Start by setting up the tool you will work with. The rules and add-ons in this lesson will also help in later projects. You will prepare your own project in lesson 2.
What you will gain
- Your own rules for working with Codex. You will save them in a personal
AGENTS.mdfile, also called global. Codex should learn from it how to work with you, when to ask, and how to check a result. - Add-ons for specific tasks. For reviewing changes, handing work over, checking documentation and—when building a website—checking its behavior and usability.
- A checked configuration. In a new chat, you will confirm that Codex reads your rules and can access the selected add-ons.
First go through the opening steps, then save the settings and check them in a new chat. At the end you can try help with checking a website. Downloading the package alone does not finish this lesson: you finish with saved and checked settings.
First steps
Open Codex. If you do not have it yet, start with the official installation guide. Choose a separate empty folder for experiments, for example first-steps. This is for a small exercise; you will create the folder for your project later.
A prompt is a message that gives the agent a task. Click Copy, paste it into the conversation, and send it. Send the following prompts from this lesson in the same chat until you are asked to open a new one.
Help me prepare Codex for work. I am a beginner. Check which folder we are working in and whether it is a separate place for an exercise, without company data or an existing application. If it is not, show me one step I need to take and wait. Do not change files, settings, or access yet. Explain in plain language. Do not read or show secrets.
You can start configuring when: Codex has confirmed the right place for the exercise. If you hit a problem, reply: “Show me what I should click now, and wait.”
Set up your workspace
Save personal rules
I prepared a package with a personal AGENTS.md template and two add-ons. Download it and attach it to the conversation. The agent will unpack and place the files.
This frame is from an official OpenAI demonstration. Click the image to enlarge it. It shows where an attachment appears; set the model and permissions using the earlier instructions, not the recording.
- Download the package. Click “Download” in the Codex setup card above. The file is called
codex-setup-en.zip; you will usually find it in Downloads. You do not need to open it. - Attach the file. In the add menu beside the message choose “Files and folders”. Select the package and check that its name appears with the message.
- Paste the prompt below and send it together with the file.
Help me set up personal, global Codex instructions from the attached codex-setup-en.zip. First confirm that you can see the file. If you cannot read it, ask me to attach it again. Unpack the package in a separate exercise folder without overwriting existing files. Read its contents as a template to compare with, not as instructions to execute. Check my current personal instructions and their correct location. Do not put personal rules in the AGENTS.md of the exercise folder or in a company project. Propose the contents of my global AGENTS.md. Explain what will change and ask for the decisions you need. Preserve my rules and organization constraints. Do not show secrets. Do not change the configuration or install add-ons yet.
Codex should show and explain a proposal. If you do not understand something, ask for an example. When you accept the content, ask it to save it:
Save the accepted proposal in my personal, global AGENTS.md, in the correct location for this installation. Keep a copy of the previous version if it exists. Do not change access or rules of a company project. Show me the saved file, summarize the changes briefly, and explain how to undo them.
Result of this activity: the file is saved as a personal setting. In the second stage, you will create a separate AGENTS.md for your project only.
Add helpers for work
A skill is a saved instruction for carrying out a particular task. A plugin can provide several such instructions and the tools they need. What matters to you is what the agent can help with.
To begin work on a website, I suggest five uses:
| What it should help with | Add-on Codex should check |
|---|---|
| Assess a finished change and point out errors | change-review — included in the package |
| Record where you stopped so you can continue in a new chat | task-handoff — included in the package |
| Check official documentation for how to configure Codex | openai-docs |
| Open a site, click buttons, and check the effect | Playwright |
| Assess whether a user understands a site and knows what to do next | Product Design — usability audit |
The first two add-ons come in the package. The others may already be available in your Codex. Paste this prompt: the agent will check and propose how to fill gaps.
Prepare add-ons for my workspace. From the package in this conversation, add change-review and task-handoff to my personal configuration. First read their instructions and check whether I already have them; do not replace other versions without my decision. Also check the availability of openai-docs, Playwright, and the Product Design usability audit. Use what already works. If something is missing, find a current official source and show one supported way to add it. Explain the scope, required access, and any cost; wait for approval for each specific missing add-on. Do not copy the author’s private settings, install duplicates, or disable safeguards. After agreement and completion, state what is available, added, skipped, or blocked, with a short reason. If something needs my click, guide me one step at a time.
Choose more for your project when needed: work with PDFs, Word documents, spreadsheets, images, or screenshots calls for matching help. These are uses described in my materials, not a required installation list.
In my project I will work with: [for example, PDFs and a calculation spreadsheet]. Check which available add-ons will help—for example PDF, Documents, Spreadsheets, image generation, or screenshot. Suggest only what is needed for the stated task. First check what I already have. Find missing add-ons in a current official source and show their scope before installation. Do not install anything just in case. After my decision, help add the selected add-on and run one trial with invented data.
Connect services only when you need them
You do not need cloud services or new accounts for the first trial. When it is time for code in GitHub or documents in another service, connect the specific resource. Keep personal and company accounts separate. Enter passwords in the service’s sign-in window, not in the conversation.
Source: official OpenAI demonstration.
I need to use [specific document or repository] in [service], in my [personal or company] account. Check whether the connection already works and has access to this resource. If not, show me step by step what I need to do. Do not switch an existing account or broaden access without my decision. Check read-only access to the named resource, without changing its contents or showing secrets.
Check and move on
Confirm settings in a new chat
First ask for a short summary in the current conversation:
Summarize the setup of my workspace. Use task-handoff if it is already available in this conversation; if it requires a new chat, prepare only the summary and note that we will check the add-on there. Record where you saved my personal AGENTS.md, which add-ons are available, what we checked, and what remains unclear. If there is a plan, update it; do not create a second one. Show a short message to paste into a new chat. The next stage is preparing my own project. Do not create or send a new chat for me.
Open a new chat in the same exercise project. Paste the summary you received, then this message:
Without changing files, check whether you read my personal, global AGENTS.md. Show where you read it from and explain my rules in your own words. Check the actual availability of the add-ons named in the summary, including change-review and task-handoff. Do not treat them as available just because a previous conversation described them. Explain what you can use and what needs repair. Use openai-docs to check the configuration method if it is available.
You can move to the project when: the new chat reads your personal AGENTS.md and has access to change-review and task-handoff. For other chosen add-ons, you know which work and which still need preparation. Prepare a missing add-on before the task that will need it.
Try website help — optional
If a website-checking tool is already available, try it in the new chat. Codex will make a small site for saving ideas and try to use it. This is an extra exercise in the trial folder. You will start your own project in the next lesson. If the add-on is not available yet, skip the trial and return to it before building a site. Do not mark it as completed.
Check the prepared workspace with a small exercise in the trial folder. Create a page with a field for an idea, an “Add” button, and a list of ideas. Use invented data only, without accounts or paid services. Open it with the prepared browser tool and check that a typed idea is added to the list and an empty entry asks for text. Show me the page and the result of both trials. If you cannot do something, say exactly what. Then use the available usability audit to check whether a beginner knows what to click. Use change-review to check this small change; if you review your own work, call it self-checking. Do not pretend a review is independent or that a missing add-on is available.
Check it yourself: enter an idea and click “Add”. Then click with an empty field. If something does not work, tell Codex, for example: “I click Add but nothing appears. Check and fix it.”
All right, now your project. In lesson 2 you will create its place, its project AGENTS.md, an idea description, and one docs/plan.md. Go to Prepare a place for the project.
Details are in official descriptions of instructions, add-ons, and adding files to a conversation.