In previous lessons, we set up the “constitution” for our Codex agent using the AGENTS.md file. We taught the agent how to navigate our project and which coding styles to follow. Now that the rules are set, it is time to learn how to drive the agent during a live work session.

Slash commands are special shortcuts that you type directly into the Codex composer. They start with a forward slash (/). Think of these as your “live” controls. While AGENTS.md provides permanent instructions, slash commands allow you to change how Codex behaves right now, without changing any files in your repository.

On CodeSignal, the Codex CLI is already installed and configured for you. You do not need to worry about setup; you can jump straight into a session by typing codex in your terminal.

It is important to remember the difference between what we have already learned and what we are doing today:

• AGENTS.md (Static Rules): These are the permanent rules for your project. For example, “Always use tabs instead of spaces.” Codex reads this every time it starts.
• Slash Commands (Live Control): These are temporary settings for your current conversation. If you want to switch to a cheaper/faster model just for the next ten minutes, you use a slash command.

Codex allows you to choose which model it uses for a task. Sometimes you need a powerful model for deep reasoning, and other times you want a cheaper/faster model for simpler questions to save time and resources.

To change the model, use the /model command:

When you run /model, Codex CLI will open an interactive picker where you select:

1. The model, then
2. The reasoning effort/level for that model

You’ll typically see options like:

1. gpt-5.1-codex-max — Codex-optimized flagship for deep and fast reasoning.
2. gpt-5.1-codex-mini — Optimized for Codex. Cheaper, faster, but less capable.
3. gpt-5.2 — Latest frontier model with improvements across knowledge, reasoning, and coding.

After you pick a model, it may prompt you to choose a reasoning level (for example, Medium vs High). This “two questions” flow is normal: first choose the model, then choose the reasoning effort.

• gpt-5.1-codex-mini: Great for exploring a codebase, quick explanations, lightweight edits.
• gpt-5.1-codex-max (or gpt-5.2): Better for heavy refactors, tricky bugs, or complex debugging.

If you want to pick a model before the session starts (rather than inside the composer), Codex CLI may also support selecting a model via:

• codex -m <model_name>, or
• a default in config.toml

These are separate from slash commands and apply at launch time.

During a session, you can control how much access Codex has using the /approvals command:

When you run it, Codex opens an interactive Approval Mode picker. You’ll see options like:

• Read Only: Safest mode. Codex can inspect the repository, but editing files and running commands requires your approval.
• Agent: A practical default for many tasks. Codex can read and edit files, and run commands in the workspace.
• Agent (full access): Most powerful and most risky. Codex can edit files outside the workspace and run commands with network access. Use this only when you intentionally need those capabilities.

A good workflow is to start more restrictive (like Read Only) while you’re exploring or planning, then move to Agent once you’re ready to implement changes. Only use Agent (full access) when there’s a specific need.

After you have changed your model or approval settings, you should verify that the changes worked. The /status command acts as your session dashboard.

When you run this, Codex will display a summary of your current session settings. For example, you might see something like this:

This output confirms which model is active, what your approval mode is set to, which directory Codex is operating in, and whether an AGENTS.md file was detected. It may also show token usage and rate-limit/credit info (depending on availability).

In this lesson, we learned how to steer a live Codex session using three core commands:

1. /model: Opens a picker to switch models (and then select reasoning effort).
2. /approvals: Opens a picker to select the session’s approval mode (Read Only, Agent, or Agent (full access)).
3. /status: Shows the current state of your session.

Combining these allows you to create a workflow that is both fast and safe. For example, you can start a session, use /model to switch to gpt-5.1-codex-mini for quick research, and then use /approvals to choose Read Only while you plan changes—switching to Agent when you’re ready to implement.