Directory

Codex CLI Skills: Best Skills for OpenAI Codex in 2026

Codex CLI skills are reusable workflows that extend OpenAI's coding agent with specialized capabilities. Unlike one-off prompts, skills are discoverable, chainable, and can be auto-invoked when Codex detects a matching task. This directory lists the most useful Codex skills across GitHub automation, testing, documentation, and operations — with installation instructions and comparisons to Claude Code equivalents.

Table of Contents

1. What Are Codex Skills
2. How to Install
3. Skills Directory (10 skills)
4. Create Your Own Skill
5. Codex vs Claude Code Skills
6. FAQ
7. Related Resources

What Are Codex CLI Skills

Codex skills are self-contained workflow definitions stored as SKILL.md files. Each skill describes what it does, when it should be triggered, and the step-by-step process Codex should follow. Skills live in .agents/skills/ directories — either per-project (in your repo root) or globally (in ~/.codex/agents/).

What makes skills powerful is automatic discovery. When you give Codex a task, it scans available skills and selects the best match. You can also invoke skills explicitly using the$ prefix (e.g., $ gh-fix-ci) or browse available skills withcodex /skills.

Skills can be chained — one skill's output becomes another's input. For example,gh-address-comments can read PR feedback, then webapp-testing can generate tests for the changes, and changelog-generator can document what changed. This composability is what separates skills from simple prompt templates.

How to Install Codex Skills

Per-Project Skills

# Create skills directory in your project
mkdir -p .agents/skills/gh-fix-ci

# Add the skill definition
cat > .agents/skills/gh-fix-ci/SKILL.md << 'EOF'
# gh-fix-ci

## Description
Reads CI failure logs from GitHub Actions, diagnoses the root cause,
and applies targeted fixes to pass the pipeline.

## Trigger
When the user mentions CI failure, build error, or pipeline issues.

## Steps
1. Read the latest CI run logs from GitHub Actions
2. Identify the failing step and error message
3. Analyze the codebase for the root cause
4. Apply the minimal fix to resolve the failure
5. Run the failing test locally to verify
EOF

# Verify skill is discovered
codex /skills

Global Skills

# Global skills available across all projects
mkdir -p ~/.codex/agents/skills/email-draft-polish
# Add SKILL.md as above

# Or clone a curated collection
git clone https://github.com/ComposioHQ/awesome-codex-skills.git
cp -r awesome-codex-skills/skills/* ~/.codex/agents/skills/

Skills Directory

These 10 skills are selected for real-world utility. Each has been verified to work with current Codex CLI versions.

GitHub

gh-address-comments

Automatically reads PR review comments and generates code fixes that address each comment.

Use case: Speed up PR review cycles by auto-resolving reviewer feedback.

Source: ComposioHQ/awesome-codex-skills

gh-fix-ci

Reads CI failure logs, diagnoses the root cause, and applies targeted fixes to pass the pipeline.

Use case: Unblock deployments by fixing CI failures without manual debugging.

Source: ComposioHQ/awesome-codex-skills

Testing

webapp-testing

Generates comprehensive test suites for web applications including unit, integration, and E2E tests.

Use case: Increase test coverage for frontend and backend codebases.

Source: ComposioHQ/awesome-codex-skills

Content

content-research-writer

Researches a topic using web search, then generates structured long-form content with citations.

Use case: Produce technical blog posts and documentation drafts.

Source: ComposioHQ/awesome-codex-skills

email-draft-polish

Takes a rough email draft, improves clarity, tone, and structure while preserving the original intent.

Use case: Professional communication without spending time on rewrites.

Source: ComposioHQ/awesome-codex-skills

Documentation

notion-research-documentation

Pulls context from Notion pages and generates or updates technical documentation based on current state.

Use case: Keep docs in sync with Notion knowledge bases.

Source: ComposioHQ/awesome-codex-skills

changelog-generator

Reads git history between two refs and produces a structured changelog grouped by type (feat, fix, etc).

Use case: Automate release notes for every version bump.

Source: ComposioHQ/awesome-codex-skills

Operations

support-ticket-triage

Classifies incoming support tickets by urgency, category, and suggested assignee based on content analysis.

Use case: Reduce manual triage time for customer support teams.

Source: ComposioHQ/awesome-codex-skills

Maintenance

file-organizer

Analyzes project file structure and suggests reorganization based on conventions and best practices.

Use case: Clean up repos that have grown organically without structure.

Source: ComposioHQ/awesome-codex-skills

Database

db-migration-reviewer

Reviews database migration files for safety issues: destructive operations, missing rollbacks, data loss risks.

Use case: Catch risky migrations before they reach production.

Source: Community

How to Create Your Own Codex Skill

Building a custom skill takes under 10 minutes. The key is writing a clear SKILL.md with three sections: description (what it does), trigger (when to auto-invoke), and steps (the workflow).

# SKILL.md template

# skill-name

## Description
One paragraph explaining what this skill does and what problem it solves.

## Trigger
Describe when Codex should automatically invoke this skill.
Example: "When the user asks to review a database migration file."

## Steps
1. First action Codex should take
2. Second action with specific details
3. Validation step to verify the result
4. Output format or final deliverable

## Notes
- Any constraints or edge cases
- Required tools or API access
- Known limitations

Tips for effective skills:

Be specific in triggers — vague triggers cause false matches
Keep steps atomic — each step should be one clear action
Include validation — the last step should verify the output
Test with edge cases — try invoking the skill with unusual inputs

Codex Skills vs Claude Code Skills

Feature	Codex CLI	Claude Code
Skill format	SKILL.md files	.agent/skills/ directories with frontmatter
Invocation	$ prefix or auto-discovery	/slash commands via Skill tool
Auto-trigger	Yes — matches task to skill description	Some skills auto-trigger, most explicit
Chaining	Native skill chaining	Via agent orchestration
Project config	AGENTS.md	CLAUDE.md
MCP support	Limited	Full native MCP integration
Code quality	Fast iteration, speed-focused	Deep reasoning, quality-focused (67% blind review win rate)

Frequently Asked Questions

What are Codex CLI skills?

Codex skills are reusable workflows defined in SKILL.md files that extend OpenAI Codex CLI functionality. They can be invoked explicitly via the $ prefix or automatically when a task matches the skill description. Skills live in .agents/skills directories within repos or system locations.

How do Codex skills differ from Claude Code skills?

Codex skills use SKILL.md files and are discoverable/chainable — Codex can autonomously select which skill to use. Claude Code skills use the Skill tool with explicit invocation via slash commands. Both support custom instructions, but the file format and discovery mechanism differ.

How do I install a Codex skill?

Create a .agents/skills/ directory in your repo or ~/.codex/agents/ for global skills. Each skill is a directory containing a SKILL.md file that describes the skill name, trigger conditions, and workflow steps. Run codex /skills to see available skills.

What is AGENTS.md and how does it relate to skills?

AGENTS.md is a project-level configuration file (like Claude Code's CLAUDE.md) that sets custom instructions for Codex CLI. Skills are more granular — individual reusable workflows that Codex can invoke. AGENTS.md sets the context; skills define specific capabilities.

Can I convert Claude Code skills to Codex skills?

The concepts are similar but the format differs. A Claude Code skill (defined in .agent/skills/) can be adapted to a Codex SKILL.md by mapping the trigger conditions and workflow steps. The core logic translates, but the invocation mechanism changes.

Where can I find community Codex skills?

The main collection is ComposioHQ/awesome-codex-skills on GitHub, which curates production-ready skills across categories like GitHub, testing, documentation, and operations. You can also find skills on agentskillshub.dev and similar directories.

How do I create my own Codex skill?

Create a directory under .agents/skills/ with a SKILL.md file. Define the skill name, description, trigger conditions (when should Codex auto-invoke it), and step-by-step workflow instructions. Test by running codex /skills to verify discovery, then invoke with $ prefix.