Anthropic announced Tuesday it’s donating its Agent Skills standard to the Agentic AI Foundation, a Linux Foundation project it co-founded with Block and OpenAI last year. The move comes less than ten months after the company first introduced Agent Skills as an “open standard” for teaching AI agents new capabilities.

Simon Willison called it back in December. When Anthropic first declared Agent Skills open, he wrote, “I wouldn’t be surprised to see this donated to a neutral foundation at some point.” Turns out some point was October.

Agent Skills is Anthropic’s specification for packaging up reusable commands that AI agents can invoke. Think of them as plugins, but with a standardized interface. Instead of every company building proprietary agent extension systems, the idea is to create a common format that works across models and platforms.

This is the second major agent-related standard Anthropic has handed to the AAIF. The company donated the Model Context Protocol (MCP) to the same foundation earlier this year. MCP handles how agents connect to data sources and tools. Skills handles what those agents can actually do once they’re connected.

The donation matters because it removes Anthropic from the position of being both participant and referee. Other companies can now contribute to the standard without wondering if they’re building on top of a competitor’s strategic asset. The Linux Foundation model has worked for keeping infrastructure projects genuinely neutral. We’ll see if it works for agent protocols.

The timing is interesting. Agent infrastructure is going from experimental to real fast. NVIDIA released OpenShell this week, a runtime specifically designed for running fleets of autonomous agents safely. The pitch is sandboxed execution with privacy controls, which matters when you’re running agents that can actually do things instead of just chatting.

The pattern is familiar if you’ve watched other infrastructure transitions. First, a company ships something proprietary. Then they open-source it. Then they hand governance to a foundation when they realize adoption matters more than control. Kubernetes went this way. So did Node.js, after some drama.

The difference with agent standards is we’re compressing that timeline. Anthropic shipped Agent Skills in December 2025 and donated it ten months later. Kubernetes took two years to land at the CNCF. The agent ecosystem is moving faster, probably because nobody wants to rebuild the browser wars with AI agents.

What this actually means for developers: if you’re building agent tooling, you can now point to a spec that isn’t owned by any one company. If you’re building agents, you’ve got a clearer path to making them work across different platforms. If you’re just watching, this is what infrastructure standardization looks like when everyone learned from the last twenty years of platform fights.

The AAIF now controls two of the three critical pieces of agent infrastructure: how agents connect to tools (MCP) and what skills they can execute (Agent Skills). The missing piece is how agents coordinate with each other, which someone will probably standardize next quarter.

Anthropic gets credit for moving fast on this. Donating a standard you created less than a year ago takes conviction that open beats closed, or at least that being the company that gave away the standard is better positioning than being the company that tried to own it.

The real test isn’t the announcement. It’s whether the foundation actually functions as neutral ground, whether other companies contribute, and whether Agent Skills becomes the standard or just another spec in a pile of competing approaches. Check back in six months.