Codex App Practical Guide: Secret Tips for Productivity and Profit

In my previous post, I covered the basics of Codex App from 0 to 1, focusing on the interface and basic configuration. Now that the foundation is laid, it's time to use Codex App for tasks that actually improve efficiency and reduce costs, fully empowering your business.

Today's article will take you into real-world scenarios to experience the hidden secrets of Codex, from basic to advanced, and from functions to scenarios, helping you master Codex in one go! 🔥

• Essential for corporate grinders: Use Codex to create the 'Office Four'—Word, PDF, PPT (Image-based vs. HTML-based), and Sheets.
• Hand-crafting cool 3D videos: Use Codex + HyperFrames / Remotion (three.js) to create 'Top Ten Chinese Animated Movies'.
• Early to bed, early to rise: Use Codex to build a beautifully typeset website analyzing the benefits of early sleep, deploy it on Vercel, and then use Codex to control OpenRouter for comparative experiments with the new Ring model.
• Essential for self-media creators: Use Codex to create long graphic tutorials—let Codex control itself, run multiple parallel dialogues, and handle screenshots and blurring in one go.
• E-commerce automation: Codex + Playwright to control the browser, achieving painless bulk product listing on Pinduoduo.

Learning while playing, and making money along the way—that's my goal, and I share it with all of you. Without further ado, let's get started!

1. Essential for Corporate Grinders: Creating Word, PDF, PPT, and Sheets

Previously, if you wanted to create an activity plan, project report, course delivery, consulting report, or employee training materials, you might have spent months mastering these tools. But now, in Codex, a single prompt can generate beautifully typeset and data-accurate Word, PDF, PPT, and Sheets files.

To demonstrate this, I created a demo:

Please perform an office suite demonstration task in the current dialogue with the theme 'Codex App Content Production Workflow'. Requirements: 1. Generate real Word, PDF, PPTX, and XLSX files; 2. Each file must have actual content, not empty; 3. Tell me the real path of each file after generation; 4. Try to generate a preview or file list for my tutorial screenshots; 5. Finally, provide an acceptance checklist to confirm all four files exist, can be opened, and are not placeholder content.

Acceptance results:

1. Word document verification screenshot:

1. PDF file verification screenshot:

1. PPT presentation verification screenshot:

1. Sheets spreadsheet verification screenshot:

Even without optimized skills or prompts, these results are excellent! Currently, the hottest PPT method on X is HTML-based drawing, which breaks tradition. Another alternative is generating image-based PPTs.

Route 1: Image-based PPT

This means letting Codex generate 16:9 visual images first, then placing them into the PPT. Each slide is essentially a full image. The effect is quite good because the underlying model is GPT2-image. These are still editable through various methods; DM me if you're interested in the documentation.

Route 2: HTML-based PPT

This is a very popular route lately. Instead of traditional software, you use HTML, CSS, and JS to write slides. The benefits include browser preview, animations, interactivity, code-controlled layouts, and easy exporting. Many open-source skills for this are available on X via Grok search.

2. Hand-crafting Cool 3D Videos: Codex + HyperFrames / Remotion for 'Top Ten Chinese Animated Movies'

Let's move from work to something cooler. I used Codex + HyperFrames / Remotion (three.js) to create a 3D video. The core logic: Codex writes code and renders, three.js handles 3D visuals, Minimax handles narration, yt-dlp downloads clips, and Remotion turns web animations into video.

Steps:

1. Explain the requirement clearly to Codex.
2. Align requirements—ask Codex to stop and ask if anything is unclear.
3. Enable 'High' and 'Full Access' for @browser and @computer use.
4. Get the final product and refine it.

This validates the chain of 'auto-scripting -> material search -> auto-download -> highlight identification -> auto-dubbing -> auto-editing'. This can be commercialized for AI traffic monetization, e-commerce, and advertising.

3. Early to Bed, Early to Rise: Build a Website with One Prompt and Deploy to Vercel

Many AI enthusiasts have messy schedules because AI is too fun. I used Codex to build a website themed 'Early to Bed, Early to Rise' to promote health. Building a small app or site is simple with a prompt like:

Please generate a website about sleeping early and waking up early. Requirements: 1. Beautiful layout; 2. 3D animations; 3. Analyze benefits; 4. Execution steps; 5. Real runnable site; 6. Start local preview and deploy to Vercel.

Codex created more than a static page (deployed at: https://codex-1-html-css-js-2.vercel.app/); it included a 3D circadian rhythm scene, plans, and self-testing scores.

You can also use Codex to test new models on OpenRouter. For example, I tested the new Ring-2.6-1T (free) model from Ant Group. Since Codex couldn't directly control my local terminal for Claude Code, I had it open the OpenRouter website and generate a single-file HTML site using the Ring model.

You can see the Ring-generated version here: https://ring-sleep-site.vercel.app/

This is a great way to test model capabilities for free. You can even use this logic to leverage GPT-4o Pro via the web interface through Codex.

4. Essential for Self-Media: Creating Long Graphic Tutorials

This is a scenario I use often. My previous viral tutorial was 80% completed by Codex, including text, screenshots, and blurring. The key is 'Prompt Accuracy' and 'Style Transfer'. You must provide your logic and structure, not just a vague direction.

I've distilled my writing habits into a 'Yichen tutorial writing' skill for Codex to call. You can even have Codex open multiple dialogues to monitor itself and run tasks in parallel, drastically increasing efficiency.

5. E-commerce Automation: Codex + Playwright for Pinduoduo Bulk Listing

For a Pinduoduo merchant, the pain point was manually filling thousands of rows of SKU data into the backend.

Initially, I suggested Computer Use, but Pinduoduo blocked it. So we switched to Playwright. Playwright controls the browser by identifying elements rather than just 'looking' at the screen, making it more stable for bulk tasks.

Codex handled the entire chain: reading Excel, mapping fields, writing Playwright scripts, and logging results. This turned a one-time task into a reusable company process.

6. Common Pitfalls and Solutions

• Pitfall 1: Computer Use isn't universal. Some sites detect automation. Use Playwright with stealth and reuse logged-in browser instances.
• Pitfall 2: Codex can't control open local terminals. It can see the screen but not send commands to external processes. Use the web version or Codex's built-in terminal.
• Pitfall 3: Giving directions instead of logic. If you just say 'write an article', the output will be garbage. Provide a framework and your own high-quality examples for style imitation.

7. My Understanding of Codex

Codex is more than a chatbot. It can produce content, analyze data (like WeChat logs), automate browsers, and supervise other AI tasks. Its potential is limitless.

8. Recommended Learning Path for Beginners

1. Generate images.
2. Generate Office documents.
3. Create a small HTML page.
4. Local preview and Vercel deployment.
5. Graphic tutorials.
6. HTML PPTs and video animations.
7. Business automation.

9. Pricing and Usage Notes

The free version is very limited. I recommend a GPT Plus or Pro subscription. If you encounter proxy issues, ensure you configure WSS_PROXY and WS_PROXY in ~/.codex/config.toml and use launchctl setenv for macOS.

Happy Codexing! If you have complex business scenarios that need optimization, feel free to reach out.