Last Thursday, Anthropic handed Agent Skills to the Agentic AI Foundation, the Linux Foundation project that already looks after the Model Context Protocol. On paper it's a governance move. In practice, the agent skills standard is turning into something more ordinary and more useful: a shared file format for the way a team does its work.
A skill is just a folder. It holds a SKILL.md file with instructions for a specific task, plus any scripts or reference files the task needs. According to Anthropic's announcement, more than 70 agent platforms now support the format, including ChatGPT, Cursor, GitHub Copilot and VS Code, and about 1 million skills have been published with nearly 280 million installs. Engineers from several companies will maintain it as individuals, and the first proposed roadmap covers versioned releases, a formal proposal process and security guidance for people who write and install skills.
Why a neutral agent skills standard matters
The skill itself isn't the interesting part. The interesting part is what goes into it. Most useful agent work depends on procedure: how your team closes the month, how you triage a support ticket, which checks run before a release. That knowledge usually lives in a few people's heads or a wiki nobody reads. Written as a skill, it becomes something an agent can follow and a colleague can review, and with a neutral standard it isn't tied to one vendor's tool. The Linux Foundation's announcement placed Agent Skills alongside three other new projects, Open Responses from OpenAI, OpenEnv from Hugging Face and OpenSandbox from Alibaba, which together cover how agents learn tasks, talk to models, get evaluated and run safely.
Portable doesn't mean identical, though. As Search Engine Watch noted, the experimental field that lists which tools a skill may use is supported differently across products, so a procedure that runs cleanly in one agent can need different permissions in another. A skill that bundles a script also needs an environment that can actually run it.
That points to the practical habit worth building now. Treat skills the way you treat software dependencies. Keep your own in version control, have someone read a third-party skill before it's installed, and check what it's allowed to touch. The format will keep changing, but a team that has written down how it works, in a form both people and agents can check, is in a better position whichever tool it uses next year.