Summary of My 8 High-Frequency Use Cases for Codex

Let me summarize the main scenarios where I use Codex. Coding is obvious—that’s its core strength. But beyond coding, I’ve found this tool gradually integrating into every aspect of my daily work.

1. Creating Illustrations

Whether I’m writing articles, preparing PPTs, or even internal company documents, I now rely on Codex for most of my illustration needs. Depending on the scenario, this can be divided into three categories.

Type 1: Informational Illustrations

I directly feed the article to Codex, then invoke my pre-built “Article Illustration” Skill.

This Skill first reads through the entire article, identifies where illustrations would be appropriate, and then confirms with me. After confirmation, it generates an HTML-based illustration page directly.

I really like this approach because the illustration is essentially a web page. If anything isn’t right—whether it’s the text, layout, or design—I can modify it quickly without needing to regenerate everything.

Moreover, the more I use it, the more I realize that for many article illustrations, you don’t actually need a model like GPT Image 2.

For conveying information, simple graphics, lines, and layouts are often sufficient. Sometimes simplicity helps readers focus on the content itself.

For example, the illustration below is actually just a web page.

Type 2: Hand-drawn Style Mind Maps, Flowcharts, Sequence Diagrams, Architecture Diagrams, etc.

In my work, I often need to draw flowcharts or organize logical relationships. I’ve always liked using Excalidraw because the output naturally has a hand-drawn feel—it doesn’t look stiff and has a nice character.

Now, my workflow is: first, throw a piece of text or an idea into Codex and let it break it down for me. For example, if this content needs to be turned into a flowchart, it identifies the required nodes, relationships, and how they should be structured. Once the overall framework is confirmed, I call my specially made Excalidraw Diagram Skill to generate it.

This Skill outputs two files simultaneously: a PNG and an Excalidraw file. The PNG can be used directly. The Excalidraw file can be re-imported into Excalidraw for further editing.

Skill link: https://github.com/coleam00/excalidraw-diagram-skill

Type 3: Creating Fun PPT Illustrations

I discovered this Skill on X (Twitter), and it’s incredibly useful. It can automatically transform dry, tedious content into highly stylized hand-drawn illustrations.

The key feature of this style is expressing originally serious or abstract content through hand-drawn cartoons. The images include various little characters, props, scenes, and exaggerated visual elements, making it feel like telling a story.

Link: https://github.com/helloianneo/ian-xiaohei-illustrations/tree/main

2. Cleaning Up Computer Disk

If your computer has limited storage space and you often need to clean up disk to free up room, I think with Codex, there’s really no need to buy specialized disk cleaning software anymore.

My MacBook at home doesn’t have a large storage. This year, I installed a lot of Agent products on this machine, but many were just for trial purposes—installed and quickly uninstalled.

However, I later discovered that simply dragging them to the Trash doesn’t completely remove them. Many Agent software packages have multiple dependencies.

Then it occurred to me: why not let Codex help me check?

So I asked it to scan the system and look for leftover files from previously installed Agent software, and identify which associated components could be safely removed. It found quite a few lingering related files that hadn’t been uninstalled properly. And they easily took up several GB.

3. Converting Long Articles into Slides

I set a rule for myself: read at least one long article carefully every week.

By “long articles,” I mean those truly worth the time—long blog posts, in-depth interviews, or research articles—not news or fragmented information.

Because I increasingly believe that in an age of information overload, simply scrolling through news has limited value. Much of it feels like you know it at the moment, but you forget after two days.

In contrast, long-form articles that have been thoroughly thought through and refined are more likely to bring new insights.

Now my approach is much simpler. After finishing an article, I directly hand it over to Codex and ask it to generate an HTML-based slide deck.

The slides don’t need to cover everything—just the core ideas, key arguments, and overall structure.

Once generated, I go through the slides again, effectively reviewing the article a second time.

This process is quite interesting. The first reading is about absorbing information. The second time, using the slides, it’s more about organizing my understanding.

Often, details that I missed during the initial reading resurface at this stage.

Skill link: https://github.com/gainubi/note-slides

4. Processing Meeting Minutes

Nowadays, various software have built-in meeting minutes features, but I still prefer to export the minutes as Markdown files and then hand them over to Codex for processing.

This method is especially useful when multiple meetings have been held on the same topic.

For example, if we have three or four meetings around a project in one week, I’ll throw all the minutes into Codex at once and ask it to analyze from an overall perspective, rather than focusing on just one meeting.

The most straightforward use is, of course, organizing Todos.

First, I ask it to extract all action items. Then I follow up with the question: “Among these Todos, which ones can you directly help me complete?”

Interestingly, many of them are things it can actually do. Tasks like organizing materials, writing drafts, doing research, analyzing data, consolidating information, generating documents, or preparing proposals—many of these can be passed directly to Codex for execution.

For client meetings or user interviews, I ask it to identify the issues users care about most, the most frequent complaints, and feedback that might be easily overlooked but worth noting.

Sometimes, to collect user feedback, we gather materials from multiple sources. These might include WeChat group chat logs, user interview notes, or survey responses. Previously, this scattered information was hard to analyze holistically.

Now, I consolidate all these materials and dump them into Codex, asking it to perform user pain point analysis, demand categorization, and insight extraction.

Especially when the data volume is large, I increasingly prefer handing this work to Codex. It can read vast amounts of material simultaneously and quickly discover correlations between different pieces of information.

5. Integrating with Feishu (Lark)

I wrote a dedicated article about this last week, so many people probably don’t know about it yet.

If you want to use Codex on your phone, besides the ChatGPT App, you can also integrate it directly with Feishu. The process is very simple—similar to how you would connect OpenClaw to Feishu—and can be done in minutes.

My original intention for connecting Codex and Feishu was to operate Codex via my phone. But after actually using it, I found myself preferring to use Codex from within the desktop Feishu client.

Because Feishu is already my work hub. Chat logs, meeting minutes, documents, project materials—a lot of content is already in Feishu.

For example, seeing a group chat discussion, I can forward it directly to Codex and ask it to summarize the key points. After a meeting, I can toss the minutes to it for analysis.

What’s even more interesting is Feishu Documents. Previously, if I wanted Codex to help me revise some content, I had to copy the content out and then throw it into Codex.

Now, I often don’t need to go through that trouble. For instance, while writing this article, if I feel a certain paragraph could be smoother or I want to explore another angle, I just @ Codex directly within the Feishu document.

6. Integrating with WeRead (WeChat Reading)

WeRead recently released a Skill that can be installed with one click. Just search for it.

Let me also share my reading habits.

Apart from literary works, I rarely read a non-fiction book from cover to cover. Many practical books aren’t meant to be read like novels; the truly valuable content is often concentrated in a few chapters.

So I often open WeRead first and look at the book’s highlighted passages. After browsing through the most popular highlights, I already have a good grasp of the core ideas and most insightful content. Then I pick specific chapters to read in detail based on my interest.

This is also why I renew my WeRead subscription every year. The popular highlights feature is incredibly useful.

In a way, it’s like thousands of readers have already done a preliminary filtering for us. Content worth pausing to think about, paragraphs that resonate most—people have already voted with their highlights.

And integrating Codex has made this even more convenient.

Since I often write articles, I frequently encounter a situation: I clearly remember reading a great point or a particularly inspiring sentence in a book, but I can’t recall exactly where. Previously, I had to go back and search through the book manually. Now, I simply ask Codex to search WeRead for it.

For example, I can ask it to find relevant content from popular highlights on a specific topic, or to locate a particular author’s perspective on a certain issue, or to compile common key concepts from popular highlights across several books.

For someone who reads and writes regularly, this experience is quite satisfying. The content in the books hasn’t increased, but the cost of accessing that content has been significantly reduced.

7. Deploying Websites

Nowadays, everyone is Vibe Coding their own small products, involving both frontend and backend. I’ve personally Vibe Coded 11 small tools myself.

Many of these tools have backends and need to be deployed to a server. I’m currently using Tencent Cloud’s CloudBase, which also has a CLI. So once authorization is done once, all subsequent deployments can be handled within Codex—no need to go to the CloudBase console for various configurations.

8. Handling Daily Company Chores

We are a small company, so many basic tasks that used to require administrative, HR, financial, or even legal staff are now mostly handled by Codex.

For example, when I receive a contract, I usually throw it directly into Codex. I ask it to check for risks, summarize key clauses, or modify it according to our needs. For many standardized contracts, it handles them quite maturely.

Similarly, documents like company policies, recruitment materials, or process documents can all be drafted by Codex first.

But what I find even more valuable is the impact of integrating it with Feishu data.

A lot of our company’s operational and financial data is stored in Feishu’s multidimensional tables. After installing the Feishu CLI for Codex, it can directly connect to this data.

This means Codex is no longer just a chat tool—it becomes an active participant in daily operations. Tasks like adding data, modifying data, or batch-checking data can all be done through it.

I also have some fixed automated workflows.

For example, every morning, I ask Codex to check the previous day’s data. It looks for missing records, anomalies, or anything clearly unreasonable.

People often make errors when entering data manually. But having Codex run a daily check means many issues are caught the same day. At the end of the month, I just run one command, and Codex pulls together the month’s operational summary.