Key takeaways
- Agent Skills are portable folders of instructions, scripts, and resources — write once, use everywhere
- Three tiers: system (auto-installed), curated (install by name), and experimental (community contributions)
- Built on the open Agent Skills standard (agentskills.io) — not locked to Codex
- Skills are installed via a $skill-installer command inside Codex and require a restart to activate
FAQ
What are Agent Skills?
Agent Skills are folders of instructions, scripts, and resources that AI agents can discover and use to perform specific tasks in a repeatable way.
How do you install a skill in Codex?
System skills are auto-installed. Curated skills install by name via $skill-installer. Experimental skills require specifying the folder path or GitHub URL.
Is this specific to Codex?
The catalog is designed for Codex, but the underlying Agent Skills format is an open standard at agentskills.io, intended for cross-agent portability.
Executive Summary
OpenAI Skills is the official skills catalog for Codex, OpenAI's coding agent. It defines a portable format — Agent Skills — where each skill is a folder containing instructions, scripts, and resources that an AI agent can discover and execute. The repo has ~9k GitHub stars and serves as both a distribution mechanism and a reference implementation of the open Agent Skills standard.
| Attribute | Value |
|---|---|
| Maintainer | OpenAI |
| GitHub Stars | ~9,000 |
| License | Per-skill (see each skill's LICENSE.txt) |
| Standard | Agent Skills (agentskills.io) |
| Primary Agent | Codex |
How It Works
Skills are organized into three tiers :
- System (
.system/) — Auto-installed in every Codex instance - Curated (
.curated/) — Vetted skills, installable by name - Experimental (
.experimental/) — Community and exploratory skills
Installation uses a built-in $skill-installer command inside Codex. After installing, a restart is required to pick up new skills.
Why It Matters
The "write once, use everywhere" framing and the open standard at agentskills.io signal OpenAI's intent to make skills a portable, ecosystem-level primitive — not just a Codex feature. If the standard gains adoption, it could become the package manager equivalent for agent capabilities.
Competitive Context
This is comparable to:
- Claude Code's CLAUDE.md / project-level instructions
- Cursor's .cursor/rules
- OpenClaw's skills/ directory pattern
The key differentiator is the explicit open standard and tiered distribution model. Most competitors treat agent customization as config files rather than installable, shareable packages.