Codex Skill for Claude Code

Delegate coding tasks to OpenAI's Codex CLI with configurable execution modes including YOLO mode for fully autonomous operation. This plugin bridges Claude Code's analytical capabilities with Codex's autonomous code execution.

Quick Start

Install via Claude Code Marketplace (Recommended)

# Add the marketplace
/plugin marketplace add veithly/codex-skill

# Install the plugin
/plugin install rick@codex-skill

Prerequisites

Before using the plugin, ensure you have:

1. OpenAI Codex CLI installed:

   npm install -g @openai/codex

2. OpenAI API Key configured:

   # Linux/macOS
   export OPENAI_API_KEY="your-api-key"
   
   # Windows PowerShell
   $env:OPENAI_API_KEY = "your-api-key"

Features

• Multiple Execution Modes: YOLO (full autonomy), Auto (balanced), Safe (conservative)
• Non-Interactive Execution: Uses codex exec for automation-friendly operation
• Git Integration: Automatic change tracking and rollback support
• Cross-Platform: Works on Windows, macOS, and Linux
• Claude Integration: Seamless workflow with other Claude Code skills

Usage

# Default auto mode - balanced safety and automation
/codex Fix the null pointer exception in src/api/handlers.ts

# YOLO mode - full autonomy (use in hardened environments only!)
/codex --mode=yolo Implement user authentication with JWT

# Safe mode - requires more oversight
/codex --mode=safe Refactor the database connection module

# With custom timeout
/codex --timeout=600 Implement complex multi-file feature

Command Syntax

/codex <task_description> [options]

Options

Option	Description	Default
--mode=<mode>	Execution mode: yolo, auto, safe	auto
--timeout=<seconds>	Maximum execution time	1800
--add-dir=<path>	Additional directory for Codex to access	-

Execution Modes

YOLO Mode (--mode=yolo)

Aspect	Details
Risk Level	HIGH
Approvals	None
Sandbox	Disabled
Best For	CI/CD pipelines, Docker containers, VMs

⚠️ Warning: Only use in externally hardened environments!

Auto Mode (--mode=auto) - Default

Aspect	Details
Risk Level	MEDIUM
Approvals	On-request
Sandbox	Workspace write
Best For	Development, refactoring, feature implementation

Safe Mode (--mode=safe)

Aspect	Details
Risk Level	LOW
Approvals	Default policy
Sandbox	Workspace write
Best For	Production code, learning, sensitive operations

Alternative Installation Methods

Via One-Click Installer

Windows (PowerShell):

git clone https://github.com/veithly/codex-skill.git
cd codex-skill
.\scripts\install\install.ps1

Linux/macOS (Bash):

git clone https://github.com/veithly/codex-skill.git
cd codex-skill
chmod +x ./scripts/install/install.sh
./scripts/install/install.sh

Via npm

npm install -g @anthropic-skills/codex

Manual Installation

# Clone repository
git clone https://github.com/veithly/codex-skill.git
cd codex-skill

# Create directories
mkdir -p ~/.claude/skills/codex
mkdir -p ~/.claude/agents
mkdir -p ~/.claude/scripts

# Copy files
cp skills/codex/SKILL.md ~/.claude/skills/codex/
cp skills/codex/reference.md ~/.claude/skills/codex/
cp agents/codex.md ~/.claude/agents/

Integration with Claude Workflow

Debug → Fix

# Use Claude to analyze
/debug Investigate why user authentication fails

# Use Codex to fix
/codex --mode=safe Fix the race condition in auth token refresh

Plan → Implement

# Use Claude to plan
/code Plan the new payment processing module

# Use Codex to implement
/codex Implement the payment module following the plan above

Implement → Review

# Use Codex for rapid implementation
/codex Implement user registration with email verification

# Use Claude for review
/review Check the registration code for security vulnerabilities

Configuration

Global Codex Configuration

Create ~/.codex/config.toml:

model = "gpt-5.2-codex"
sandbox = "workspace-write"
reasoning_effort = "high"

instructions = """
Follow project coding standards.
Write tests for new functionality.
"""

Project-Level Configuration

Create codex.toml in your project root:

model = "gpt-5.2-codex"
sandbox = "workspace-write"
additional_dirs = ["../shared-libs"]

instructions = """
This is a TypeScript project.
Use functional components with React hooks.
"""

Troubleshooting

"stdin is not a terminal"

The skill uses codex exec for non-interactive execution. If running manually, use:

codex exec --full-auto "your task"

"Codex CLI not found"

npm install -g @openai/codex

"API key not configured"

export OPENAI_API_KEY="your-api-key"

Security Considerations

• YOLO mode bypasses ALL safety measures - use only in isolated environments
• API keys should never be committed to version control
• Always review AI-generated code before committing
• Use Git for rollback capability

File Structure

codex-skill/
├── .claude-plugin/
│   ├── marketplace.json    # Marketplace definition
│   └── plugin.json         # Plugin manifest
├── skills/
│   └── codex/
│       ├── SKILL.md        # Main skill definition
│       └── reference.md    # CLI quick reference
├── agents/
│   └── codex.md            # Sub-agent definition
├── scripts/
│   ├── install/            # One-click installers
│   ├── codex-runner.*      # Helper wrappers
│   └── codex-yolo.*        # Quick YOLO launchers
├── examples/               # Configuration examples
├── README.md               # This file
├── REQUIREMENTS.md         # Detailed requirements
├── CHANGELOG.md            # Version history
└── LICENSE                 # MIT License

Requirements

Component	Requirement
Claude Code	Latest version
Node.js	v18.0.0 or higher
OpenAI Codex CLI	v0.65.0 or higher
Git	v2.0+ (recommended)

See REQUIREMENTS.md for detailed requirements.

Contributing

Contributions are welcome! See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.

Links

• GitHub Repository
• OpenAI Codex CLI Documentation
• Claude Code Documentation