SkillCharged
Codex Crash Course
Agentic AI & Super Apps
Codex
AI Agents
Skills & SOPs
Plugins
Computer Use
Automations

Codex Crash Course (2026): The Complete AI Agent Super App Masterclass

Guided by Nitin Khatri (Lead Software Architect & Designer)
Reviewed by Sarah Miller (Principal AI Engineer, Reviewer)
Updated August 23, 2026
30 min
Level: Beginner

Course Roadmap: What We'll Build & Learn

Learn what OpenAI's Codex is and master its core superpowers: working with local files, persistent memory, external plugins, custom skills, computer & browser automation, and recurring scheduled tasks.

What You Need Before Starting:

  • Codex desktop app installed on your Mac or Windows PC
  • No coding experience required — just a curiosity to learn AI agent workflows
  • A local folder on your computer with a few sample files or documents to practice with
Core Concept

What is Codex? The AI Agent Super App Explained

Why Codex represents a massive leap forward from standard web chatbots.

At first glance, Codex might look similar to ChatGPT or Claude because you can open a chat window and type prompts. But under the hood, they are completely different technologies.

When you chat with a standard web AI, everything is locked inside a cloud browser window. When you upload a file, it lives on remote servers, and the AI cannot touch your actual computer, desktop apps, or local hard drive.

Codex is OpenAI's AI Agent Super App designed to perform real knowledge work and coding tasks directly on your computer. It has full filesystem access to read and write local files, controls your desktop interface (mouse, keyboard, apps), tests web pages inside a real browser, and runs background automated jobs on a schedule.

Traditional Methods vs. Codex

File Storage & Access

Traditional Approach:Uploaded to a remote cloud sandbox; cannot read your local hard drive.
Codex:Direct local file access. All created files, code, and spreadsheets live right on your computer.

Execution & Desktop Control

Traditional Approach:Limited to generating text and code blocks inside a chat window.
Codex:Full computer & browser use. Can move the cursor, click buttons, test web apps, and control desktop tools.

Memory & Custom SOPs

Traditional Approach:Isolated chats that lose detailed workflow context across different sessions.
Codex:Persistent Memory (agents.md), reusable Skills (/), external Plugins (@), and scheduled cron automations.

The 6 Core Superpowers We Will Master

📁

Native Local Filesystem

Read, edit, batch-process, and create real files (Excel, Word, HTML, Code) right on your machine.

🧠

Dual Persistent Memory

Teach the agent your exact formatting rules and style preferences in `agents.md` once, and it remembers forever.

🔌

Connected Tool Plugins (@)

Connect to Email, Notion, GitHub, and Databases with two clicks and query them in natural language.

Reusable Skills & SOPs (/)

Package your best workflows into repeatable slash-command playbooks that run in seconds.

🖱️

Computer & Browser Use

Let the agent control your mouse and keyboard to manipulate desktop apps or test live web UI.

⏱️

Scheduled Automations

Put recurring tasks on autopilot so your agent wakes up and runs your workflows on a schedule.

Quick Cheat Sheet: Essential Commands

Keep these core syntax triggers handy as you follow the walkthrough modules below:

Command / TriggerWhat It Does
@<filename>Tag a local project file to bring its full content into the prompt context
@<plugin-name>Attach an external tool connection (Email, Notion, Database, GitHub)
/<skill-name>Trigger a saved, repeatable procedural SOP recipe
@computer-useLet the agent control your mouse and keyboard to manipulate desktop apps
@browser-useLaunch automated in-browser navigation and click-through UI testing
Turn this into a skillConvert the current chat's verified output into a permanent reusable skill
Module 1

Getting Started: Local File Access & Setting Up Your Workspace

Module Learning Goal

Understand how Codex interacts with your local drive, how to organize project folders, and how to multitask with parallel agents.

If you've used standard AI chat apps on the web, you're used to uploading files into a cloud window. Codex is fundamentally different: it is an AI agent that lives on your computer. It can read folders from your Downloads directory, create real Excel sheets or code files in your Documents, and organize everything directly on your hard drive.

1Step 1: Create a Dedicated Project Workspace

Try this prompt in your agent:
$ Create a new project folder in ~/Documents/my-workspace
What you will see on screen:
✓ Project workspace created at ~/Documents/my-workspace ✓ Linked to active agent session.
Under the hood:Pro Tip: Instead of opening random chats, always create a Project Folder in the left sidebar. Any spreadsheet, document, or code the agent builds will automatically be saved into this local folder on your computer.

2Step 2: Tell the Agent to Process Local Files

Try this prompt in your agent:
$ Look at the files in my ~/Downloads/raw-reports folder. Analyze the contents, categorize each entry, and create a structured analytical Excel spreadsheet with summary charts.
What you will see on screen:
Scanning ~/Downloads/raw-reports (50 files detected)... Extracting data points -> Categorizing transactions -> Creating local spreadsheet workbook at ~/Documents/my-workspace/outputs/analysis.xlsx.
Under the hood:Notice what happened: You didn't manually upload 50 files! The agent navigated to your folder, read every file, and built a real Excel file that you can immediately double-click and open in Excel or Finder.

3Step 3: Reference Existing Documents Using '@'

Try this prompt in your agent:
$ Please review @analysis.xlsx and turn the key takeaways into an interactive landing page outline for my team.
What you will see on screen:
Reading local file analysis.xlsx... Synthesizing metrics -> Building responsive HTML landing page in ~/Documents/my-workspace/outputs/index.html.
Under the hood:Whenever you want the agent to look at a file inside your project folder, simply type '@' followed by the file name. The agent will read it directly from your computer.

Do:Organize your work into Project Folders

Keep each goal (e.g., 'Quarterly Research', 'Website Redesign', 'Financial Audit') in its own project folder. This keeps all generated documents cleanly sorted in Finder or File Explorer.

Avoid:Creating loose standalone chats in the root directory

Avoid starting chats outside of a project folder, as it makes it harder to track where the agent created your documents.

Do:Run multiple agent chats in parallel

You don't have to wait for one task to finish! You can open 2 or 3 chats in the same project simultaneously. One agent can process data while another drafts copy or builds code.

Pro Tip for Beginners

💡
Where are my files stored?:Click the three dots (...) next to any chat or document in Codex and select 'Open in Finder' (or File Explorer) to see the exact folder on your hard drive where your files live.

Hands-On Practice Checklist

Try each of these steps on your computer and check them off as you complete them:

Module 2

Memory Mastery: Teaching the Agent How You Like Things Done

Module Learning Goal

Learn how to use manual memory (agents.md) so the agent permanently remembers your preferences, formatting styles, and quality standards.

Have you ever found yourself repeating the same instructions to an AI over and over? ('Make it concise', 'Use bullet points', 'Include a sidebar'). Codex solves this with Persistent Memory. There are two types of memory you need to know about: Manual Memory (which you control) and Automatic Memory (which the agent manages in the background).

agents.md (Your Personal Manual Memory File)markdown

This living file lives in your project or global root. Codex reads it on every single prompt to tailor its output to your preferences.

# My Personal Guidelines & Style Preferences

## Document & Layout Rules
- Layout: When creating learning docs, always include a left navigation sidebar with a table of contents.
- Formatting: Use clear comparison boxes (Do vs. Avoid) and numbered step-by-step walkthroughs.
- Tone: Instructional, encouraging, and easy for beginners to follow. Avoid jargon without explaining it.
- Code Standards: Write clean, well-commented code with zero placeholders.

Do:Tell the agent to remember styles you love

When the agent produces a great output, simply say: 'I love this layout. Please remember to always use this format in agents.md whenever I ask for similar documents'.

Avoid:Manually editing the background 'memories/' folder

Codex maintains an internal telemetry memory folder for its own observations. Let the agent manage that automatically—only edit `agents.md` yourself.

Pro Tip for Beginners

🧠
How Manual Memory Works:Whenever you update `agents.md` (or tell the agent to update it), every new chat automatically adopts those rules. You never have to repeat your formatting preferences again!

Hands-On Practice Checklist

Try each of these steps on your computer and check them off as you complete them:

Module 3

Connecting Your Tools: Supercharging Codex with Plugins (@)

Module Learning Goal

Connect external tools like Email, Notion, GitHub, and Databases so your agent can pull live context and take real actions.

An AI agent is only as helpful as the tools it can access. With Codex Plugins, you can connect your agent to the software you already use in two clicks. Typing '@' in your prompt brings up a menu of connected tools that the agent can read from and write to.

1Step 1: Enable Plugins in the Plugins Tab

Try this prompt in your agent:
$ Open Plugins tab -> Connect 'Email' and 'Notion'
What you will see on screen:
✓ Connected: Email Plugin (Read/Write permissions granted) ✓ Connected: Notion Plugin (Workspace linked).
Under the hood:Navigate to the Plugins tab in Codex. Connect the apps you want your agent to interact with.

2Step 2: Ask the Agent to Research and Synthesize Across Tools

Try this prompt in your agent:
$ Look at my recent incoming messages from the past week using @email. Find all actionable requests, cross-reference them with our project notes in @notion, and build a summary table.
What you will see on screen:
Searching messages via @email... Cross-referencing notes via @notion... ✓ Created ActionPlan.md with categorized priorities and draft responses.
Under the hood:The agent queries your tools, connects the dots between different data sources, and presents you with a finished action plan.

Do:Use '@' to see available tools

Just type '@' in the prompt box anytime to see the quick menu of all installed plugins and active project documents.

Avoid:Sending emails or modifying databases without review

Ask the agent to draft changes or summaries first so you can review the output before confirming external actions.

Hands-On Practice Checklist

Try each of these steps on your computer and check them off as you complete them:

Module 4

Skills & SOPs: Turning Good Outputs into Reusable Playbooks (/)

Module Learning Goal

Master the 'Iterative Workflow Distillation' technique to turn successful tasks into permanent slash-command recipes (/skill).

What if you could package a multi-step task that took you 30 minutes into a single command that runs in 30 seconds? That is what Skills are. A skill is a standardized recipe (stored as a `SKILL.md` instruction file) that teaches your agent exact steps to execute a recurring workflow.

1Step 1: Do the Task and Iterate Until You Love the Result

Try this prompt in your agent:
$ Analyze these raw data logs, build an executive dashboard with color-coded risk levels, and include a 3-step action checklist.
What you will see on screen:
Agent creates draft -> You request tweaks: 'Make the risk categories red/yellow/green and add a summary table' -> Agent updates output -> Output is now perfect.
Under the hood:Always work interactively with the agent first until the output matches your exact vision.

2Step 2: The Magic Prompt: Reverse-Engineer into a Skill

Try this prompt in your agent:
$ I love this output! Please turn this entire process into a reusable skill called data-audit-report so I can run it anytime.
What you will see on screen:
Analyzing workflow steps... ✓ Created reusable skill: data-audit-report.md ✓ Saved to your local skills library.
Under the hood:The agent reviews the entire conversation, extracts the prompt logic, formatting rules, and validation checks, and writes a permanent `SKILL.md` playbook.

3Step 3: Run Your New Skill with a Slash Command

Try this prompt in your agent:
$ /data-audit-report analyze the latest batch in ~/Downloads/new-batch
What you will see on screen:
Executing skill /data-audit-report... Applying color-coded risk dashboard -> Formatting summary table -> ✓ Finished in 18 seconds.
Under the hood:In any future chat, just type '/' and select your skill. The agent runs the full verified playbook instantly.
Example: data-audit-report.md (Generated Skill)markdown

The agent generates clean, reusable instruction playbooks for you automatically.

---
name: data-audit-report
description: Scans data logs, categorizes risk levels with color formatting, and outputs an action plan.
---

# Operational Steps
1. Ingest target CSV/JSON files from specified folder.
2. Validate data fields and calculate summary metrics.
3. Generate formatted dashboard with High/Medium/Low indicators.
4. Output a 3-step recommended action plan.

Do:Build skills via 'Workflow Distillation'

The easiest and most reliable way to create a skill is to do the task manually first, get the exact output you want, and then tell the agent: 'Turn this into a skill'.

Avoid:Writing abstract skill instructions from scratch

Avoid guessing instructions in advance. Let the agent extract instructions from real, verified work.

Do:Improve skills every time you get a great result

If a skill run produces an exceptionally neat diagram or layout, tell the agent: 'Update the skill so it always formats charts this way'.

Hands-On Practice Checklist

Try each of these steps on your computer and check them off as you complete them:

Module 5

Visual Assets & Diagrams: Generating Media Right on Canvas

Module Learning Goal

Learn how to generate technical diagrams, flowcharts, and custom imagery directly inside your project without leaving Codex.

Great technical and business work requires visual clarity. Codex includes built-in multimodal generation and diagramming capabilities. You can ask for system architecture diagrams, process flowcharts, or visual illustrations, and the agent saves them directly as image files in your project folder.

1Step 1: Generate Visual Flowcharts & Diagrams

Try this prompt in your agent:
$ Please create a visual flowchart diagram showing our 3-step user onboarding process. Save it as a clean image in my project folder.
What you will see on screen:
Invoking diagram generation skill... ✓ Generated onboarding-flowchart.png (Saved to project directory).
Under the hood:The agent generates the graphic locally. You can immediately preview it in Codex or embed it into your documents.

2Step 2: Generate Custom Imagery for Presentations or Web Pages

Try this prompt in your agent:
$ Generate 3 clean, professional product mockup illustrations with modern lighting and a dark background for our landing page.
What you will see on screen:
Generating visual assets... ✓ Created asset-1.png, asset-2.png, asset-3.png in project folder.
Under the hood:All images are saved directly on your computer so you can use them in your slide decks, websites, or reports.

Hands-On Practice Checklist

Try each of these steps on your computer and check them off as you complete them:

Module 6

Hands-Off Automation: Computer Use & In-Browser UI Testing

Module Learning Goal

See how the agent controls your mouse, keyboard, and browser to test apps and automate repetitive desktop tasks.

One of Codex's most groundbreaking capabilities is Computer Use and Browser Use. When you enable `@computer-use` or `@browser-use`, the agent can literally see your screen, move the mouse cursor, click buttons, type text, and test web pages just like a human operator.

1Step 1: Automated Desktop Software Control (@computer-use)

Try this prompt in your agent:
$ Open my presentation software, create a new 5-slide presentation, and place our generated charts on each slide using @computer-use.
What you will see on screen:
Scanning desktop applications... Launching presentation software -> Creating slide deck -> Inserting chart images -> Saving presentation.
Under the hood:Watch your screen: The agent moves the mouse, creates the deck, inserts the images, and saves the file automatically.

2Step 2: Automated In-Browser Testing & QA (@browser-use)

Try this prompt in your agent:
$ Build a responsive quiz app in index.html, then open the browser and test that every button, navigation link, and quiz answer works correctly using @browser-use.
What you will see on screen:
Opening browser preview at localhost:3000... AI taking control of browser -> Scrolling through layout -> Clicking start button -> Testing quiz answers -> Validating results screen -> ✓ All UI tests passed.
Under the hood:The agent opens the browser, interacts with every component, verifies smooth scrolling, tests button clicks, and confirms there are no broken links.

Common Beginner Pitfall & How to Fix It

What happens: The agent clicks the wrong area on desktop apps

How to solve it: Keep the target application window clearly visible on your screen and specify exact button names in your prompt.

Hands-On Practice Checklist

Try each of these steps on your computer and check them off as you complete them:

Module 7

Putting Tasks on Autopilot: Scheduled Background Automations

Module Learning Goal

Learn how to turn verified skills into hands-free recurring background jobs that run every hour, day, or week.

Once you have built a skill you love (Module 4), you don't even have to remember to run it manually! With Automations, you can tell Codex to run that exact skill on a schedule in the background, updating your spreadsheets or summaries automatically.

1Step 1: Schedule Your Reusable Skill

Try this prompt in your agent:
$ Please run our /data-audit-report skill every Friday at 9:00 AM automatically and update the summary spreadsheet.
What you will see on screen:
✓ Created recurring automation: 'Weekly Data Audit' ✓ Schedule: Every Friday at 9:00 AM ✓ Target Skill: /data-audit-report ✓ Status: Active in Automations dashboard.
Under the hood:Codex sets up a background cron schedule. Every Friday morning, your agent wakes up, runs the skill, processes the data, and updates your file.

2Step 2: Manage & Inspect Your Automations

Try this prompt in your agent:
$ Navigate to the Automations tab in Codex
What you will see on screen:
View active schedules, check last run status, review generated files, or pause/edit frequencies anytime.
Under the hood:You have full visual control over all running schedules from the Automations tab.

Do:Test skills manually before scheduling them

Always run a skill 2 or 3 times in chat to make sure it produces flawless results before putting it on a recurring schedule.

Avoid:Scheduling uncontrolled write operations

For important workflows, configure the automation to write reports to a review folder so you can verify changes when you start your day.

Hands-On Practice Checklist

Try each of these steps on your computer and check them off as you complete them:

Module 8

Bonus Superpower: Ambient Screen Context (Zero-Upload Assistance)

Module Learning Goal

Learn how ambient visual context allows your agent to understand what's on your screen in real time without taking manual screenshots.

Have you ever wanted to ask an AI a question about what you're currently looking at without having to take a screenshot and upload it? With Ambient Visual Context enabled in settings, Codex observes your active workspace in real time and can answer questions or generate companion documents instantly.

Pro Tip for Beginners

👁️
How to Use Ambient Context:Simply have your active app or document open on your screen, switch to Codex, and prompt: 'Look at what is currently on my screen and write a 5-bullet summary of the key action items'.

Hands-On Practice Checklist

Try each of these steps on your computer and check them off as you complete them:

Crash Course Starter Files & Templates

Codex Beginner-to-Pro Starter Pack

Includes copyable `agents.md` memory templates, standard `SKILL.md` recipes, and recommended plugin configuration blueprints.

Download Starter Pack (.zip)

Ready to Take Your Skills Further?

Explore our developer guides, reference playbooks, and video courses to keep leveling up.