Four companies, four layers of the stack
On October 22, the Agentic AI Foundation (AAIF), part of the Linux Foundation, took in four projects at once at AGNTCon + MCPCon North America in San Jose. In the foundation's framing, each covers one layer of the agent stack:
- Agent Skills (Anthropic): an open format for giving agents new capabilities, packaging instructions, scripts and resources into folders agents load on demand.
- Open Responses (OpenAI): an open specification based on the OpenAI Responses API that defines a shared request schema, streaming events and tool extensions, so the same calling code can switch model providers with little or no translation.
- OpenEnv (Hugging Face): a framework for reinforcement learning environments that exposes both MCP and the Gym interface training libraries use through a containerized service; its committee includes Microsoft, Reflection, Meta-PyTorch and Prime Intellect, among others.
- OpenSandbox (Alibaba): a general-purpose sandbox platform for AI applications with multi-language SDKs, a unified sandbox API and Docker / Kubernetes runtimes, covering coding agents, GUI agents, evaluation and reinforcement learning training.
Together with MCP, goose, AGENTS.md, A2A, agentgateway and Agent Router, that takes AAIF from 3 hosted projects to 10 in its first year.
What Agent Skills handed over
Of the four, Agent Skills has by far the largest install base. In a post the same day, Anthropic said the format grew out of Claude Code, launched in October 2025 and was published as an open standard that December. Since then more than 70 agent platforms have integrated it, including ChatGPT, Cursor, GitHub Copilot and VS Code, with about 1 million skills published and nearly 280 million installs.
What changes hands is the name and logo, the GitHub organization and the agentskills.io domain. Engineers from Anthropic, Google, OpenAI and other companies will maintain the standard as individuals, with David Soria Parra and Mahesh Murag as lead maintainers; you can keep using and contributing to it without joining AAIF. Anthropic's proposed first roadmap covers versioned releases, a formal proposal process and security guidance for people who write and install skills.
What it means for developers
The most direct change is at the skills layer. Until now, whether an Agent Skill worked elsewhere depended on whether each vendor kept up with Anthropic's spec; spec changes now go through a public proposal process. Security guidance making the first roadmap is worth noting too: skills are, at bottom, scripts and instructions an agent will execute, so a skill of unknown origin carries much the same risk as an npm package of unknown origin. If Open Responses is widely adopted, the cost of switching model providers keeps falling, while OpenSandbox and OpenEnv give teams building their own agent sandboxes and training environments neutral, ready-made options.
via: AAIF: Agentic AI Foundation Announces Four New Projects, Anthropic: Donating Agent Skills to the Agentic AI Foundation