Best Codex skills: where to find them and which to install first
A shortlist of Codex skills by job from OpenAI’s own catalog, where Codex skills come from, how to install them and how many to keep. Read 21 November 2026.
- For
- Codex users choosing skills to install
- Written for
- Codex
- Time
- 7 min read · updated Nov 21, 2026
The short answer
There is no ranking of Codex skills to trust, because nobody publishes install counts or ratings for them, so this shortlist is by job. OpenAI’s own catalog has 39 curated skills, and Codex ships 5 more built in. Install a curated one with $skill-installer <name> and restart Codex. Good first picks are gh-fix-ci and gh-address-comments for pull requests, playwright for browser work, openai-docs for current documentation, and one deploy skill for the host you use. Keep the list short: Codex only reads skill descriptions up to about 2% of its context window.
Searching for the best Codex skills turns up lists of everything, and no list can say which skills are best, because there is no download count or rating to sort by. What can be checked is what exists, where it comes from and what each one does. This page uses OpenAI’s documentation and its skills repository, read on 21 November 2026, and groups skills by the job you want done. The picks are ours; the descriptions are from each skill’s own file.
Where Codex skills come from
The catalog notice matters if you are comparing lists: the curated folder is still there and $skill-installer still installs from it, but OpenAI’s own README says to look at its plugins repository for current examples. Agent skills marketplaces compares how Codex and the other agents add catalogs, and Claude skills on GitHub covers public repositories, since the folder format is the same.
A shortlist by job
These come from the 39 curated skills. Each description is a short version of the skill’s own description line, which is the text Codex uses to decide when to run it. Several need an account, a command-line tool or an API key of their own; the description says which.
The curated folder also holds screenshot, migrate-to-codex (which moves instruction files, skills, agents and MCP config into Codex) and a set of Figma skills. If you are choosing skills that act on outside services, read what they will touch first. Safety checks before running a skill is the short list.
Install one
Inside Codex, run $skill-installer with the skill’s name. OpenAI’s README gives the same for a curated skill and a form for a skill in the experimental folder or at a GitHub directory URL. Restart Codex afterwards so it picks up the new skill.
To see what you have, run /skills in the Codex CLI or the IDE extension, or type $ to pick one. To call a skill yourself, put it in the prompt. Codex can also choose a skill when your task matches its description, unless the skill sets allow_implicit_invocation: false.
Keep the list short
OpenAI’s documentation says the initial skills list Codex reads is limited to at most 2% of the model’s context window, or 8,000 characters when the window is unknown. With a long list, descriptions get shortened or dropped, and Codex may stop choosing a skill on its own. skills.max_context_tokens raises the limit, up to 10,000 tokens. The simpler fix is to switch off the skills you don’t use.
Each entry takes one selector, path or name, and when entries overlap the last one wins. Codex looks for skills in .agents/skills from the folder where you launch it up to the repository root, then $HOME/.agents/skills, then /etc/codex/skills, then the skills built into Codex. Where skills live has the folder for every agent, and the Codex page covers the problems people hit.
Skills that make a finished file
Most of the skills above change how Codex works with code and services. Lookbk’s skills make something you can open: a title video, a slide deck, a link-preview image, a landing page. Each skill page shows what it rendered, and the installer writes to ~/.agents/skills, the folder Codex reads. These four are free.
Run the installer with the last part of a skill page’s address, restart Codex, then name the skill with $.
Common questions
- What are the best Codex skills?
- No ranking exists to rely on, because there are no public install counts or ratings. For pull requests start with
gh-fix-ciandgh-address-comments; for browser workplaywright; for current OpenAI documentationopenai-docs; and the deploy skill for your host. - How do I install a Codex skill?
- Run
$skill-installer <name>inside Codex for a curated skill, or copy a skill folder into.agents/skillsin a repository or~/.agents/skillsat home. Restart Codex afterwards. - How do I list my Codex skills?
- Run
/skillsin the Codex CLI or the IDE extension, or type$to pick one. - Is there a Codex skills marketplace?
- OpenAI’s documentation says plugins distribute skills, through a plugin directory that ChatGPT and Codex share, and its
openai/pluginsrepository holds Codex plugin examples. The olderopenai/skillsrepository still has a curated folder, but its README says it is deprecated. - Are there awesome lists for Codex skills?
- Codex reads the same
SKILL.mdfolders as other agents, so public lists of skills for those agents work too. Check each skill’s license and read it before you install. - Why does Codex not use one of my skills?
- OpenAI’s documentation says the initial skills list is limited to about 2% of the context window, so with many skills some descriptions are shortened or dropped. Name the skill with
$, or switch off skills you don’t use. - How do I turn a Codex skill off without deleting it?
- Add a
[[skills.config]]entry to~/.codex/config.tomlwith the skill’spathornameandenabled = false, then restart Codex.
Sources
What this guide says about other companies’ tools comes from their own documentation, checked on Nov 21, 2026.
Did this guide help?
Read next
All guidesWhere skills live: folders for every agent
The folders Claude Code, Codex, Cursor and other agents read skills from, which copy wins when two share a name, and what cloud agents can see.
Install a skill in any agent
Install a Lookbk skill in Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI or OpenCode: where each looks, how to check it loaded and how to run it.
Agent skills marketplaces: how each agent adds a catalog
There is no one skills marketplace for every agent. How Claude Code, Cursor, OpenCode, Gemini CLI, Codex and Copilot each add a catalog, from their own docs.
Claude skills on GitHub: main repos and how to use them
The GitHub repositories that hold Claude skills, checked on the GitHub API: license, last commit, skill count. Then how to copy one into your agent.
Safety checks before running a skill
A skill is instructions and code your agent may run. Six checks to make first: its source, its commands, its network use and what it may do without asking.
See a skill’s result before you install it.
Every skill page shows what it makes, what it needs and which agents can run it.



