OpenAI Agent Skills repo catalogs reusable instruction folders, scripts, and resources for Codex

What an Agent Skill is

The README defines Agent Skills as folders of instructions, scripts, and resources that AI agents can discover and use to perform specific tasks. The tagline is "write once, use everywhere," meaning a skill is authored one time and then reused across contexts. Codex uses skills to package capabilities that teams and individuals can apply to complete specific tasks in a repeatable way, and this repository catalogs skills for use and distribution with Codex.

The framing treats a skill as a portable unit of capability rather than a model or a prompt. By bundling instructions, scripts, and resources into a folder, the skill becomes something an agent can locate and run without the user rewriting the logic each time. The repository's job is to hold that catalog so skills can be shared and installed.

The repository is deprecated

The README opens with a clear deprecation notice. It states that this repository is deprecated and that for current Codex skill and plugin examples, users should move to the OpenAI Plugins repository at github.com/openai/plugins. It also points anyone wanting to add their own skills to Codex to the Build plugins guide from developers.openai.com, which includes instructions for creating a skill only plugin.

This matters for anyone who finds the repo through search. The skills it holds may still be readable as examples, but the supported path for new work is the plugins repository. The README keeps the migration instructions at the top so a visitor is not misled into building against a frozen catalog.

How skills are installed in Codex

The README explains that skills in the .system folder are installed automatically in the latest version of Codex. To install curated or experimental skills, you use the $skill-installer command inside Codex. Curated skills install by name and default to the skills/.curated folder, while experimental skills need the skill folder named. After installing a skill, you restart Codex to pick up the new skills.

The install flow for a curated skill looks like this:

$skill-installer gh-address-comments

The README also shows installing an experimental skill by naming the folder, and installing directly from a GitHub directory URL. Each method routes through the same $skill-installer, and the common step after any install is a Codex restart so the new capability is registered.

Where to learn more and how licensing works

The README links three learning resources: using skills in Codex, creating custom skills in Codex, and the Agent Skills open standard at agentskills.io. These point readers to the live documentation and to the cross vendor standard behind the format, so the repo acts as an entry point rather than the full manual.

On licensing, the README states that the license of an individual skill can be found directly inside the skill's directory in the LICENSE.txt file. That keeps licensing per skill instead of one blanket license for the whole catalog, which fits a repo that aggregates work from many authors.

What this repo is and is not

This repository is a catalog and a set of examples, not a runtime or a service. It documents the skill format, shows how Codex discovers and installs skills, and points to the open standard and the newer plugins home. The content is short and procedural, aimed at getting a reader from "what is a skill" to "how do I install one."

It is not a place to file issues about the live Codex product or to find the current set of supported skills, since the deprecation notice redirects both to other locations. A reader should treat the README as a snapshot of the skill concept and an install walkthrough, with the plugins repo as the forward path.

Position in the OpenAI tooling story

Agent Skills sit in OpenAI's broader push to make models callable through packaged, reusable units. The README ties the format to Codex and to an open standard, which suggests the intent is interoperability across agent runtimes rather than a single tool. By moving the catalog to openai/plugins, OpenAI folded skills into a wider plugin model that covers more than instruction folders.

For a developer, the practical takeaway is that the skill pattern, a folder of instructions, scripts, and resources, is the reusable building block, and the install command inside Codex is the delivery mechanism. The deprecated repo preserves the early examples while the active work continues under the plugins banner.

Editorial conclusion

OpenAI Agent Skills repository catalogs folders of instructions, scripts, and resources that AI agents discover and run inside Codex, under the "write once, use everywhere" idea. The repo is deprecated and now points to openai/plugins, with curated and experimental skills installed through the $skill-installer command and a restart of Codex to register them. Licensing lives per skill in each directory LICENSE.txt file. Treat the README as a procedural snapshot of the skill format and an install walkthrough, not as the live home for new work. The catalog still serves as a set of reference examples after the move to plugins, useful for anyone learning the skill format.

Frequently asked questions

Official sources

1. Issues
2. openai/skills on GitHub
3. README