OpenAI Codex Training

A hands-on training course for using OpenAI Codex across the CLI, app, IDE extensions, and cloud workflows.

Course Overview

This 5-hour hands-on workshop covers installation, authentication, safe agent workflows, configuration, MCP servers, skills, cloud tasks, and practical development labs.

What You’ll Learn

• 🚀 Installation & Setup: CLI, app, IDE, ChatGPT login, and API-key authentication
• 🛡️ Safety & Security: Sandbox modes, approval policies, trusted projects, and reviews
• 🎯 Core Features: AGENTS.md guidance, custom prompts, profiles, and session resumption
• ⚡ Agent Skills: Reusable workflow bundles with scripts, references, and assets
• 🔧 Advanced Capabilities: MCP integration, Codex Cloud, local OSS models, and CI/CD workflows
• 💻 Practical Skills: Real-world exercises in Java, Python, and TypeScript

Prerequisites

• Command-line experience
• Basic programming knowledge in at least one language
• Git familiarity
• Docker (for advanced exercises)

Repository Structure

codex-training/
├── slides.md                     # Slidev presentation (160+ slides)
├── exercises/                    # Hands-on labs with starter code
│   ├── plan-mode-warmup/        # Lab 0: Plan mode and steering warm-up
│   ├── java-spring-boot/        # Lab 1: Spring Boot REST API
│   ├── python-refactoring/      # Lab 2: Legacy code refactoring
│   ├── react-forms/             # Lab 3: React TypeScript forms
│   ├── microservices/           # Lab 4: Multi-language microservices
│   ├── skills-creation/         # Lab 5: Create custom Codex skills
│   └── mcp-context7/            # Lab 6: Connect current docs through MCP
├── config-examples/              # Sample configurations
│   ├── basic-config.toml        # Minimal setup for beginners
│   └── advanced-config.toml     # Full features with MCP, profiles
├── prompts/                      # Custom prompt templates
│   └── *.md                     # security-audit, test-gen, etc.
├── docs/                         # Reference documentation
│   ├── TROUBLESHOOTING.md       # Common issues and solutions
│   └── MODEL-SELECTION.md       # Cost/performance guide
├── .github/workflows/            # CI/CD examples
│   └── codex-review.yml         # Automated PR review with Codex
└── exports/                      # Slide exports (PDF)

Quick Start

1. Install Codex

# Official install script (macOS/Linux)
curl -fsSL https://chatgpt.com/codex/install.sh | sh

# Via npm
npm install -g @openai/codex

# Via Homebrew (macOS/Linux)
brew install --cask codex

# Verify installation
codex --version

2. Authenticate

# ChatGPT account (recommended)
codex login

# Or store an API key with Codex
printenv OPENAI_API_KEY | codex login --with-api-key

3. Run Your First Command

codex "Create a hello world function in Python"

4. Start the Training

# Clone this repository
git clone https://github.com/kousen/codex-training
cd codex-training

# Install dependencies for slides
npm install

# Start the presentation
npm run dev

# Open browser to http://localhost:3030

Exercises

Lab 0: Planning and Steering Warm-Up

Learn the current Codex interaction model before the larger labs:

• Ask Codex to propose a plan before editing
• Steer an in-flight task with follow-up messages
• Inspect permissions, model choice, and diffs with slash commands

Time: 15-20 minutes

Lab 1: Spring Boot REST API

Build a complete task management API with Spring Boot, including:

• CRUD operations with validation
• H2 database integration
• OpenAPI documentation
• Comprehensive test suite

Time: 60-90 minutes

Lab 2: Python Code Refactoring

Transform legacy Python code using modern best practices:

• Add type hints and documentation
• Implement design patterns
• Create pytest test suite
• Setup code quality tools

Time: 45-60 minutes

Lab 3: React TypeScript Forms

Create a production-ready registration form with:

• React Hook Form + Zod validation
• Accessibility compliance
• Multi-step workflow
• Full test coverage

Time: 45-60 minutes

Lab 4: Microservices Architecture

Build an event-driven microservices system with:

• Multiple languages (Node.js, Python, Go, Java)
• RabbitMQ message queue
• Docker orchestration
• API gateway

Time: 90-120 minutes

Lab 5: Skills Creation

Create a custom Codex skill using $skill-creator:

• SKILL.md format with YAML frontmatter
• Conventional commit message generator
• Team sharing via repository

Time: 30 minutes

Lab 6: MCP Servers with Context7

Connect Codex to current library documentation:

• Add Context7 as an MCP server
• Inspect configured MCP tools
• Use current React Hook Form, Zod, pytest, and Spring docs in prompts

Time: 15-20 minutes

Key Codex Features Covered

Core Features

• ✅ Terminal UI navigation
• ✅ Sandbox modes and approval policies
• ✅ Project memory with AGENTS.md
• ✅ Custom prompts and profiles
• ✅ Session management and resumption

Agent Skills

• ✅ Creating reusable skill bundles
• ✅ SKILL.md format and structure
• ✅ Explicit ($skill-name) and implicit invocation
• ✅ Built-in skills: $skill-creator, $skill-installer
• ✅ Skills vs prompts comparison

Advanced Features

• ✅ Model Context Protocol (MCP)
• ✅ OpenAI models plus local OSS providers through Ollama or LM Studio
• ✅ Running as MCP server
• ✅ Codex App, IDE extensions, and Codex Cloud workflows
• ✅ CI/CD integration
• ✅ Non-interactive execution

Configuration

• ✅ TOML configuration files
• ✅ Project-local config (.codex/config.toml)
• ✅ Environment variables
• ✅ Shell environment policies
• ✅ Notification systems

Tips for Success

1. Start Conservatively: Use workspace-write with on-request, or read-only for review-only tasks
2. Use AGENTS.md: Provide context for better results
3. Create Profiles: Separate development/production configurations
4. Review Generated Code: Never blindly accept AI suggestions
5. Leverage MCP: Pull current documentation and project data into prompts
6. Test Thoroughly: Always verify generated code works correctly

Useful Commands Reference

# Basic usage
codex                          # Interactive mode
codex "prompt"                 # Start with initial prompt
codex exec "prompt"            # Execute and exit
codex resume                   # Resume session picker
codex resume --last            # Resume most recent session
codex update                   # Update the CLI

# Configuration
codex --profile development    # Use specific profile
codex --sandbox read-only      # Set sandbox mode
codex --model gpt-6-sol      # Pin a model (/model or codex debug models lists your catalog)
codex --strict-config          # Fail fast on stale/unknown config keys
codex --search "latest docs"   # Enable live web search for that run

# Advanced
codex mcp list                 # List configured MCP servers
codex mcp login server-name    # OAuth login for supported MCP servers
codex doctor                   # Diagnose install, config, auth, and runtime health
codex cloud exec "prompt"      # Launch a cloud task
codex cloud diff               # Inspect cloud task changes
codex cloud apply              # Apply cloud task changes locally
codex login --device-auth      # Device-auth flow for remote/headless environments
codex review --uncommitted     # Review local changes
codex review --base main       # Review your branch against main

Inside an interactive Codex session, you can also invoke skills directly:

$skill-creator
$skill-installer linear
$create-plan

Resources

Course Materials

• Course Slides - Full Slidev presentation
• Exercises - Hands-on labs with starter code
• Troubleshooting Guide - Common issues and solutions
• Model Selection Guide - Cost/performance optimization

Official Documentation

• Codex CLI Documentation
• Codex App Documentation
• Codex MCP Documentation
• Codex GitHub Repository
• Agent Skills Documentation
• Skills Catalog
• Model Context Protocol
• agentskills.io Specification

Instructor

Kenneth Kousen

• President, Kousen IT, Inc.
• Author & Technical Trainer
• [email protected]
• https://www.kousenit.com

Contributing

Found an issue or want to contribute? Please:

1. Open an issue on GitHub
2. Submit a pull request
3. Share your experience

License

This training material is licensed under the MIT License. See LICENSE file for details.

Acknowledgments

• OpenAI for Codex
• The open-source community for MCP tools

---

Ready to start? Navigate to the exercises directory and begin with Lab 1!