Gave Up the Name, the Logo, and the Domain—Keeping Only the Implementation: Anthropic Donates Agent Skills to a Neutral Foundation, Turning It Into a 'Common Standard' After MCP
Anthropic has donated the name, logo, GitHub, and domain of "Agent Skills" to the neutral AAIF foundation. It's the second donation after MCP, and on the same day OpenAI, Hugging Face, and Alibaba each contributed components of their own. The implementation and pricing, however, stay in each company
What It Gave Up Was the "Brand"; What It Kept Was the "Substance"
On October 22, 2026, the Agentic AI Foundation (AAIF), part of the Linux Foundation, announced that it had taken on four new projects. At the center is "Agent Skills," donated by Anthropic. It's a mechanism that hands an AI agent a bundle of instructions, scripts, and reference material, loading them only when needed; a single "skill" is represented as a folder containing a file called SKILL.md.
What changed hands in this donation was not the technology itself, but the "brand" side of things. Specifically, the following four items moved to the AAIF.
- The Agent Skills name
- The logo
- The GitHub organization
- The agentskills.io domain
The contents and behavior of skills, on the other hand, are unchanged. "A skill is still a folder containing a SKILL.md," and the pricing, the way it's delivered, and Claude's own implementation all remain in Anthropic's hands. The brand moves to the commons, while the product implementation stays in-house—that's the dividing line.
A "Second Donation" Following MCP
The AAIF is a foundation launched jointly by Anthropic, Block, and OpenAI in December 2025. CTO and co-founder Manik Surtani describes the foundation's role as "one place to build common standards, protocols, and foundations where no single company controls the outcome." It already holds the Model Context Protocol (MCP), Agent2Agent, AGENTS.md, goose, and more; with these four additions, it has grown from three to ten projects within a year of its launch.
For Anthropic, this is a second donation, following its handover of MCP to the same foundation. Developer Simon Willison recalls that back in December 2025 he had predicted "Skills will eventually end up at the AAIF, just like MCP."
Four "Common Components" Lined Up on the Same Day
What's notable is that this wasn't about Agent Skills alone: competing companies each brought different components to the table on the same day. Four projects came together, mapping onto the various stages of agent development.
| Project | Donated by | Area covered |
|---|---|---|
| Agent Skills | Anthropic | A portable "skill" format for giving agents capabilities and domain expertise |
| Open Responses | OpenAI | A spec that standardizes API calls across providers (reducing rewrites when switching) |
| OpenEnv | Hugging Face | Portable, containerized training and evaluation environments for reinforcement learning |
| OpenSandbox | Alibaba | A sandbox where coding/GUI agents can safely execute code |
"Gain capabilities," "talk to the model," "train and evaluate," and "run safely"—the lineup aims to fill in each stage an agent goes through before it can run, using common standards that aren't tied to any particular vendor.
What This Means on the Ground for Development and Operations
In practical terms, the significance is a loosening of vendor lock-in. Agent Skills moves to a structure where engineers from Anthropic, Google, OpenAI, and others maintain it "in a personal capacity" under a neutral foundation, with David Soria Parra and Mahesh Murag serving as Lead Maintainers. Because the spec becomes harder to change unilaterally at the convenience of any one company, it's easier to assume that a skill you write once can be reused across multiple agents.
From a management and adoption standpoint, the bigger implication is that skills and API integration shift in status from "product" to "infrastructure (common foundation)." As was the case with MCP, standardization lowers the cost of switching tools and makes it easier to keep leveraging the instructions and agent assets you've built in-house even when the tools you adopt change.
Read the "Widespread Adoption" Numbers With a Grain of Salt
That said, it would be a mistake to view this with pure optimism. Since its December 2025 release, Agent Skills is said to have been adopted by more than 70 platforms, including ChatGPT, Cursor, GitHub Copilot, and VS Code—but these are vendor-reported figures. One analysis cautions that "this is the number of platforms that have announced support, not the number of people actually using it, and it says nothing about how well each integration is built out." It's safer not to read the size of the adoption count directly as the depth of real usage.
It's also worth keeping in mind that while the brand and neutral governance have moved, the implementation, pricing, and delivery model stay on Anthropic's side. The banner of an "open common standard" and the implementation and billing that each company continues to hold within its own products exist on separate layers. Making a standard neutral does not immediately erase competition or lock-in among the companies—that is a sober view that also holds up.
References: Linux Foundation (AAIF four-project announcement) / Simon Willison, "Donating Agent Skills to the Agentic AI Foundation" / SiliconANGLE / orcarouter.ai (notes on what was transferred and the figures)
