Agentic AI Foundation adds four projects from Anthropic, OpenAI, Hugging Face and Alibaba

The Agentic AI Foundation adds four tools to its hosted open-source portfolio: Anthropic's Agent Skills, OpenAI's Open Responses, Hugging Face's OpenEnv and Alibaba's OpenSandbox. Announced October 22 at AGNTCon + MCPCon North America in San Jose, the additions bring the foundation's roster to 10 projects, from three at its December 2025 launch.

The projects span distinct parts of the agent-development stack. AAIF describes that split as reusable agent capabilities, application interfaces for language models, environments for training and evaluation, and sandboxed execution. The announcement is a governance move, not evidence that the projects work together as one product or that their stated technical benefits have been independently tested.

What are the four new Agentic AI Foundation projects?

• Agent Skills, contributed by Anthropic, is an open format for reusable agent expertise. It packages instructions, scripts and other resources that an agent can load when needed.

• Open Responses, contributed by OpenAI, is an open specification for multi-provider LLM interfaces. Based on OpenAI's Responses API, it sets out a shared schema, streaming events and extensible tooling intended to let applications describe model requests and outputs in a common format.

• OpenEnv, contributed by Hugging Face, is an open framework and community hub for reinforcement-learning environments. AAIF says it uses a containerized service that exposes an agent interface, the Model Context Protocol and Gym, a training protocol used by training libraries.

• OpenSandbox, contributed by Alibaba, is a general-purpose sandbox platform for AI applications. It offers multi-language SDKs, unified sandbox APIs, and Docker and Kubernetes runtimes for uses including coding agents, agent evaluation, AI code execution and reinforcement-learning training.

The four additions join Agent2Agent, Agent Router, agentgateway, AGENTS.md, goose and the Model Context Protocol, according to AAIF. That list now covers protocols, gateways, agent instructions and development tools alongside the newly added projects.

AAIF was formed under the Linux Foundation in December 2025. Its launch contributions were Anthropic's Model Context Protocol, Block's goose and OpenAI's AGENTS.md. OpenAI said at the time that the foundation was intended to provide neutral stewardship for open, interoperable agent infrastructure. AAIF similarly characterizes its current portfolio as operating under open, vendor-neutral governance.

For engineering teams, the practical distinction is where each project sits: Agent Skills concerns reusable context and task knowledge; Open Responses concerns application-to-model API formats; OpenEnv concerns training environments; and OpenSandbox concerns the sandbox layer for AI applications. Whether these specifications gain broad support across vendors and production deployments remains separate from their transfer to the foundation.

This story draws on original reporting from SiliconANGLE.