Anthropic gave away Agent Skills, OpenAI gave away its Responses shape, and they now sit in the same foundation
The Agentic AI Foundation under the Linux Foundation took in four new projects on October 22. Agent Skills comes from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face and OpenSandbox from Alibaba. The foundation goes from 3 projects to 10, next to MCP, A2A, AGENTS.md, goose, agentgateway and Agent Router, and it passed 300 members the same day. Link: https://www.linuxfoundation.org/press/agentic-ai-foundation-announces-four-new-projects-agent-skills-open-responses-openenv-and-opensandbox
The numbers on Skills explain why Anthropic let go of it. More than 70 agent platforms already read the format, including ChatGPT, Cursor, GitHub Copilot and VS Code, and the foundation counts about 1 million skills published and close to 280 million installs. Anthropic hands over the name, the logo, the GitHub org and agentskills.io. The first items on the proposed roadmap are versioned releases, a formal proposal process and security guidance for people who write and install skills. That last one is the tell. A format with 280 million installs and no signing story is a supply chain, and everyone involved knows it. Anthropic's post: https://www.anthropic.com/news/donating-agent-skills-to-the-agentic-ai-foundation
The quieter headline is OpenAI. Open Responses is a cross-provider request schema based on the OpenAI Responses API, and OpenAI put it into a foundation that Anthropic co-runs. OpenEnv packages RL environments as containers that speak both MCP and Gym. OpenSandbox ships Docker and Kubernetes runtimes plus a lifecycle protocol for where agents run. Read together, the four cover how an agent learns a skill, how it talks to any model, how it gets trained and graded, and where it is allowed to execute.
A year ago each of those layers was a moat somebody was trying to own. Now the labs are competing on the model and the harness and handing the plumbing to a neutral body, which is what happened to HTTP and Kubernetes once the fight moved up the stack. The thing to watch is the security guidance for skills. If it ends up as signed, versioned packages with a review process, skills become a real package ecosystem. If it stays a document, they stay the easiest way to slip instructions into an agent.
← Back to all articles
The numbers on Skills explain why Anthropic let go of it. More than 70 agent platforms already read the format, including ChatGPT, Cursor, GitHub Copilot and VS Code, and the foundation counts about 1 million skills published and close to 280 million installs. Anthropic hands over the name, the logo, the GitHub org and agentskills.io. The first items on the proposed roadmap are versioned releases, a formal proposal process and security guidance for people who write and install skills. That last one is the tell. A format with 280 million installs and no signing story is a supply chain, and everyone involved knows it. Anthropic's post: https://www.anthropic.com/news/donating-agent-skills-to-the-agentic-ai-foundation
The quieter headline is OpenAI. Open Responses is a cross-provider request schema based on the OpenAI Responses API, and OpenAI put it into a foundation that Anthropic co-runs. OpenEnv packages RL environments as containers that speak both MCP and Gym. OpenSandbox ships Docker and Kubernetes runtimes plus a lifecycle protocol for where agents run. Read together, the four cover how an agent learns a skill, how it talks to any model, how it gets trained and graded, and where it is allowed to execute.
A year ago each of those layers was a moat somebody was trying to own. Now the labs are competing on the model and the harness and handing the plumbing to a neutral body, which is what happened to HTTP and Kubernetes once the fight moved up the stack. The thing to watch is the security guidance for skills. If it ends up as signed, versioned packages with a review process, skills become a real package ecosystem. If it stays a document, they stay the easiest way to slip instructions into an agent.
Comments