Agent skills marketplaces › Codex skills
Codex skills: where Codex finds them, where to get more, and how to share them
Codex loads Agent Skills from .agents/skills/ in your repository (every folder from where you start Codex up to the repo root), from $HOME/.agents/skills/ and /etc/codex/skills/, plus two built-ins, $skill-creator and $skill-installer. Get more with $skill-installer, from the plugin directory Codex shares with ChatGPT (/plugins in Codex CLI), or with npx skills add owner/repo -a codex. Type $ or /skills to use one.
Where Codex looks
| Scope | Folder | Use it for |
|---|---|---|
| Repo | $CWD/.agents/skills/, each parent folder's .agents/skills/, and $REPO_ROOT/.agents/skills/ | Skills for one module (in its folder) or the whole repository (at the root), checked in so the team gets them |
| User | $HOME/.agents/skills/ | Your own skills, in every repository |
| Admin | /etc/codex/skills/ | Defaults for everyone on a machine or container, and automation |
| System | Bundled with Codex | skill-creator and skill-installer, on by default |
Codex follows symlinked skill folders, and if two skills share a name it shows both rather than merging them. Standalone skills work in Codex CLI, the IDE extension and the ChatGPT desktop app. Plugins, which carry skills too, work in Codex CLI and the desktop app but not in the IDE extension. Claude Code reads .claude/skills/ instead; the folders for every agent are in How to install agent skills.
Where to get Codex skills
| Source | What's there | How you install |
|---|---|---|
$skill-installer | OpenAI's curated skills, such as linear, gh-fix-ci and pdf; it can also fetch skills from other repositories | $skill-installer linear inside Codex |
| OpenAI's plugin directory | One catalog for ChatGPT and Codex, in tabs: OpenAI, your workspace, and Personal. The openai/plugins repo held 62 plugin folders on November 13, from Figma, Linear, Notion, Sentry and Stripe to Superpowers | /plugins in Codex CLI, or the Plugins tab in the ChatGPT desktop app; start a new session after |
| Any plugin marketplace | A git repo with .agents/plugins/marketplace.json; the desktop app also reads a repo's .claude-plugin/marketplace.json | codex plugin marketplace add owner/repo (add --ref to pin), then install from /plugins |
| skills.sh | Any public skill on GitHub, ranked by installs | npx skills add owner/repo -a codex writes to .agents/skills/, or ~/.agents/skills/ with -g |
| Tessl | Skills with eval scores and Snyk findings | npx tessl i github:owner/repo --skill name |
Claude-first repos (anthropics/skills and others) | The same SKILL.md format | Copy or symlink the folder into .agents/skills/ |
The old openai/skills repository is marked deprecated in favor of openai/plugins, though OpenAI's docs still link its curated skills as examples. For the wider market (Anthropic's catalogs, graded directories, paid skills), see the full marketplace comparison.
Using a skill
- By name: type
$and pick the skill, or run/skills. - Automatically: Codex picks a skill when your task matches its
description, so a skill with a vague description rarely fires. - Making one:
$skill-creatorasks what the skill does and when it should trigger, and writes the folder. By hand, a SKILL.md needsnameanddescriptionin its frontmatter. - Not showing up: Codex detects changes on its own; restart it if a new or edited skill doesn't appear.
How many skills Codex can list
At the start of a session Codex shows the model each skill's name, description and path, and that list gets at most 2% of the context window (8,000 characters when the window size is unknown). With many skills installed, Codex shortens descriptions first and may then leave skills out, with a warning. skills.max_context_tokens in config.toml raises the budget to as much as 10,000 tokens. Claude Code works the same way with a 1% budget. The practical rule is the same in both: keep the skills you use, and remove the rest.
Switching skills off
# ~/.codex/config.toml: disable one skill without deleting it
[[skills.config]]
path = "/path/to/skill/SKILL.md"
enabled = false
# a project's trusted .codex/config.toml: no bundled system skills here
[skills.bundled]
enabled = false
Each [[skills.config]] entry takes one selector: an absolute path for one skill, or a name for every skill with that name. Restart Codex after a change. To keep a skill available by $name but stop Codex choosing it on its own, add agents/openai.yaml to the skill with policy: allow_implicit_invocation: false; the same file sets its display name, icon and MCP dependencies.
Sharing Codex skills with a team
- In the repo: commit
.agents/skills/. Copilot and Cursor read the same folder. - A repo marketplace:
$REPO_ROOT/.agents/plugins/marketplace.jsonmakes plugins discoverable, and the repo's.codex/config.tomlturns each one on with[plugins."name@marketplace"] enabled = true. - A ChatGPT workspace: on Business, Enterprise and Edu, admins import and sync a GitHub marketplace and set each plugin to installed by default or available.
The trade-offs, across Codex, Claude Code and Cursor, are in Sharing skills across a team. Whatever the route, read a skill's scripts first: when a plugin's code runs through Codex, Codex's sandbox and approval policy applies, but the instructions in a SKILL.md can still steer the agent. Are agent skills safe? has a ten-point check.
Zest's Codex plugin, from the company that publishes this site, records which skills, models and tokens each Codex session used, per engineer, next to the pull requests they led to; it isn't a marketplace, and it's free for 3 seats. Zest for Codex
Questions
Where are Codex skills stored?
In .agents/skills/ folders: every one from the folder where you start Codex up to the repository root, then $HOME/.agents/skills/ for you and /etc/codex/skills/ for everyone on the machine. Two built-in skills, $skill-creator and $skill-installer, ship with Codex.
How do I install a skill in Codex?
Run $skill-installer with a skill's name for OpenAI's curated ones, install a plugin with /plugins in Codex CLI or the Plugins tab in the ChatGPT desktop app, or run npx skills add owner/repo -a codex for any skill on GitHub. Codex picks up new skills on its own; restart it if one doesn't appear.
Can Codex use Claude Code skills?
Yes, the format is the same open standard, but Codex doesn't read .claude/skills/. Copy or symlink the skill folder into .agents/skills/ (Codex follows symlinks), and keep the name and description fields, which Codex requires.
What's the difference between a Codex skill and a plugin?
A skill is one folder with a SKILL.md. A plugin packages one or more skills, optionally with MCP servers and other files, so it can be installed from a marketplace. OpenAI's docs say to author with skills and distribute with plugins.
How do I turn off a Codex skill?
Add a [[skills.config]] entry with the skill's path or name and enabled = false to ~/.codex/config.toml, then restart Codex. To stop Codex choosing a skill on its own but keep $name working, set policy.allow_implicit_invocation: false in the skill's agents/openai.yaml.