The best skills for Codex CLI
Codex writes code fast, but out of the box it ignores your conventions and makes you repeat the same explanations every session. Skills fix that: they are reusable instruction folders that Codex loads at startup and that you call by name. This guide picks the best Codex-compatible skills, sorted by use case, and shows exactly where to drop them and how to trigger them. You will also see the difference between a skill and the AGENTS.md file, which are often confused.
Key takeaways
- A Codex skill is a folder with a SKILL.md file: reusable instructions for one precise task.
- You drop it in .agents/skills (repository) or $HOME/.agents/skills (personal), and call it with $name.
- AGENTS.md carries the project's always-on rules; a skill loads on demand. The two complement each other.
- The most useful skills frame your stack, secure your tests, hold long jobs together and plug in your tools.
What is a Codex skill and where should you put it?
A Codex skill is simply a folder containing a SKILL.md file. That file starts with two pieces of information, a name and a description explaining when to use it, then gives the instructions to follow. Codex reads these folders at startup and knows what to do when a task matches.
Where you put it depends on the scope you want. For a skill shared with your team, create an .agents/skills folder at the root of the repository: it travels with the project and benefits everyone. For your personal skills, available across all your projects, use $HOME/.agents/skills. Careful: several tutorials still point to a .codex/skills folder, which is not the path the official documentation uses.
To trigger a skill, type $ followed by its name in Codex. You can also let Codex pick it itself: if your request matches the skill's description, it activates without you naming it. OpenAI also publishes an official skills catalogue you can install straight from Codex.
Skill or AGENTS.md: which one, and when?
These two mechanisms look alike but serve different purposes. AGENTS.md holds your repository's always-on rules: the command to install the project, the one to run the tests, your coding conventions, what must never be touched. Codex reads them systematically, every session.
A skill is the opposite: a capability loaded on demand, for one repeatable, precise task. Writing tests in your style, preparing a database migration, auditing an interface. You only pay the context cost when you need it.
The simple rule: what applies to the whole project, all the time goes in AGENTS.md; what describes a task you repeat becomes a skill. The two stack up, and that is often the right combination.
Which skills make Codex respect your conventions?
The first move is to give Codex the patterns of your stack. A back-end patterns skill sets the architecture, API design and data access. A front-end patterns skill holds your React or Next.js components and interface structure. A PostgreSQL skill recalls schema and query best practices.
The payoff is immediate: instead of producing generic code learned from the internet, Codex writes in the style your team chose. That is the difference between code that works and code that passes review on the first try.
Three skills to align Codex with your stack:
backend architecture patterns, API design, database optimization, and server-side best practices for routes API from Nod…
BackendView the skill →Modern frontend patterns for React, Next.js, and performant user interfaces.
FrontendView the skill →Quick reference for PostgreSQL best practices. For detailed guidance, use the `database-reviewer` agent.
BackendView the skill →Which skills make Codex actually test its code?
An agent that announces 'all done' without checking anything wastes your time. A test-driven development skill enforces the right order: write the test, watch it fail, write the minimum code to make it pass. A verification loop skill forces Codex to check its work before calling it finished. An end-to-end interface testing skill covers what unit tests never see.
These skills change the trust relationship. You stop rereading every line as a precaution, because verification is part of the agent's job rather than yours.
Three skills for tests that hold up:
Write the test first. Watch it fail. Write minimal code to pass.
Tests et qualitéView the skill →A comprehensive verification system for Kiro sessions.
Tests et qualitéView the skill →Use this for Control UI changes that need a real browser flow with deterministic Gateway data.
Tests et qualitéView the skill →Which skills let you hand Codex a big piece of work?
On a migration or a refactor, the hard part is not writing the code: it is keeping the thread from start to finish. An execute plans skill has it draft a plan, criticize it, then work through it step by step. A checkpoints skill records progress so it can resume after an interruption. An autonomous loop skill frames the tasks where Codex must iterate on its own, with clear limits.
This is where Codex is worth the most: you describe the goal, it moves forward in verifiable steps, and you review the result rather than driving every command.
Three skills for work that spans hours:
Load plan, review critically, execute all tasks, report when complete.
Workflows automatisésView the skill →Create, verify, or list workflow checkpoints after running the validation checks.
Workflows automatisésView the skill →Run autonomous loops safely with clear stop conditions, observability, and recovery actions.
Workflows automatisésView the skill →Which skills plug Codex into your everyday tools?
An agent locked inside your editor stays limited. A browser automation skill lets it navigate, fill in a form or check a page in real conditions. A Trello skill gives it access to your boards, lists and cards. A command-line email skill lets it read and send messages from the terminal.
The idea is to remove the manual back-and-forth: Codex fetches the context it needs itself and acts where the work actually happens. Just review the permissions you grant each tool before installing it.
Three skills to connect Codex to your environment:
Use this skill when you need the `browser` tool for anything beyond a single page check.
Workflows automatisésView the skill →Manage Trello boards, lists, and cards directly from OpenClaw.
Connecteurs et intégrationsView the skill →Use `himalaya` for IMAP/SMTP email from shell.
Connecteurs et intégrationsView the skill →Which skills help with documentation and design in Codex?
That leaves what everyone postpones. An interface design skill helps Codex produce a front end with real character, instead of the same generic template. A diagram creation skill turns an architecture into a readable diagram rather than a paragraph. A documentation inventory skill lists what already exists before writing, to avoid duplicates.
These are the tasks nobody asks for but everyone regrets skipping. Handed to a skill, they cost a few minutes instead of an afternoon.
Three skills for documentation and design:
This skill guides creation of distinctive, production-grade frontend interfaces that avoid generic "AI slop" aesthetics.…
FrontendView the skill →Create diagrams as artifacts, not prose. Choose one output mode:
Documentation techniqueView the skill →You are the inventory sub-agent for technical documentation.
Documentation techniqueView the skill →Where should you start with Codex skills?
Start small. Create the .agents/skills folder in a real project, add a single skill matching your stack, and work with it for a day. You will see straight away whether the code produced moves closer to your conventions.
Then add a verification or testing skill, the one that pays off fastest: it turns a talkative agent into one that proves what it claims. Fill in your AGENTS.md alongside, with your repository's commands and rules, so Codex starts well every session.
And keep the rule that runs through this blog: the agent proposes, you decide. A well-chosen skill makes Codex's first draft much closer to what you would have written, but your review is still what validates it.
Frequently asked questions about Codex skills
Where do you put a skill for Codex?
In the .agents/skills folder at the root of your repository if the skill should serve the whole team, or in $HOME/.agents/skills for your personal skills, available across all your projects. Each skill is a folder containing a SKILL.md file. Be wary of tutorials pointing to .codex/skills: that is not the official path.
How do you call a skill in Codex?
Type a dollar sign followed by the skill name, for example $code-reviewer. Codex can also activate it on its own when your request matches the description written in the SKILL.md, without you naming it.
What is the difference between a skill and AGENTS.md?
AGENTS.md holds the repository's always-on rules: install and test commands, conventions, things to avoid. A skill is a capability loaded on demand for a repeatable task. Permanent rules go in AGENTS.md, procedures become skills.
Do Claude Code skills work with Codex?
Often yes, because in both cases the format rests on a SKILL.md file with a name and a description. What changes is the install location and how you call the skill. Check the compatibility shown on each page before adopting it.
Ready to take action?
Explore the skills and plugins matched to your needs and install them in minutes in your AI coding tool.
