Nearly 280 million installs, 1 million published skills, more than 70 supporting platforms — and until recently, this format belonged to a single company. That has now changed. On October 22, 2026, Anthropic announced it is handing the Agent Skills standard over to the Agentic AI Foundation (AAIF), a neutral organization under the Linux Foundation. For developers everywhere, the move is another step toward open, vendor-neutral foundations for AI agents.
Table of Contents
Key Takeaways
- On October 22, 2026, Anthropic handed the Agent Skills standard to the Agentic AI Foundation, including the name, logo, GitHub organization, and the agentskills.io domain.
- The AAIF already governs the Model Context Protocol (MCP), another Anthropic donation.
- More than 70 platforms already support Agent Skills, including ChatGPT, Cursor, GitHub Copilot, and VS Code.
- Roughly 1 million skills have been published with nearly 280 million installs, according to Anthropic.
- David Soria Parra and Mahesh Murag are now the lead maintainers; engineers from multiple companies will continue contributing as individuals.
What the Agent Skills Standard Actually Is
The Agent Skills standard defines a simple but effective format that teaches AI agents how to reliably carry out specific tasks. At its core, a “skill” is just a folder containing a file called SKILL.md. That file describes in plain text what the skill is for and how an agent should use it. It’s often paired with scripts, sample data, or reference material the agent can pull in as needed.
The concept works like a specialized manual: instead of retraining a large language model for every niche use case, the agent gets a compact, readable package it can load on demand. Developers can build skills for almost any domain, from accounting rules to a specific company’s coding conventions.
The format originated inside Claude Code, Anthropic’s software development tool, where it first appeared in October 2025. Just two months later, in December 2025, Anthropic released it as an open standard — a clear signal that the idea was never meant to stay locked to a single product.
Why Handing It to a Neutral Foundation Matters
An open format only stays genuinely open as long as control doesn’t sit with a single company. That’s exactly the gap this handoff to the Agentic AI Foundation closes. The AAIF operates under the Linux Foundation and was co-founded by Anthropic together with Block and OpenAI — direct competitors in the AI market.
That structure is deliberate. When several competing companies jointly steward a foundation, the risk that the standard gets quietly bent to favor one vendor drops sharply. For developers, that translates into more planning certainty: teams building skills for their product today don’t have to worry that Anthropic will change the rules tomorrow.
“A standard controlled by a single company isn’t really a standard in the eyes of most developers — it’s a product feature with a marketing name.”
The benefits of the foundation structure boil down to three points:
- Interoperability: Agent Skills works across platforms, not just inside Anthropic’s own ecosystem.
- Avoiding vendor lock-in: Companies using Skills aren’t tied to a single provider.
- Shared development: Engineers from multiple companies now maintain the standard together.
Concretely, the donation covers the “Agent Skills” name, its logo, the GitHub organization, and the agentskills.io domain. That means the symbolic and infrastructural pieces of the project now sit with the foundation, not with Anthropic alone.
Agent Skills vs. MCP
Anyone following AI agents has probably already encountered the Model Context Protocol (MCP) (background: MCP, Memory, Skills, and Hooks: How AI Agents Really Work), another open standard that Anthropic developed and later handed to the AAIF. MCP and Agent Skills solve different but related problems.
| Aspect | Model Context Protocol (MCP) | Agent Skills Standard |
|---|---|---|
| Purpose | Connects an agent to external data sources/tools | Teaches an agent task-specific knowledge |
| Basic form | Protocol for server-client communication | Folder structure with a SKILL.md file |
| Typical use | Access to databases, APIs, file systems | Instructions for specific tasks or domain knowledge |
| Governance | Agentic AI Foundation | Agentic AI Foundation (new) |
| Originated at Anthropic | Yes | Yes |
While MCP handles the technical connection between an agent and the outside world, Agent Skills supplies the content-level “how do I do this.” The two standards complement each other and now live under the same institutional roof. That makes it easier for developers to build agents that both reach external systems and apply specialized task knowledge, without depending on proprietary interfaces.
Adoption by the Numbers
The figures Anthropic published show that Agent Skills had already achieved meaningful reach even before the foundation handoff. Worth noting: these numbers come from Anthropic itself and haven’t been independently verified.
- More than 70 agent platforms have already integrated Agent Skills, including well-known names like ChatGPT, Cursor, GitHub Copilot, and Visual Studio Code.
- Roughly 1 million skills have been published so far.
- Nearly 280 million installs have been recorded.
That reach is notable given the format only launched in October 2025 and was published as an open standard in December 2025. Within roughly a year, Agent Skills has grown into a format with relevance well beyond Anthropic’s own ecosystem, even though the exact counting methodology hasn’t been made public.
For developers and companies weighing whether to invest in AI agents, this adoption curve offers a practical argument: a format already supported by tools like GitHub Copilot or VS Code is easier to fold into existing development workflows than a one-off, isolated solution.
Four New Projects in One Day
The Agent Skills handoff didn’t happen in isolation. On the same day, October 22, 2026, the Agentic AI Foundation announced four new projects at the AGNTCon + MCPCon conference in San Jose:
- Agent Skills (Anthropic)
- Open Responses (OpenAI)
- OpenEnv (Hugging Face)
- OpenSandbox (Alibaba)
That bundled announcement shows that standardization in agentic AI is now being pushed forward by several major players at once, not just Anthropic. The fact that companies like OpenAI, Hugging Face, and Alibaba contributed their own projects to the same foundation on the same day points to a growing appetite for shared, open foundations instead of competing walled gardens.
Worth clarifying: other news published that same day, such as Google’s Agent Compute Engine or AWS’s Infinity runtime, is unrelated to the Anthropic donation. Those are separate, independent announcements from other companies.
Who Maintains the Standard Now
With the handoff to the Agentic AI Foundation, the structure behind ongoing development also changes. The lead maintainers of the Agent Skills standard are now David Soria Parra and Mahesh Murag, both of whom played a central role in shaping the format at Anthropic.
The way further contributions are organized is notable: engineers from Anthropic, Google, OpenAI, and other companies will keep maintaining the standard, but as individuals, not as representatives of their employers. That model is familiar from other open-source projects and is meant to prevent any single company’s interests from dominating technical direction. Decisions about the format’s future are intended to rest more on technical expertise than on corporate affiliation.
What It Means for Developers
For developers already working with Claude Code, GitHub Copilot, or similar tools, little changes day-to-day right away: existing skills keep working, and the SKILL.md file structure stays the same. The handoff matters mainly in the medium term, because it builds trust. Anthropic is also expanding its own agent lineup, for example with Claude Duet.
Anyone deciding whether to invest time building their own Agent Skills can now be more confident that investment isn’t tied to the business strategy of a single vendor. That’s especially relevant for organizations integrating AI agents into internal processes and relying on long-term stability.
Bottom Line
Handing the Agent Skills standard to the Agentic AI Foundation is more than a symbolic gesture. It shifts control over a format with nearly 280 million installs away from a single company and toward a neutral foundation backed by several competitors. Combined with the Model Context Protocol already governed there, this creates a shared, open foundation for agentic AI where tools, knowledge, and communication can come together without being tied to one specific vendor.
Developers and companies evaluating AI agent strategy now have good reason to take a closer look at the official documentation at agentskills.io and at the Agentic AI Foundation’s project structure. Anyone planning their own agents or integrations should check whether existing workflows map onto the Agent Skills standard, especially since broad support from tools like VS Code and GitHub Copilot lowers the barrier to entry.
