AAIF Adds Four Open-Source Agent Projects, Reaching Ten Hosted Projects in a Year

The Agentic AI Foundation (AAIF), a Linux Foundation project, said on 22 October 2026 that it had added four new open-source projects, bringing its total to ten hosted projects in its first year, according to the foundation's announcement. The additions are Agent Skills, contributed by Anthropic; Open Responses, contributed by OpenAI; OpenEnv, contributed by Hugging Face; and OpenSandbox, contributed by Alibaba.

AAIF was formed on 9 December 2025 by the Linux Foundation, anchored by three founding contributions: Anthropic's Model Context Protocol (MCP), Block's goose, and OpenAI's AGENTS.md, according to the Linux Foundation's formation announcement. Its stated purpose is to give agentic AI infrastructure "a neutral, open foundation" so it develops "transparently, collaboratively", per that same announcement.

Each of the four new projects sits at a different layer of what the foundation calls the agentic stack: how an agent acquires a new capability, how an application talks to multiple model providers through one interface, how an agent is trained and evaluated, and where an agent runs its code safely, according to the AAIF's project announcement.

What each project does

Agent Skills is an open format that packages instructions, scripts and resources so an agent can load a capability or domain expertise on demand, designed to work across different agent platforms, according to the AAIF announcement. In its own post on the donation, dated 22 October 2026, Anthropic says that since the format's publication in December 2025, "1 million skills have been published and nearly 280 million installed" across "more than 70 agent platforms", naming ChatGPT, Cursor, GitHub Copilot and VS Code as examples. Anthropic does not describe how those totals are counted, so it is not possible to tell from the post how many of the published skills are in active use, or whether one skill installed on several machines is counted more than once.

Open Responses, contributed by OpenAI, is a specification for building LLM interfaces that work the same way across multiple model providers: it defines a shared request-and-response schema, consistent streaming events and extensible tooling so a developer can write an integration once rather than per provider, according to the AAIF announcement.

OpenEnv, contributed by Hugging Face, is a framework for building the reinforcement-learning environments used to train and evaluate agents; the AAIF describes it in the same announcement as interoperable and plug-and-play, built on containerized services that support both the Model Context Protocol and the Gym training interface.

OpenSandbox, contributed by Alibaba, provides multi-language SDKs and unified sandbox APIs, with runtimes for Docker and Kubernetes, aimed at isolating the code that coding agents and GUI agents execute, per the same announcement.

Why shared governance matters to builders

For students, researchers and engineers building or studying agents, the practical change is less about any single tool than about who controls it. MCP, AGENTS.md, Agent Skills, Open Responses, OpenEnv and OpenSandbox were each developed inside one company before moving into AAIF's governance, which the foundation's chief technology officer, Manik Surtani, described in the announcement as a bet made at the foundation's December 2025 launch that "agentic AI needed a neutral home." Anthropic's David Soria Parra, co-creator and lead maintainer of MCP, said in the same release that the company is "committed to continuing to invest in open source agentic tooling for builders."

That distinction matters for anyone building coursework, research tooling or production agents on top of these formats: a specification governed by a multi-company foundation is harder for any one of its contributors to change unilaterally than one owned by a single lab, so code written against MCP, Agent Skills or Open Responses is intended to keep working as the underlying models change. Researchers training or evaluating agents get a separate, practical benefit from OpenEnv and OpenSandbox: shared environment and sandbox specifications make it easier to reproduce another group's training or evaluation setup instead of rebuilding it from a paper's description alone.

What the numbers don't yet show

The scale figures in this story come from the organisations themselves. AAIF's own count of ten hosted projects and Anthropic's count of 1 million published skills and nearly 280 million installations have not been accompanied by published underlying data or a counting method, and no independent audit of either figure exists yet. The independent coverage of the announcement published so far largely restates the foundation's own figures rather than testing them.

Membership growth is on separate, somewhat firmer footing: the Linux Foundation has reported that AAIF passed 300 member organisations within less than a year of its December 2025 launch, in a separate release. Whether the four newly added projects see adoption comparable to MCP and AGENTS.md will depend on take-up by agent platforms and model providers beyond the four companies that contributed them.