Claude Code Debugging for Beginners: Fixing an AI eBay Listing App

By Beau Johnson··8 min read

Quick answer

On day 14 I use Claude Code to debug my AI eBay listing app. I run the photo analysis built on Google Vision and GPT vision APIs, paste errors and console logs back to Claude, add eBay item specifics and a price optimizer, and start moving from the eBay sandbox to the production API.

In this video I catch you up on day 14 of building an AI listing app for resellers, then spend most of the time debugging it with Claude Code. It's for beginners who want to see what vibe coding looks like when things break and you have to keep prompting until they work. I recorded this in August 2025, and tools change fast, so check the current docs before you copy any of my setup.

What I cover in this video

  • 0:00 Where the project stands after 14 days, and why I redid my sub agents directory in Claude Code.
  • 0:46 The landing page, the waitlist, and what the product does for resellers.
  • 2:16 The main workflow: upload photos, assign a SKU number, and generate a listing.
  • 3:01 The first AI analysis using Google Vision and GPT vision APIs, and what it got wrong.
  • 4:32 My first prompt to Claude asking for eBay item specifics and condition detection.
  • 8:18 Hitting an error and pasting it straight back to Claude.
  • 10:37 Sending console logs, two failed analyses, and then a much better title.
  • 13:41 Moving from the eBay sandbox API to the production API with Netlify environment variables.

What the app does after 14 days

I got the idea from Idea Browser, and I picked it because I'm already in the reselling niche and know the pain points. Listing items takes a lot of time, and some people pay virtual assistants to do it. My app lets you take photos of an item and get it listed on marketplaces like eBay, Poshmark, and Mercari much faster than filling out item specifics and descriptions by hand.

Day one was the landing page. It isn't final copy. It explains the idea and collects waitlist signups, and we already have emails coming in. After you sign in you land on a dashboard. The vision for the overview page is to show your sell-through rate and the products selling well in your store, connected to your eBay account, with AI analyzing those numbers so you can make better decisions.

The main feature works like this:

  1. Upload your photos.
  2. Assign a SKU number for inventory tracking.
  3. Go to the generate listings page, select the item, and run the AI analysis.

The analysis uses Google Vision and GPT vision APIs to pull keywords, brand, size, and color into a listing. My first test returned Wall Street bull hoodie men medium black pullover graphic long casual cotton. It got the size and the right category, but the keywords weren't specific enough, the pricing wasn't good, and it didn't detect the condition.

Why I moved from Bolt to Claude Code

I started this project in Bolt and made a lot of progress there. The front end, the back end, Supabase, and GitHub all got set up. Then it got hard to make small changes without breaking the back end or Supabase, so two days before this video I started working in Claude Code.

I also had to redo my sub agents directory because I hadn't set it up correctly. Now I have three sub agents connected to the project. I plan to keep asking Claude Code which sub agents and hooks make sense for this app and document what I learn.

The debugging loop: test, paste the error, explain what I want

I'm not a veteran software engineer, so my process is simple. I run the workflow, look at what's wrong, and tell Claude what happened and what I want instead. Here is the first prompt I gave it:

So the AI analysis returned successfully, but I don't see any item specifics filled out. Is there a way that we could pull item specifics that are required and recommended on eBay for listing clothing? I don't see it anywhere in our dashboard when the AI analysis happens. Also, the AI analysis failed to get the condition of the item.

Claude added an eBay item specifics section to the item details page. It told me the data was already being generated by the analysis pipeline and was not being displayed in the UI.

When I tested again, I got an error. I copied it and pasted it back to Claude. The logs showed the analysis was extracting the information but flagging it for manual review, so Claude went to work on the condition detection.

The next run came back without the color in the title and without the full item specifics in the UI. I told Claude exactly that. It asked me to open the debug section on the item details page so it could see the raw AI analysis data, and I pasted in the console logs. The analysis failed twice more. I sent the logs each time. Then it came back with a great title that included the color and the size.

From there I gave Claude everything I was thinking at once: the edit listing modal still didn't show the required and recommended eBay item specifics, the item wasn't showing up on my dashboard overview page, and I wanted an eBay price optimizer in the modal so users can see what an item sells for, review the listing, and publish straight to eBay. Claude added basic info, eBay item specifics, and a price optimizer as three tabs. I liked the changes and asked it to combine them onto one page.

Switching from the eBay sandbox to the production API

Some recommended item specifics were still missing, and I asked Claude whether the eBay API was returning accurate sales data. We had been using the eBay sandbox API, which was good for testing. For real live data we need the production API, which I was approved for a couple of weeks earlier.

This matters because each item type on eBay, like a t-shirt, a hoodie, or a jacket, has a different set of item specifics. The AI analysis has to read those and prefill the required fields before you list.

To make the switch I added the production environment variables in Netlify, asked Claude how to enable production mode, and redeployed the project so it picks up the changes. The video ends there, so I don't show the results of the switch yet.

Key takeaways

  • Pick an idea in a niche you already know. I know the listing problem because I've lived it as a reseller.
  • Get your main feature working as well as it can before adding more. I have plenty of other ideas, but the AI listing analysis comes first.
  • When you hit an error, paste the error or the console logs back to Claude and explain what you expected to happen.
  • Give Claude all your thoughts and corrections in one message. It summarized mine into a clear list of fixes.
  • A little progress every day adds up. If you want to build an app with AI, expect errors along the way and keep going.

FAQ

Why did you switch from Bolt to Claude Code?

I built the front end, back end, Supabase, and GitHub setup in Bolt, but it got hard to make small changes without breaking the back end or Supabase. Two days before this video I started working in Claude Code instead.

How do you fix errors in Claude Code if you can't code?

I copy the error or the console logs, paste them back to Claude, and explain what isn't working and what I want it to do. In this video that took several rounds before the AI analysis returned a good title with the color and size.

Which APIs does the listing app use to analyze photos?

The project has Google Vision and GPT vision APIs plugged in. They pull keywords, brand, size, and color from the photos to build the listing.

Why move from the eBay sandbox API to the production API?

The sandbox was good for testing, but I want real live data and the full set of item specifics for each category. I added my production environment variables in Netlify and redeployed the project to start the switch.

Transcript

Read the full transcript

0:00We started this project just 14 days ago and I want to catch you up to speed as the progress we have made so far. We have a lot of the features that have been implemented and I am excited just to share you where we've come from in just 14 days and show people just like yourself that you can do this as well if you just have an idea and you want to put it into action. So, we've been using a lot of Claude code lately, and we had to go back and redo our sub aents directory because we actually um really need to put these sub aents to work, but we hadn't set it up correctly. So, I had fixed that and did that. And now we have three sub aents that are connected to our program and to our product that's going to help us just make more progress

0:46quicker. So, we started out in Bolt. Now, we're using Cloud Code, and it's been really helpful. I want to show you the progress we've made on the actual front-end side and kind of the workflow on how this vision is coming to life. Day one was just working on the landing page, and this is not a final copy. So, this was just built to portray the idea, what our product does, and have people join the weight list. So, that's it. It's basically a way for people to take photos of items and get them listed onto marketplaces like eBay, Poshmark, Merkari, much quicker than actually manually filling out item specifics and descriptions. So, after you sign in, you then get access to the dashboard. And I want to take you through the workflow as to what we have built out so far. The

1:31dashboard is still going to be worked on, but at least we have a working product for the most part. So, after you enter the dashboard, the overview page is going to give you a kind of breakdown of your current sell-through rate, products that are selling well in your store, you're going to be able to have this connected up to eBay and your account. So, you're able to track analytics and then also have AI analyze those analytics so you can make better decisions as a reseller. So, that's where the overview page, the vision for it is. and basic function. The Halo feature of our product is basically taking photos that you can input into our AI analysis and get a title, get the item specifics done and filled out, and

2:16then also get it priced up. So, all of this is going to be done for you by you just simply uploading your photos. And it's going to make it a better user experience for people that want to just flip items real quick. And um you know it's going to be for the casual user and also for the full-time resellers out there. This is a huge pain point in the industry by the way. So yes, we're going to want to upload our photos. Here are some test photos we are uploading. After they get uploaded, that is when you assign a skew number for inventory tracking. We will go with one. Once the skew number is assigned, then it takes you to the generate listings page. You can see here nothing has been generated. There's no price. There's no knowledge of this item as at all. So, you're going

3:01to go ahead and select the item and then have it do the analysis. We have Google Vision and GPT vision APIs plugged into this project. So, by using that technology, you're hoping to get keywords, brand, size, color, obviously, all of that information put in to a listing essentially. So this is the main feature of the product and we want to focus on making this the best it can be. So it's analyzing you can see here it got Wall Street bull hoodie men medium black pullover graphic long casual cotton. So not the best keywords but still it's much more progress than we've

3:46made just a few days ago. So let's see here. I wanted it to be really specific um with the keywords. The pricing isn't all that good. Um but at least I got the size. It didn't analyze the condition either. So in this video, we're kind of just going to be troubleshooting a lot of the things that we want implemented into our project and seeing us work through those. So it got it in the right category, basic info. Um that's great. So once that is done, um it then would essentially be listed onto eBay. So we want to go back to Claude and say, "So the AI analysis returned successfully, but I don't see any item specifics

4:32filled out. Is there a way that we could pull item specifics that are required and recommended on eBay for listing clothing?" I don't see it anywhere in our dashboard when the AI analysis happens. Also, the AI analysis failed to get the condition of the item. And just to give you some context, this idea was just started 14 days ago. Ideabrows.com, that's where I got the idea. Why we started to implement this idea was because I'm already involved in the niche. I've already done it. I've already known the pain points and the problems. This solves a big problem which is listing items and people don't have to pay virtual assistants to list items now. They don't have to waste time

5:18listing them their items by themselves. That's where the AI integration comes in. People can just upload the photos, have listings done in seconds. And that's kind of the problem we're solving within this niche. And I'm showing people like yourself that you can build something similar to maybe you're involved in a niche already and you have an idea that solves a painoint. Today's the day to be building it. We started this project off in bolt.new and we ended up making major progress getting the front end, the back end superbase integrated uh GitHub all set out and then we just got to a point where it was really difficult to make subtle changes without breaking like the backend or breaking superbase things like that. So then we just started two days ago starting working with Claude Code and

6:03just getting familiar with using it and you're going to see a lot more videos like this in the series of me just figuring it out asking Claude code what is the best sub agents for our program what is you know maybe some good hooks and hacks that we could be using for our project and that's all going to be documented here. Amazing. So we've made major progress. I love seeing it come together and I do believe that we're going to be able to launch this product within the next 30 days as I'm going to be working on it every day, releasing a video every day, keeping you up to date on the progress that we're making. We already have an email list. We already have email signups. Right now, we're working with kind of just enhancing the

6:48listing process, building in the features and the UI, pulling the data from eBay and making it a great user experience. Like I said, it's very important for me. Um, and I think for most people to focus on the main selling point of your product and until that is like the best it can be, then I I think is good to start implementing more features, which I have tons of ideas, but I want to make sure that this feature works so good. Then we can move on to other things, but I want to just make sure it works the best it can. Um, so you see, uh, perfect. I've successfully added the eBay item specifics display section to the item details page. So, it implemented all of that. Um, AI quality metrics, smart

7:33filtering. Okay. Um, now when users run AI analysis on clothing items, they'll able to see all the comprehensive eBay specific data that was previously hidden. This addresses your request. The data is already being generated by the AI analysis pipeline. it was just not displaying in your UI until now. Well, let's check it out. All right, let's see if it implemented those changes. I keep telling myself, as long as we're just making progress, that's all that matters. So, that's why I'm hopeful and happy to share all of everything that we're going through because it doesn't really matter, you know, where you're at in your project. As long as you're just like making a little bit of progress, a little bit of progress, a little bit, little bit every day worked on, over

8:18time, you're going to see results. So, let's go through the workflow again and see what changes it made. All right, so back in the generate listings page, we're just working with the same item because I want to see what changes it makes. Error. All right, so let's copy paste the code. Just something we've been good at recently. And whenever there's an error like that, I just give it back to Claude and have it fix it because like I said, I'm not a 30-year veteran of software engineering and coding. But this is the nice thing to be alive in this time to be able to process errors just like this and learn along the way. I think it's important to get at least some understanding on why it wasn't working. So the logs show that the AI analysis

9:03was working and extracted the information and it was successfully extracted. However, I notice the AI analysis is being flagged for manual review. Let me check the condition detection issue and work on improving it. I love it. I love it. All right. So, it has implemented those changes. Last time we got an error. See what happens. All right. Well, it came back. I was hoping in this table it would have all the item specifics filled out. Also didn't put the color of the item in the title. So, we're going to go back to Claude. The AI analysis worked, but I don't see the color of the item in the title. And also I don't see full eBay item specifics displayed in the UI after the item is analyzed. I can click edit

9:50but I don't see anything on the page. And this is why it's so good for new beginners to work with Claude and just basically do what it you know like explain why it's not working and what you want it to do and why you want it to do it. So, it's going to investigate why eBay item specifics are not displayed in the UI. Fix color detection in for the title. Check why edit page shows nothing. So, I'm just going to go ahead and continue. All right. So, Cloud wants me to navigate to the item details page and go to the debug section showing the raw AI analysis data. And then it will help it understand what is actually stored in the AI analysis field and whether the eBay item specifics has any data. So I'm going to go ahead and go

10:37back in. I copy pasted that info from the console logs and I'm copying it into here. Also some other things that I noticed. So I see that the title was very generic and also it didn't capture the size. Here's the console logs. All right, so we fixed some of the errors and let's see how it goes. Analysis failed. Analysis failed. Here's the logs. Let's give it another shot. See what happens. Much better. Much better. Great title. Amazing. So, the edit listings page is not it's not it's just not how if I were to publish this listing to eBay, it's not putting what

11:22eBay requires into this popup modal. So, uh let's fix that. For the most part, it looks pretty good, though. Um, so I will tell Claude that. All right, a couple things. The AI analysis worked. The title looked great. It got the color. It got the size. But the pop-up modal when clicking edit listing after the AI generation still does not show the required and recommended eBay item specifics for that particular item in that particular category. Also, after generating the AI analysis, the item fails to appear in my dashboard overview page. Is there any way we can fix that and also implement the eBay analysis

12:09price optimizer agent within the edit listings modal so users can know what that item sells for so they could price it accurately. Ideally, after the AI generation, we would want them to be able to review that listing and publish straight to eBay right after the AI analysis. Sometimes you just got to give Claude all the things you're thinking about. All right, after kind of just dumping all the ideas and the corrections we needed to make, it summarized edit listing modal eBay specifics. So it should show that um dashboard refresh issue and eBay price optimizer is now integrated into that workflow after the AI analysis direct

12:56eBay publishing. Amazing. Let's go check it out. Okay, so it did add in some pretty good features. Um I do like, you know, the basic info, the eBay item specifics. I think that these should be combined in one page. And then the price optimizer. So, very cool. And I'm just going to make those suggestions um to Claude. I like the changes. They look great. But I think basic info, eBay item specifics, and price optimizer should just be on one page instead of three different tabs. Also, are we getting all of the item specifics from the eBay API? because it looked like we were missing

13:41some recommended item specifics according to the item category we are listing. And also is the eBay API returning accurate sales data based on the item we are analyzing and listing? We are still using the eBay API sandbox. I wonder if we need to switch to the production model. Just give it all. Give it all the information, all your thoughts. Okay, so in this project, we've been working on the eBay sandbox API, which has been good for testing things out, but where we're seeing where we want like real live data coming in is we need to switch to the production API that we have. We just got approved a couple weeks ago. So, I just need to update that in Netlefi and hopefully

14:28that will help with the item specifics not coming through on our end, which is what we really want. And each item that you list on eBay, say it's a t-shirt versus a hoodie versus a jacket, is going to have a different set of item specifics that need to be filled out in red by the AI analysis to be able to bring that information and prefill out all of that required information before you list an item. So, that's something that we're going to switch into and see how it all works. So to add in the production model, I need to add in those environmental variables in netify like I just said and that is going to enable us just to get much more data. So I'm going to go ahead and do that.

15:13Okay, I added those environmental variables into Netlefi and I do have a production API key. I do have a eBay developer account. Now it says to enable production mode. Can you do that on my end or how do I do it? Okay, after adding those environment variables into Netlefi, I'm going to deploy project so it accepts the changes. All right, so we're making progress during this project. And I know it may seem slow, but I'm going to continue to document the journey. If you have questions, please comment down below. Let me know what I'm doing wrong. Let me know what I could be doing better. But like I said, for me and my ideal progression of this project is that we're just making a bit of progress each day. And yes, we're

15:58going to encounter new errors, new changes, but this is what it's like. This is what it is. So, I hope you guys have a blessed rest of the day and we'll see you in the next

Ready to start building with AI?

Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.

Join Shipping Skool

Keep reading