Agentic AI Foundation adds Agent Skills, Open Responses, OpenEnv and OpenSandbox: what each project is and who donated it

The Agentic AI Foundation's new projects, announced on October 22, 2026, are Agent Skills, Open Responses, OpenEnv and OpenSandbox, contributed by Anthropic, OpenAI, Hugging Face and Alibaba. The Linux Foundation-hosted group, which already looks after the Model Context Protocol, now has 10 projects covering how AI agents get instructions, call models, train and run code. Its first three projects, MCP, goose and AGENTS.md, are covered in our MCP explainer and our AGENTS.md guide.

This story is based on the Linux Foundation's October 22 announcement, Anthropic's post on donating Agent Skills, the AAIF's project, board and technical committee pages, the projects' GitHub repositories and sites, and the AAIF's December 9, 2025 launch announcement, checked October 25, 2026.

The short version

• Four new projects: Agent Skills (Anthropic), Open Responses (OpenAI), OpenEnv (Hugging Face) and OpenSandbox (Alibaba).
• Ten in total: the AAIF launched with three on December 9, 2025, and also lists Agent2Agent (A2A), Agent Router and agentgateway.
• Timing: announced at AGNTCon + MCPCon North America in San Jose, October 22–23.
• Biggest adoption claim: Anthropic says more than 70 agent platforms have integrated Agent Skills. That's Anthropic's figure, not an independent count.
• Governance: a governing board chaired by AWS's David Nalley and a technical committee chaired by Anthropic's David Soria Parra.

What are the four new projects?

Last verified: October 25, 2026

What is Agent Skills, and what changes now that Anthropic has donated it?

A skill is a folder with a SKILL.md file that explains a task, plus any scripts or reference files, which an agent reads only when a task needs it. Anthropic built skills for Claude Code, launched them in October 2025 and published Agent Skills as an open standard in December 2025. Google's Gemini also uses the same SKILL.md packaging for its own skills, as our Gemini skills story explains.

Anthropic says more than 70 agent platforms, "including ChatGPT, Cursor, GitHub Copilot, and VS Code," have integrated the standard, and that "1 million skills have been published and nearly 280 million installed." It's donating the Agent Skills name and logo, the GitHub organization and the agentskills.io domain. Engineers from Anthropic, Google, OpenAI and other companies will maintain it as individuals, with David Soria Parra and Mahesh Murag as lead maintainers. Proposed next steps are versioned releases, a formal proposal process and security guidance for people who write and install skills. That last item matters: a skill can carry scripts that an agent runs.

What is Open Responses?

Open Responses is "an open specification for multi-provider, interoperable LLM interfaces" based on the OpenAI Responses API, according to the Linux Foundation. It defines a shared request and response schema, consistent streaming events and tool-calling patterns, so one integration can work across model providers and gateways. It launched on January 15, 2026, and its latest revision, dated April 24, 2026, added a WebSocket transport and a compaction endpoint, according to its changelog. OpenAI's Steve Coffey said the project wants "simple, clean, and unified interfaces." OpenAI's own product built on the Responses API is covered in our Agents API story.

What are OpenEnv and OpenSandbox?

OpenEnv gives agent developers a standard way to package training environments. Its repository describes one API (reset, step and state) for environments shipped as Docker images, more than 40 ready-made environments, and integrations with training libraries such as TRL and Unsloth. Meta's PyTorch team and Hugging Face launched it on October 23, 2025; the Linux Foundation lists Hugging Face as the contributor, and the repository says a technical committee including both governs it. The README says it's in early development and its APIs may change.

OpenSandbox is Alibaba's sandbox platform for running agent code. It defines a sandbox protocol for creating and managing environments, with SDKs in Python, Java/Kotlin, TypeScript, C# and Go, plus Docker and Kubernetes runtimes. Alibaba's Xiaobin Xu said "we don't want its future to be defined by a single company." It joins a growing set of open sandboxes for agents, including NVIDIA's runtime covered in our OpenShell explainer.

Who runs the Agentic AI Foundation?

The AAIF was announced on December 9, 2025, with Anthropic, Block and OpenAI as co-founders and AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI as platinum members. Its site lists two bodies:

• Governing board: 10 members listed, chaired by David Nalley of AWS, with representatives from Google, Microsoft, OpenAI, Cloudflare, Anthropic, Bloomberg, Block, Shopify and the Gates Foundation.
• Technical committee: chaired by David Soria Parra of Anthropic, with members from Microsoft, Block, OpenAI, Google, Cloudflare, Bloomberg and AWS.

The AAIF's pages don't describe how projects are accepted or graduate. One point worth watching is concentration: Soria Parra chairs the technical committee and is a lead maintainer of both MCP and Agent Skills. Anthropic says Agent Skills will stay "open, community-driven, and vendor-neutral," and that anyone can use or contribute to it without joining the foundation.

What's confirmed, reported and uncertain

• Confirmed (Linux Foundation, AAIF, Anthropic): the four projects and their contributors, the 10-project total, the board and committee chairs, and the terms of the Agent Skills donation.
• Vendor-reported: the 70-plus platforms, 1 million skills published and nearly 280 million installs (Anthropic).
• Uncertain: how the AAIF admits and graduates projects, and whether Open Responses will be adopted by providers that compete with OpenAI.