The quiet story here isn't the skills themselves — it's that a major model vendor adopted a portable, cross-vendor format. This repo packages "Agent Skills" (folders of instructions, scripts, and resources tied to the open standard at agentskills.io) for Codex, signaling that model-agnostic skills are becoming shared infrastructure rather than one company's proprietary feature. It has since been deprecated in favor of OpenAI's plugins model, which makes it a clean snapshot of a format mid-transition.
What Sets It Apart
- Three tiers encode trust, not just maturity:
.systemskills ship and auto-install with the latest Codex;.curatedare vetted and curated by OpenAI, installed on demand by name via$skill-installer;.experimentalare early-stage, installed by specifying the folder path. You can judge provenance before running someone else's scripts. - A skill is just a folder — "write once, use everywhere" — with no bespoke plugin API, which is precisely why the same skill can travel across any agent that honors the standard.
- Licensing lives in a per-skill
LICENSE.txt, tracked at the unit of reuse rather than the whole repo — the detail that matters the moment you lift one skill into a commercial workflow.
Who It's For
Great fit if you want to study how OpenAI structured agent skills for Codex, or to salvage individual .curated/.experimental skills as reference implementations. Look elsewhere if you need a maintained, install-and-go source: the catalog is deprecated, and OpenAI now points to its Plugins repository and "Build plugins" guide. Treat this as an archive to learn from, not a dependency to build on.
