Ecommerce Product Image Suite
Role
Act as the single ecommerce image entry point. Classify the user’s requested image type, load only the needed reference file, then use $gpt-image-2 from @zjxuyunshi/gpt-image-2 for the current image-generation or image-editing execution path.
This skill supplies ecommerce routing, product-truth rules, fact discipline, prompt frames, output naming, and marketplace constraints. $gpt-image-2 from @zjxuyunshi/gpt-image-2 supplies the executable image contract and may change over time; follow its current instructions instead of copying stale connector payload details here.
Quick Triage
- Confirm the user supplied at least one product image. If not, ask for it.
- Identify the requested output type from the user’s words, references, and platform target.
- Read
references/shared-execution.md.
- Read exactly one capability reference from the router below. Read a second one only when the request truly combines two output types, such as a main image with size annotations.
- Ask one focused question only when a missing fact would materially change the reusable output or create an unsupported claim.
Capability Router
| User intent |
Read |
| Hero image, listing cover, white-background main image, product-first commercial cover |
references/main-images.md |
| Replace, clean, or upgrade the background while keeping the product intact |
references/background-replace.md |
| Lifestyle scene, product in home/outdoor/studio/travel/office environment |
references/scene-images.md |
| Where, when, why, how, or for whom the product is used |
references/use-case-images.md |
| Text callout or benefit graphic for a selling point |
references/selling-point-images.md |
| Texture, hardware, stitching, packaging, component, craftsmanship close-up |
references/detail-images.md |
| Dimensions, size chart, specification table, capacity, weight, parameters |
references/size-images.md |
| Poster, campaign visual, launch graphic, promotion banner, live-commerce cover |
references/marketing-poster.md |
| Generate a new model wearing, holding, carrying, or using the product |
references/ai-model-images.md |
| Replace the model in a supplied model/reference image |
references/model-replace.md |
| Create new poses or angles from an existing model/product setup |
references/pose-variant.md |
| Replace a product in a supplied reference scene with the user’s product |
references/product-replace.md |
Fact Discipline
Separate facts before prompting:
confirmed: supplied by the user or printed visibly on the product/packaging.visible: safe observations from images, such as color, shape, compartments, packaging form, or included visible parts.missing: facts needed for the requested output but not supplied.do not claim: material, certifications, performance, waterproofing, load-bearing, medical, slimming, safety, child-safety, food-contact, compatibility, discounts, or test data unless confirmed.
Use placeholders only when the user explicitly asks for a draft. Otherwise, ask for missing factual values.
Result Handling
Save generated files under a clear folder in /Users/yunshi/Downloads/ecommerce-product-image-suite/<short-product-name>-<timestamp>/. Preview the first generated image when practical. Report what was created, where files were saved, and which confirmed facts or assumptions were used.
Failure Handling
Stop and report the exact blocker for missing product images, missing required factual values, auth, billing, upload failure, connector/schema rejection, task timeout, result download failure, or excessive product drift. Do not create repeated tasks blindly.
---
name: ecommerce-product-image-suite
description: 'Route ecommerce product image generation and editing tasks from product photos into the right specialized workflow. Use when the user asks for marketplace-ready main images, background replacement, lifestyle scenes, use-case images, AI model shots, model replacement, pose variants, product-in-scene replacement, selling point graphics, detail close-ups, size/specification images, or marketing posters for Amazon, Shopify, TikTok Shop, Xiaohongshu, Taobao, Tmall, 1688, Shopee, Temu, JD, or DTC stores.'
metadata:
icon: ':lucide:shopping-bag'
title: Ecommerce Product Image Suite
packageName: '@zjxuyunshi/ecommerce-product-image-suite'
version: 0.0.2
dependencies:
- packageName: '@zjxuyunshi/gpt-image-2'
skill: gpt-image-2
---
# Ecommerce Product Image Suite
## Role
Act as the single ecommerce image entry point. Classify the user's requested image type, load only the needed reference file, then use `$gpt-image-2` from `@zjxuyunshi/gpt-image-2` for the current image-generation or image-editing execution path.
This skill supplies ecommerce routing, product-truth rules, fact discipline, prompt frames, output naming, and marketplace constraints. `$gpt-image-2` from `@zjxuyunshi/gpt-image-2` supplies the executable image contract and may change over time; follow its current instructions instead of copying stale connector payload details here.
## Quick Triage
1. Confirm the user supplied at least one product image. If not, ask for it.
2. Identify the requested output type from the user's words, references, and platform target.
3. Read `references/shared-execution.md`.
4. Read exactly one capability reference from the router below. Read a second one only when the request truly combines two output types, such as a main image with size annotations.
5. Ask one focused question only when a missing fact would materially change the reusable output or create an unsupported claim.
## Capability Router
| User intent | Read |
| --- | --- |
| Hero image, listing cover, white-background main image, product-first commercial cover | `references/main-images.md` |
| Replace, clean, or upgrade the background while keeping the product intact | `references/background-replace.md` |
| Lifestyle scene, product in home/outdoor/studio/travel/office environment | `references/scene-images.md` |
| Where, when, why, how, or for whom the product is used | `references/use-case-images.md` |
| Text callout or benefit graphic for a selling point | `references/selling-point-images.md` |
| Texture, hardware, stitching, packaging, component, craftsmanship close-up | `references/detail-images.md` |
| Dimensions, size chart, specification table, capacity, weight, parameters | `references/size-images.md` |
| Poster, campaign visual, launch graphic, promotion banner, live-commerce cover | `references/marketing-poster.md` |
| Generate a new model wearing, holding, carrying, or using the product | `references/ai-model-images.md` |
| Replace the model in a supplied model/reference image | `references/model-replace.md` |
| Create new poses or angles from an existing model/product setup | `references/pose-variant.md` |
| Replace a product in a supplied reference scene with the user's product | `references/product-replace.md` |
## Fact Discipline
Separate facts before prompting:
- `confirmed`: supplied by the user or printed visibly on the product/packaging.
- `visible`: safe observations from images, such as color, shape, compartments, packaging form, or included visible parts.
- `missing`: facts needed for the requested output but not supplied.
- `do not claim`: material, certifications, performance, waterproofing, load-bearing, medical, slimming, safety, child-safety, food-contact, compatibility, discounts, or test data unless confirmed.
Use placeholders only when the user explicitly asks for a draft. Otherwise, ask for missing factual values.
## Result Handling
Save generated files under a clear folder in `/Users/yunshi/Downloads/ecommerce-product-image-suite/<short-product-name>-<timestamp>/`. Preview the first generated image when practical. Report what was created, where files were saved, and which confirmed facts or assumptions were used.
## Failure Handling
Stop and report the exact blocker for missing product images, missing required factual values, auth, billing, upload failure, connector/schema rejection, task timeout, result download failure, or excessive product drift. Do not create repeated tasks blindly.