Anthropic donates Agent Skills to open-source foundation

Giving AI agents real-world business skills is no longer under Anthropic's exclusive control.

Anthropic has officially donated Agent Skills, its open standard for equipping AI agents with specialized workflows, to the Agentic AI Foundation (AAIF) under the Linux Foundation. Along with the standard itself, Anthropic is handing over the brand, logo, GitHub organization, and the agentskills.io domain.

From here on out, the project will be governed as vendor-neutral open source. Engineers from Anthropic, Google, OpenAI, and other tech firms will maintain the standard as individuals, led by maintainers David Soria Parra and Mahesh Murag.

From Claude Code hack to 280 million installs

The whole project started with a practical headache inside Anthropic.

Back when building Claude Code around October 2025, Anthropic noticed that while the model was sharp on general knowledge, it stumbled on practical procedural steps that real business tasks demand. Their fix was dead simple: a basic folder holding a SKILL.md file explaining the task, alongside whatever scripts or reference docs the agent needed.

Developers ran with it immediately. Within weeks, the community published thousands of skills on GitHub and built loaders for Codex, Cursor, and OpenCode. That pushed Anthropic to spin Agent Skills out into a formal open standard in December 2025.

Today, over 70 platforms support it, including heavyweights like ChatGPT, Cursor, GitHub Copilot, and VS Code. The ecosystem has racked up 1 million published skills and nearly 280 million installs.

Moving forward under AAIF, the team plans to roll out formal versioning, a standard proposal process, and much-needed security guidance for authors and users alike.

Why it matters

Anthropic is running the open-source playbook to perfection. While most platform races end up in walled gardens, Anthropic is turning core agent plumbing into public infrastructure. First it was the Model Context Protocol, and now it is Agent Skills.

By surrendering corporate control to the Linux Foundation and getting rivals like OpenAI and Google on board as maintainers, Anthropic ensures this standard becomes the universal default. Competitors will find it almost impossible to sell developers on a proprietary plugin format when one folder already works across Cursor, Copilot, and ChatGPT.

For builders, this is a massive win. You can package your team's internal workflows or deploy routines once without having to rewrite them every time a new coding tool takes over the hype cycle.

That said, the security stakes are high. When near 280 million installs involve running scripts inside developer environments, supply-chain vulnerabilities and prompt injection become immediate threats. The incoming governance team will need to prove they can police malicious skills without ruining the raw simplicity that made the standard take off in the first place.