Anthropic Gave Away a Standard With 280 Million Installs. Here's What 'Agent Skills' Actually Does
A 'skill' is plainer than it sounds: a folder with a SKILL.md file that spells out how to do one specific job, plus whatever scripts or reference material the task needs. Anthropic built the format for Claude Code because a general-purpose coding agent is good at reasoning and bad at remembering your team's specific deploy checklist or API quirks. A skill is how you hand it that checklist once, instead of re-explaining it in every conversation.
Developers started publishing skills and writing loaders for other agents almost immediately after the October 2025 launch. Anthropic published the format as an open standard that December, still under its own GitHub org and domain. Handing it to a neutral foundation is the next step in that same direction: taking Anthropic's name off a format other companies' products now depend on.
Who actually runs it now
| Thing | Detail |
|---|---|
| Home | Agentic AI Foundation, a directed fund under the Linux Foundation |
| Co-founders of AAIF | Anthropic, OpenAI, and Block |
| Lead maintainers | David Soria Parra and Mahesh Murag, as individuals, not as Anthropic employees |
| Other maintainers | Engineers from Anthropic, Google, OpenAI, and other companies, serving as individuals |
| Also donated to AAIF | The Model Context Protocol (MCP), separately, by Anthropic |
Anthropic's stated reasoning is that it wants Agent Skills to stay 'open, community-driven, and vendor-neutral.' A format one company owns is a format that company can change the rules on later; a format a neutral foundation owns is harder for any single vendor to steer to its own advantage, at least in theory. It's the same logic behind handing Kubernetes to the Cloud Native Computing Foundation a decade ago, or MCP to this same AAIF earlier.
What actually changes if you use ChatGPT, Cursor, or Claude Code
- ✦Nothing is required of you. Existing skills keep working, and you can keep writing or installing them exactly as before.
- ✦The roadmap gets more formal. Anthropic says AAIF plans versioned releases, a real proposal process for spec changes, and security guidance for people writing and installing skills, something a single-company project hadn't needed until adoption got this wide.
- ✦Governance opens up, slightly. Feedback on the specification now goes through the project's public Discord and GitHub rather than directly to Anthropic.
- ✦Skill security is worth a second look. A skill can bundle executable scripts. As the catalog scales past a million published skills, treat an unfamiliar one the way you'd treat an unfamiliar browser extension: check what it actually runs before installing it.
Is 'open standard' the same as 'independent'
Worth being honest about: AAIF's founding members are Anthropic, OpenAI, and Block, three companies with every incentive to keep agent tooling interoperable with their own products. That's not the same as a genuinely independent standards body, and reporting on the broader push to open up agent tooling has flagged adoption claims as worth treating with some caution since several come from vendor-adjacent sources rather than neutral audits, per coverage of the standard's spread across coding tools. What's real is that Anthropic no longer unilaterally controls the format's future. What's unproven is whether that actually produces different outcomes than Anthropic running it alone would have.
What is an Agent Skill, in plain terms?
A folder containing a SKILL.md file that teaches an AI agent how to do one specific task, often with supporting scripts or reference documents. It's a way to give an agent a standing procedure instead of re-explaining the same steps every time.
Who owns Agent Skills now?
The Agentic AI Foundation, a project under the Linux Foundation that Anthropic co-founded along with OpenAI and Block. Anthropic donated the standard's name, logo, GitHub organization, and domain to it on October 22, 2026.
Do I need to do anything if I already use Claude Code, Cursor, or GitHub Copilot?
No. Existing skills and integrations keep working the same way. The donation changes who governs the standard's future, not how it works today.
How widely adopted is the standard?
Anthropic says more than 70 agent platforms have integrated it, including ChatGPT, Cursor, GitHub Copilot, and VS Code, with over 1 million skills published and nearly 280 million installs as of the October 2026 announcement.
Is a skill safe to install from an unfamiliar source?
Treat it like a browser extension. A skill can bundle executable scripts, and the catalog is large and largely unaudited. Check what a skill actually runs before installing it from anyone you don't already trust.
Try the tools ✦
Free browser tools that never upload your files.