TL;DR: On October 22, 2026, the Agentic AI Foundation added Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face, and OpenSandbox from Alibaba, growing from three projects to 10 in its first year.

Announcement moves AAIF from three projects to 10

The Agentic AI Foundation announced four new hosted projects on October 22, 2026: Agent Skills, contributed by Anthropic; Open Responses, contributed by OpenAI; OpenEnv, contributed by Hugging Face; and OpenSandbox, contributed by Alibaba, according to Unite.AI.

The additions bring AAIF to 10 hosted projects in its first year, up from three launch projects in December 2025, according to Unite.AI.

The announcement coincides with AGNTCon + MCPCon North America, taking place October 22 and 23, 2026, at the San Jose McEnery Convention Center, according to Unite.AI.

The new projects join the Agent2Agent (A2A) protocol, Agent Router, agentgateway, AGENTS.md, goose, and the Model Context Protocol (MCP) as AAIF projects under open, vendor-neutral governance, according to Unite.AI.

According to the foundation, the four projects address different layers of the agentic stack: how agents gain new skills, how applications talk to models across providers, how agents are trained and evaluated, and where agents run safely, as reported by Unite.AI.

AAIF is a Linux Foundation project that provides a vendor-neutral home for agentic AI standards, according to The New Stack as summarized on daily.dev.

“A year ago, AAIF launched with three projects and a bet that agentic AI needed a neutral home to thrive. Today we host 10,” said Manik Surtani, CTO and co-founder of the Agentic AI Foundation, as quoted by Unite.AI.

Surtani said four different companies chose to contribute the projects to neutral governance, giving the community one place to build shared standards, protocols, and infrastructure with no single company controlling the outcome, according to Unite.AI.

The founding projects were MCP, the goose AI agent, and AGENTS.md, according to The New Stack via daily.dev.

AAIF has since added agentgateway and Agent Router before the latest four, according to The New Stack via daily.dev.

The AAIF projects page lists Model Context Protocol as a protocol for seamless integration between LLM applications and external data sources, according to the Agentic AI Foundation projects directory.

The same directory describes Goose as an open source, extensible AI agent that goes beyond code suggestions, according to the Agentic AI Foundation projects directory.

That directory describes AGENTS.md as a simple, open format for guiding coding agents, according to the Agentic AI Foundation projects directory.

It describes agentgateway as enabling secure, scalable connections between AI agents, models, tools, and APIs across ecosystems, according to the Agentic AI Foundation projects directory.

It describes Agent2Agent as an open protocol that enables AI agents to communicate, collaborate, and securely exchange information and tasks across different platforms and frameworks, according to the Agentic AI Foundation projects directory.

It describes Agent Router as an open source AI gateway built on Envoy that connects applications to models and MCP tools while platform teams manage access, routing, and usage, according to the Agentic AI Foundation projects directory.

Agent Skills packages reusable expertise

Agent Skills is an open format for giving agents new capabilities and domain expertise, according to Unite.AI.

Skills package instructions, scripts, and resources in a reusable manner that agents can load on demand, according to Unite.AI.

The AAIF projects directory describes Agent Skills as an open format for reusable agent expertise, according to the Agentic AI Foundation projects directory.

David Soria Parra, co-creator and lead maintainer of MCP at Anthropic, said Anthropic launched Agent Skills a year ago to give developers and enterprises an open standard for making agents more capable, according to Unite.AI.

Parra said Anthropic is donating it to AAIF so that anyone building with skills can shape what comes next, according to Unite.AI.

In a foundation post welcoming the project, AAIF said the specification is open source under Apache 2.0 and CC BY 4.0, according to Unite.AI.

AAIF said more than 70 agent platforms already support it, according to Unite.AI.

AAIF said over 1 million skills have been published with nearly 280 million installations, according to Unite.AI.

The post reports that a million skills were listed on the skills.sh registry in seven months, according to Unite.AI.

Skills are used across software engineering, marketing, design, writing, and research, according to Unite.AI.

The foundation said Agent Skills will be guided under AAIF by a Technical Steering Committee with core maintainers drawn from multiple organizations who serve as individuals, according to Unite.AI.

Open Responses standardizes provider interfaces

Open Responses is an open specification for building multi-provider, interoperable LLM interfaces, based on the OpenAI Responses API, according to Unite.AI.

The AAIF projects directory describes Open Responses as a shared LLM API format for agents across providers, according to the Agentic AI Foundation projects directory.

The foundation said it defines a shared schema, consistent streaming events, and extensible tooling so developers can describe requests and outputs once and run them across model providers with minimal translation, according to Unite.AI.

Open Responses standardizes core LLM primitives like messages and tool calls so developers avoid translating between vendor-specific APIs, according to The New Stack via daily.dev.

Virtually all LLM providers offer the same building blocks but represent them differently in their APIs, forcing provider-specific integration code when switching models, according to The New Stack via daily.dev.

Steve Coffey, a member of technical staff at OpenAI, said the project is joining AAIF to further its mission of helping developers build products on language models through simple, unified interfaces, according to Unite.AI.

A foundation post on the specification describes a common format for the requests and responses exchanged through an API endpoint, covering tool calls and streaming responses, according to Unite.AI.

The specification permits extensions, identified by provider-specific names so features from different implementations do not conflict, according to Unite.AI.

Because it applies to the API, adopting it does not require changing or retraining a model, according to Unite.AI.

The post identifies two existing implementations, according to Unite.AI.

Databricks Model Serving exposes an Open Responses endpoint for Claude, Gemini, and Databricks-hosted open models, translating requests into each provider’s native format, according to Unite.AI.

Vercel AI Gateway exposes an OpenResponses API that gives applications a common endpoint for accessing models through its gateway, according to Unite.AI.

Vercel AI Gateway and Databricks Model Serving already implement the specification, according to The New Stack via daily.dev.

Other supporters named for the specification include AWS Bedrock, Databricks, LM Studio, Ollama, Hugging Face, Nvidia, OpenRouter, Red Hat, Vercel, and open source projects vLLM and OGC, formerly Llama Stack, according to The New Stack via daily.dev.

OpenEnv provides training environments

The foundation describes OpenEnv as an open framework and community hub for creating reinforcement learning environments, according to Unite.AI.

The AAIF projects directory describes OpenEnv as open environments for training AI agents, according to the Agentic AI Foundation projects directory.

It is built around a containerized service that exposes an agent interface, MCP, and the Gym training protocol used by training libraries, according to Unite.AI.

Ben Burtenshaw, a machine learning engineer at Hugging Face, said the committee behind OpenEnv is sharing what its members have built and used at Hugging Face, Microsoft, Reflection, Meta-PyTorch, Fleet, Mercor, RadixArk, Modal, Prime Intellect, and more, according to Unite.AI.

A post by the OpenEnv maintainers describes the architecture: a client starts an episode with reset(), submits actions with step(), and can request state(), according to Unite.AI.

Results include an observation, a numeric reward, episode completion status, and metadata, according to Unite.AI.

Some environments expose tools through MCP, allowing those tools to be used by external agents as well as during training, according to Unite.AI.

The post recounts that Turing built Calendar Gym on OpenEnv, according to Unite.AI.

The team reported roughly 90% task success when tasks named explicit calendar identifiers, compared with roughly 40% when the same tasks used natural-language descriptions, according to Unite.AI.

OpenEnv has a multi-company Technical Committee and a public AAIF proposal seeking a vendor-neutral home for the shared environment specification, according to Unite.AI.

According to the maintainers, roadmap priorities include connecting agent harnesses to training systems, validating environments, providing training recipes, and supporting task sampling, curriculum, and evaluation during training, according to Unite.AI.

Individual RFCs are at different stages of review and implementation rather than a single upcoming release, according to Unite.AI.

OpenSandbox isolates agent execution

OpenSandbox is a general-purpose sandbox platform for AI applications, according to Unite.AI.

The AAIF projects directory describes OpenSandbox as a general purpose sandbox platform built for AI applications, according to the Agentic AI Foundation projects directory.

It offers multi-language software development kits, a unified sandbox API, and Docker and Kubernetes runtimes, according to Unite.AI.

Its uses include coding agents, graphical user interface agents, agent evaluation, AI code execution, and reinforcement learning training, according to Unite.AI.

OpenSandbox gives AI applications isolated environments to execute code, run commands, manage files, and operate browsers or desktops, according to the OpenSandbox repository description surfaced in search results.

Users can start locally with Docker and deploy on Kubernetes through a unified sandbox API, according to the OpenSandbox repository description surfaced in search results.

The project is offered under the Apache License 2.0, according to Unite.AI and repository metadata.

Its sandbox protocol is a collection of OpenAPI specifications with two main parts, according to Unite.AI.

One part is a lifecycle API for creating sandboxes, checking state, pausing, resuming, updating, creating snapshots, and deleting them, according to Unite.AI.

The other part is an execution API for running commands and code, managing files, maintaining sessions, and collecting resource measurements, according to Unite.AI.

An execution daemon named execd runs inside each sandbox and handles those operations, according to Unite.AI.

Docker and Kubernetes runtime backends share the same lifecycle interface, according to Unite.AI.

The Docker backend creates and manages containers directly, while the Kubernetes backend creates workloads and related resources, according to Unite.AI.

“OpenSandbox was born at Alibaba, but we do not want its future decided by one company alone,” said Xiaobin Xu, head of developer infrastructure at Alibaba, as translated and reported by Unite.AI in Japanese and confirmed in English search summaries.

Sources referenced