The Codex CLI for Agentic Coding Crash Course

A coding agent that can edit your files and run your commands is a completely different tool from a chatbot that suggests code, and it deserves a different kind of training. This crash course teaches you to work with the Codex CLI the way a careful professional does: inside one small, safe project, with every change planned, tested, reviewed, and recoverable.

The whole course runs inside a single tiny Python project, a command-line expense tracker, so nothing real is ever at risk. You install Codex cleanly, sign in and understand exactly where your credentials live, and tour the repository before touching a single file. From there you work one complete, repeatable loop again and again: inspect, plan, edit, test, read the diff, checkpoint with Git, and recover calmly when something goes wrong.

Inside, you will learn:
• What the Codex CLI is and is not, and how its agentic loop of reading, editing, and running commands actually works
• How to install Codex without guesswork and sign in safely with a ChatGPT account or an API key
• How to read a repository before changing it and set lasting project rules with an AGENTS.md file
• How to plan before you prompt, make a first safe edit, and read a diff line by line like a reviewer
• How to use tests as a contract, with red-green cycles, regression tests, and honest reading of failures
• How Git checkpoints, branches, restore, and revert turn agent mistakes into short detours instead of disasters
• How permissions, sandboxes, secrets, and prompt injection define the real boundaries of safe agentic coding

Safety is not a chapter at the end; it is the method. You will demonstrate a denied write and a denied network call, spot a hostile instruction hidden inside a training fixture and refuse it, and learn why generated changes are proposals that earn trust only through tests and review. A second, independent review pass closes the course, followed by a realistic first-week plan for building the habit.

This book is for developers meeting a terminal coding agent for the first time, students and career changers who can follow simple Python, and experienced programmers who are curious about agentic coding but unwilling to hand an assistant a real codebase blind. Every chapter ends with a checkpoint that names exactly what you should see, so you always know whether you are on track.

Finish the course and you will own a supervised, repeatable workflow for AI-assisted development that transfers to any project you work on next. This is an independent guide, not affiliated with, sponsored, or endorsed by OpenAI.