21-Day ChatGPT Codex Challenge
The same structured 21-day path, built around ChatGPT Codex. Practical, daily, beginner-friendly.
21-day roadmap + bonus week
Week 1
Install Codex and run your first safe prompt
Install Codex, sign in using the easiest current path, and run one safe prompt without changing files.
ChatGPT vs Codex - know which tool to use
Using the wrong tool for the wrong job wastes time and causes mistakes. Today you learn the one simple rule, walk through three worked examples, practice classifying real tasks, and run the two-tool workflow from idea to safe Codex prompt.
Plan Mode - make Codex think before it builds
Learn how to ask Codex for a written plan before it touches any file, how to read and evaluate that plan, and how to reduce it if it looks too large.
Open a project without breaking anything
Find the correct project folder, open a terminal inside it, and confirm you are in the right place before Codex touches a single file.
Approval modes - when to use Suggest, Auto Edit, and Full Auto
Learn the current approval modes in Codex, when each one is safe, and how to stop a run immediately if something looks wrong.
Read Codex output without panicking
Learn to read terminal error messages calmly, find the useful line quickly, ask for an explanation before anyone tries to fix anything, and save an error record you can use if the first fix attempt fails.
Week 1 build day - one safe homepage change from start to finish
Today you use everything from Days 1 to 6 together on one real task. You will pick one small visible change, ask Codex to plan it, approve only what you understand, preview it in your browser, check the diff, and save only when it looks right.
Week 2
The AGENTS.md file - give Codex permanent project rules
Every Codex session starts fresh with no memory of your project. Today you create one file that fixes this — AGENTS.md. It sits in your project folder and tells Codex exactly how to behave before it touches a single file. You will create it, paste a starter template, and confirm Codex actually reads it.
What to put in AGENTS.md so Codex follows your rules every session
Learn the six essential sections of a strong AGENTS.md, study a complete working example, customise at least one rule for your project, and confirm Codex reads it accurately.
Slash commands - how to find and use shortcuts safely
Codex has a set of slash commands — shortcuts you type starting with "/" that control the session. Today you learn how to find the commands available on your version of Codex, identify which ones are safe for beginners, try two of them, and build a personal cheat sheet. Important: command names can differ between Codex versions, so today you learn how to find your local list rather than memorise a fixed set.
Permissions and approvals - do not let AI run wild
Learn how to read permission prompts, approve only what you understand, point Codex at the right file before it starts, and reject or revise actions that are too broad.
How to review a diff before accepting changes
Learn to read what Codex actually changed, spot unexpected edits before committing, check user-facing text, and handle the case where two sessions may have touched the same file.
When Codex goes wrong - the exact recovery process
Learn what to do the moment something breaks: stop immediately, collect the right information, build a recovery summary, use GitHub Desktop to undo safely, and know exactly when to stop trying to fix and start fresh.
Week 2 build day - add a contact section with no backend email sending
Today you use the full safe Codex workflow to add one real feature: a simple contact section. No email sending. No backend. Just a visible contact section with your details or a basic form that does not submit anywhere yet.
Week 3
Codex skills - reusable instructions for repeated work
Build a reusable instruction block for one repeated task, store it somewhere you can find it, and test it on a live example.
Create your first personal Codex workflow
Build one complete reusable workflow from scratch — a pre-session checklist, a standard opening prompt, and a closing report request — then test it on a small real task so you know it works before your next session.
Big-build planning - handle complex multi-file tasks safely
Day 3 taught you how to ask for a plan on a small task. Today you handle something bigger — a feature that touches multiple files across multiple phases. You will learn how a big-build plan looks different from a small one, how to spot when a plan is too large, and how to split it into phases you can approve one at a time.
GitHub safety - branches, commits, and how to undo changes
When Codex changes files, those changes happen directly on your computer. Without GitHub, one wrong Codex session could overwrite weeks of work with no way back. Today you learn to use GitHub Desktop — the safest, most visual way to manage your project — to create branches, commit working changes, review what Codex did, and safely undo changes when something goes wrong.
Use Codex for SEO fixes, UI fixes, and bug fixes
Learn three of the most common Codex job types: improving one SEO element, fixing one visible UI problem, and diagnosing one bug. Each job type needs a different prompt structure — today you build and use all three.
Build your pre-session safety checklist
Create the complete pre-session safety checklist you will use before every Codex session for the rest of your building career. By the end of today, the checklist is saved, tested, and added to AGENTS.md.
Final build day - plan, build, test, and ship one real improvement
This is the last day of the challenge. Today you use every skill from all 21 days together — on one real, small improvement that you will take from an idea all the way to a tested, committed result. You will plan before building, approve only what you understand, preview on desktop and mobile, check the diff, and write a short summary of what you shipped and what safety habit helped you most.
Bonus Week — Advanced Skills
Safe automation in Codex — use repeatable commands, not mystery hooks
Build repeatable Codex workflows with visible commands, AGENTS.md rules, and reusable skills instead of undocumented hooks.
The context window — why Codex goes weird and how to fix it
Spot the warning signs of a full context and fix it before things go wrong.
Agents.md checklist — the session rules that protect you
The five rules that keep every Codex session safer and more predictable.
What is MCP and how to connect Codex to your tools
Connect Codex to GitHub, Supabase, and more — without switching apps.
Specialist role prompts — give Codex one job at a time
Use narrow reviewer, debugger, and writer roles so Codex thinks like a specialist instead of a vague general assistant.
Permissions and safety rules before you let Codex touch your files
The permissions and approval settings every non-developer should review.
How to read a diff — and why you must before accepting any change
Read a diff in 30 seconds, catch unexpected changes, and revert only what you did not want.