers. If you’re building production skills, this is worth watching — the governance process is about to get real. The October 22 Sweep Anthropic wasn’t the only one writing checks that day. OpenAI donated Open Responses — a multi-provider LLM API spec designed to let developers switch between model providers without rewriting their code. Hugging Face contributed OpenEnv, a framework for reinforcement learning environments. Alibaba contributed OpenSandbox, a general-purpose sandbox platform for agent workloads. All four moved into neutral governance on the same day. The AAIF now hosts ten projects in under a year. If you’re building agents in 2026 and haven’t reviewed what AAIF hosts, the list is worth your attention: MCP, AGENTS.md, goose, Agent Skills, Open Responses, OpenEnv, OpenSandbox, and three others. The Real Play This is MCP all over again, and that’s not a criticism. Anthropic donated MCP to the AAIF in December 2025. Within ten months, MCP had 10,000+ servers and had become the de facto connection protocol for AI agents across the industry — including from companies that compete directly with Anthropic. The pattern is deliberate: donate the standard, stay active in development, let the ecosystem grow without the friction of “one company controls this.” Open interoperability standards remove adoption barriers. Fewer barriers means more agents get built. More agents mean more API calls — and Anthropic’s business is selling inference, not owning spec governance. Agent Skills follows the same logic. If a skill you write today works in Cursor, Claude Code, ChatGPT, and GitHub Copilot without modification, the market for skills expands. The market for Claude inference expands with it. It’s not altruism. It’s leverage. Read the Agent Skills specification https://agentskills.io/specification to see exactly what the standard covers now that the AAIF owns it. The AAIF is quietly building the TCP/IP layer for agentic AI — infrastructure nobody owns but everybody depends on. October 22 moved four more pieces of that into neutral hands. Watch what gets contributed next.