AAIF adds Agent Skills, Open Responses, OpenSandbox, OpenEnv

Anthropic, OpenAI, Hugging Face and Alibaba each just handed a piece of their agent stack to someone else. Today the Agentic AI Foundation (AAIF) took in four new projects: Agent Skills, Open Responses, OpenEnv and OpenSandbox. That is the skill format, the model API, the training environment and the sandbox, all moving under neutral governance. For a small business building agents, that is the list of layers you can stop worrying about getting locked into.

What actually happened

Per the AAIF announcement, released at AGNTCon + MCPCon North America in San Jose on October 22:

• Agent Skills, from Anthropic. Reusable packages of instructions, scripts and resources that an agent loads when it needs them, so the same know-how works across agent products.
• Open Responses, from OpenAI. A spec for interoperable, multi-provider LLM interfaces, based on the OpenAI Responses API: one shared schema, consistent streaming events, extensible tools.
• OpenEnv, from Hugging Face. A framework and hub for reinforcement learning environments, each one a container that exposes an agent interface, MCP and the Gym protocol.
• OpenSandbox, from Alibaba. A sandbox platform for AI apps with multi-language SDKs, one sandbox API, Docker and Kubernetes runtimes, and a protocol for custom runtimes.

With these four, AAIF hosts 10 projects, up from three when it launched in December 2025. The others include MCP, A2A, AGENTS.md, goose and agentgateway. Alibaba's Xiaobin Xu summed up the reason to donate: "we don't want its future to be defined by a single company."

The release does not name a license for each project. Check each repo before you ship on it.

Why open agent standards matter for your business

Your skills are now an asset, not a lock. If you write your SOPs as Agent Skills, they move with you when you change agent vendors. Write the procedure once. Keep it in your repo.

Open Responses is the escape hatch for your model bill. Code that speaks one shared request and streaming schema can swap providers when the price changes. Most of our builds already route through a thin adapter. A neutral spec makes that adapter thinner.

Sandboxes are where agent risk lives. An agent that runs code needs a box with walls. A standard sandbox API means you can run that box on your own Docker host or Kubernetes cluster, not only inside a vendor's cloud.

Governance is not a guarantee. A foundation owns the spec. It does not promise every vendor implements it the same way. Test the parts you depend on.

Sources: Agentic AI Foundation via PR Newswire.