A ready-made universal prompt and display template — and the story of how it was built
What This is.
A downloadable prompt and display template that analyze listings across almost any eBay category — electronics, games, clothing, collectibles. Paste them into a profile, add your API key, and you have working analysis in about ten minutes, without writing a prompt of your own first.
Table of Contents #
- What the Starter Listing Analyzer Does
- Download the Files
- Set It Up
- How the Func State Score Works
- How It Was Built — a Real Prompt-Building Session
- Key Lessons
- Building Your Own
What the Starter Listing Analyzer Does #
For every listing it analyzes, it gives you:
- Product identification — what it is, and items per purchase vs. stock quantity
- Condition assessment — from the condition fields, description, and photos
- Seller evaluation — original owner or reseller, with the signals that triggered the call
- An ai Func State grid column: functional-resale confidence, Very High / High / Medium / Low / Risk
- A formatted dashboard in the AI Sku Analysis panel
Download the Files #
What the display template will look like:
Both files are also included in the SKU Manager Knowledge Package, so if you’ve built the SKU Manager Assistant project, your AI assistant already knows them and can modify them for you.
Set It Up #
About 10 minutes. This assumes SKU Manager is installed and whitelisted — if not, do the Quick Start Guide first.
1. Build the profile #
- Open the Data tab and, in the Ai group, click AI Settings. On a first install there is no profile yet, so the 3-step wizard opens by itself and names the profile Default; otherwise create a new profile. The wizard is shown step by step in Build the Profile — the 3-Step Wizard.
- AI Configuration: Config Mode Standard, Provider OpenAI (ChatGPT), Model
gpt-6-luna, your API key, and the whole prompt file pasted into System Prompt in place of the sample text. Click Next Step →. - Fields & Columns: tick Title, Total Price, Condition, Condition Description, Quantity, Returns, Category Name, Seller Business, Seller Registration, Seller Country. Leave Send pictures and Send description ticked under Send Prompt Fields. Under User-Defined Custom Columns (Results), tick Func State — it’s auto-detected from the prompt; ignore any other detected keys. Click Next Step →.
- Display Template: paste the whole template file in place of the sample, then click Finish Setup. The profile is saved and the AI server starts; Server Status shows Running.
A new profile opens on Google (Gemini), so change Provider to OpenAI (ChatGPT) first. The app then selects gpt-6-luna, listed under Recommended. It reads listing photos. Leave Send pictures and Send description ticked under Send Prompt Fields. Finish Setup saves the profile and starts the AI server.
2. Create the filter #
- Home → Filters, then click the + button at the top of the Filters panel.
- Condition:
TitleIs not blankto analyze everything, or narrow it to specific searches. - Action: Apply AI Prompt → Profile: the one you just built.
- Set a Filter alias, click Ok, then turn on the filter’s Enabled switch in the Filters list.
AI analysis runs only on listings a filter catches. The filter’s Action must be Apply AI Prompt, with your profile selected. A new filter starts switched off, so turn on its Enabled switch.
3. Test #
- Make sure the AI server is running (Data → Ai → Start AI Server if not) and start your feed.
- Click a result. The dashboard renders in the AI Sku Analysis panel and the ai Func State column fills in.
With Analyze new listings only selected, the first batch a search pulls when you press Start stays Not analyzed, except the row you have selected, which is analyzed. Grid → Initial Results Max sets the size of that batch. Click any other row in that batch to analyze it. Rows found later are analyzed as they arrive, and each one is a paid call.
With Analyze initially loaded listings selected, starting the AI server analyzes every Not analyzed row a filter catches, including rows found while the server was stopped. With the server already running, pressing Start also analyzes the first batch a search pulls. Each one is a paid call. When you start the server, the app asks first if the estimate is over $10.
How the Func State Score Works #
| Score | When |
|---|---|
| Very High | Explicit “works”, “tested”, “fully functional” stated |
| High | Normal listing — condition implies working, nothing contradicts it |
| Medium | “Untested” stated, but nothing says broken |
| Low | Functional issues mentioned (needs repair, intermittent) |
| Risk | “For Parts”, “doesn’t work”, or critical flags |
What deliberately does NOT affect the score: returns policy (that’s buyer protection, not functionality), cosmetic wear, lack of explicit “works great” language, and the age of the items. Most honest listings never say “tested and working” — that’s normal, not suspicious. Getting that baseline right was the hardest part of building this, as the next section shows.
How It Was Built — a Real Prompt-Building Session #
This template was built by talking to Claude in the SKU Manager Assistant project — no technical specifications, just plain-English requests and iteration. The short version:
The initial ask #
“I want a universal template that works across all product categories. Identify what the product is first, then adapt — quantity detection, condition assessment, seller evaluation. It should work for electronics, video games, clothing, collectibles, anything.”
Claude asked clarifying questions (inventory table for lots? how detailed on sellers? AI columns or just the template?) and produced a working first version: six analysis sections, product-type identification first, an adaptive inventory table.
Refining the seller assessment #
“If the seller name has ‘wholesale’ or ‘liquidation’ in it, that’s a reseller. Inventory numbering like ‘SKU:’ in the description — that’s a business. Professional language like ‘consistent with wear’ instead of ‘has some scratches’ — reseller.”
Claude turned those observations into detection patterns, and the assessment now returns an original-owner likelihood with the signals that triggered it.
Tightening the layout #
“The layout is too spread out. Tighten everything up — header on one line, less padding, more information-dense.” One message; done.
The scoring bug that mattered #
A test listing — six retro games, condition Good, no returns, no “tested working” language — scored Low. That was wrong, and the reason was instructive: the AI was treating the absence of positives as a negative. No returns? Irrelevant to whether items work. No “tested” claim? Normal listings don’t say that.
“The core assumption should be: if the condition is Used/Good — not ‘For Parts’ — assume items work unless something explicitly says otherwise.”
That one correction produced the scoring table above, and the same listing now scores High.
Naming the column #
Asked for short name options for the grid header, got a list, picked Func State.
Key Lessons #
- You can just talk to it. “The layout is too spread out” and “this scored Low but should be High because…” were enough — the project’s knowledge files handle the implementation.
- Test with real listings. The retro-games listing exposed the flawed baseline; made-up examples never would have.
- Iterate, don’t rewrite. Each version built on the last — got something working, found a problem, described it, got an update.
- Be specific about what’s wrong. “This is wrong” doesn’t help. “Scored Low, should be High, because the condition is Good and nothing says it doesn’t work” — that’s actionable.
- The baseline assumption matters most. Decide what the default verdict should be when the listing says nothing either way.
Building Your Own #
The Starter Listing Analyzer is a solid universal default, but category-specific prompts do better on the categories you actually work. In the SKU Manager Assistant project, start with:
I want to create a prompt for [PRODUCT CATEGORY].
The main things I care about are: [LIST 3–5 THINGS].
I'll be sending these fields: [LIST FIELDS].
I'm leaving Send pictures and Send description ticked.
Can you help me build this?
Answer the clarifying questions, get a first version, test on real listings, and iterate — exactly like the session above.
See also: Quick Start Guide · Building Your AI Analysis · Improving with Feedback