Are AI agent skills tied to one product?

Skills were originally developed by Anthropic, but released as an open, freely implementable format rather than kept as a single-vendor feature, and a growing list of agent products has adopted them since.^[1] The format itself is deliberately plain — a folder with a SKILL.md file at its root, metadata plus instructions, optionally bundling scripts, references, and assets — which is exactly what makes it easy for other products to implement on their own, without needing Anthropic’s cooperation or a shared runtime.^[1]

That breadth is the point. Cursor supports skills in its editor,^[2] GitHub Copilot supports them in VS Code,^[3] and the same open listing shows adoption spanning Google’s Gemini CLI and OpenAI’s Codex — four separate model vendors’ tools reading the same folder format, not a feature bolted onto one product.^[1] The practical caveat is that there is no automatic syncing between products: nothing pushes a skill you wrote for one agent into another’s library. A skill written for one may need small adjustments — a different bundled-script convention, a different way the host discovers the folder — to run cleanly in another.

The adopter list runs well past the four named above. The standard’s own site lists more than forty agent products under its open banner, spanning coding tools like Goose, OpenCode, and VS Code as well as agents built for entirely different jobs — Databricks’ Genie Code and Snowflake’s Cortex Code for data work, Pulumi’s Neo for cloud infrastructure — all reading the same SKILL.md folder.^[1] Governance is public too: the specification lives in an open GitHub repository with a public Discord for discussion, so a change to the format goes through the same pull-request process any other open-source project uses rather than one vendor’s private roadmap.^[1]