MODULE 2.1
Terminal and Codex in practice
Open a training project in Codex and produce a verifiable change.
What it is
The terminal is the window where you write commands for the computer. The shell is the program that interprets those commands. Here the terminal examples use Bash on Linux or macOS; on Windows, use a Bash environment via WSL or follow the official installer for Windows. Do not paste Linux commands directly into PowerShell without adaptation.
Why learn
Knowing which environment you are in avoids errors that seem like AI failures. The cd command changes the current folder; pwd shows the location in Bash. You do not need to memorize dozens of commands to get started.
Key concepts
Terminal; shell; current folder; command and response.
What it is
The official Codex page offers installers for macOS/Linux and specific guidance for Windows. Installing means adding the program to the machine. The installation command downloads and runs an official script; read the source, check the domain and use your own account. Do not run commands received from unknown pages.
Why learn
Installation, login and execution are different steps. A program that is installed still needs authentication. A completed login does not mean you opened the correct folder.
Key concepts
Official source; installation; version; diagnostics.
What it is
Run codex login and complete the browser flow to sign in with ChatGPT. If you choose the API, the key must be in the OPENAI_API_KEY variable; pipe it via standard input, without typing it in the command. A .env file alone does not authenticate Codex: some mechanism must load the variable.
Why learn
Pasting the key in examples, messages, or history can expose the account. It is also possible to be authenticated by the wrong method and consume a different modality than expected.
Key concepts
codex login; codex login status; standard input; active account.
What it is
The working folder is the task bench. Create a small area with training files before allowing changes. A README.md describes the purpose for people; AGENTS.md gives operational instructions to the agent. You do not need to open your entire personal folder to experiment.
Why learn
A small scope reduces ambiguity and makes reviewing differences easier. When something goes wrong, it is clear which files should have been affected.
Key concepts
Local scope; README; AGENTS; input files.
What it is
Start by requesting inspection: list the structure, read instructions and explain pending items. Ask the agent to cite which files it used. After checking, authorize a small, named change, such as creating plano.md with three next steps.
Why learn
Separating diagnosis and change creates a reference for review. You learn the flow without mixing installation, large refactoring, and publishing in a single attempt.
Key concepts
Inspect; plan; change; validate.
What it is
The client’s permissions control access to files, network, and execution. Natural‑language instructions guide behavior, but they do not replace technical isolation. Start with permissions restricted to the training folder. Do not teach removing all protections to work around any error.
Why learn
Recovering a change requires knowing what existed before. Git and file copies provide rollback points; they will be practiced later. Read what the command will do before expanding permissions.
Key concepts
Minimum permission; small changes; comparison; rollback point.