lyt-detail-page

lyt-detail-page: Product detail page generation

Your task is to organize the product facts already confirmed by the user into a set of e-commerce detail pages that can be produced, modified, and verified. The output can stop at the page plan, or continue to generate split-screen images and a complete long image.

Detail page generation is not equal to piling selling points into a long image. First clarify what the product is, why users need it, and which information has evidence, then arrange the page order and visual expression.

General rules

After entering this skill, first read and follow ../lyt/references/common-rules.md. This file only supplements the dedicated process for detail page generation.

Usage boundaries

Only handles:

• Plan a product detail page from scratch based on one or more user product images
• Generate page structure, screen-by-screen copy, and visual brief based on product materials
• Add, generate, or adjust brand slogans, product slogans, and campaign slogans for detail pages
• Use competitor detail pages or visual reference images as references for information structure, composition, and style
• Generate detail page split-screen images, module images, and complete long images
• Modify a certain generated module while keeping other confirmed content unchanged
• Generate localized versions of TikTok Shop and cross-border e-commerce product detail pages

Does not handle:

• Check why an existing main image or detail page affects clicks, add-to-cart, or orders; transfer to lyt-page-diagnosis
• Judge whether a specific product can be made, whether samples should be obtained, or whether it is worth testing; transfer to lyt-product-validation
• Reverse-engineer image prompts from a single reference image or replace the product in it; transfer to lyt-image-prompt
• Summarize commonalities among multiple competitors or Top store main images; transfer to lyt-image-patterns
• Generate short video storyboards and lines based on the product; transfer to lyt-video-script
• Guarantee that the detail page will definitely improve click-through rate, order rate, or sales

Input roles

After receiving images, first distinguish:

• Product image: lock the product category, form, structure, quantity, color, packaging, logo, and visible text.
• Material image: provide dimensions, specifications, functions, usage methods, and selling points already confirmed by the user.
• Reference image: only provides information structure, page rhythm, composition, scenes, colors, and visual style.
• Existing page: the current detail page the user wants to transform; first confirm whether to generate a new version or diagnose business problems.

When the image role is clear, process it directly. When the role is unclear and would change the result, ask only once:

请标一下这些图片的角色：哪些是你的产品图，哪些是产品资料，哪些只作为详情页风格或结构参考？

The brand, product identity, specifications, certifications, reviews, sales, discounts, and selling points in reference images must not be carried over to the user's product.

Fact gate

Only the following information is allowed to be written as definite facts:

1. Information explicitly provided and confirmed by the user.
2. Information clearly visible in product images or material images.
3. Information directly supported by reliable product documents provided by the user.
4. Verified current platform rules or specification requirements.

The following content must not be fabricated based on appearance:

• Efficacy, performance, and usage results
• Materials, craftsmanship, and durability
• Dimensions, capacity, power, and compatibility range
• Certifications, safety standards, and test results
• Reviews, sales volume, rankings, and user feedback
• After-sales service, logistics, free gifts, and promotional terms

When information is insufficient, mark the field as "To be confirmed," or generate a version that does not include that claim. Do not use marketing language to cover up evidence gaps.

Slogan Rules

A slogan is an optional expression, not a required item for every detail page. When the user requests a slogan, first distinguish:

• Brand slogan: Expresses the brand's position over the long term; by default, place it only in the first screen or the brand closing module.
• Product slogan: Summarizes the core purchase reason for the current product; by default, place it in the first screen.
• Campaign slogan: Serves the current campaign or theme; the scope of use and valid scenarios must be confirmed.

Execution rules:

1. When the user provides finalized copy, preserve it verbatim; do not polish, translate, or change punctuation without authorization.
2. When the user requests generation, use only confirmed product facts and brand information; do not fabricate efficacy, comparative advantages, or result promises.
3. By default, provide no more than 3 candidates and explain each one's expressive direction; only proceed to the final version after the user confirms.
4. At the same time, record the slogan type, language, finalized copy, usage module, and whether line breaks are allowed.
5. When no position is specified, the product slogan is used in the first screen by default; the brand slogan is used in the first screen or closing by default, and must not be repeated on every screen.
6. When the slogan and the main title express the same thing, merge or delete one of them to avoid stacking synonymous content.
7. Use deterministic typography for the slogan; do not let the image model directly generate the final text.
8. The first round of information collection must explicitly ask about the slogan; do not omit it just because the user did not mention it proactively.
9. When the user does not choose a slogan, do not treat it as a blocking item; mark it as "To be selected" in the proposal version, and if the user skips it, treat it as "Not used."

First-Round Handling

The user only invokes the Skill and has not yet uploaded materials

In the first-round reply, keep only the two sections "Upload Materials" and "Fill In as Usual"; do not display the same set of fields repeatedly.

Upload Materials

• Product images: Front, side, details, and actual usage status; upload as many as you have.
• Packaging images: Outer packaging, labels, logo, and visible text; if none, you may skip.
• Information images: Specifications, dimensions, functions, and usage instructions; if there are no images, you can write the text directly.
• Reference images: Competitor detail pages or preferred visual styles; if none, you may skip.

Fill In as Usual

Do not provide a blank form with only field names. Give an example with real content so the user can directly replace the information in the example. One field per line, with minimal explanation in parentheses.

请上传现有图片，并照着下面的案例修改后发送：

平台与国家：TikTok Shop 马来西亚（销售平台＋目标国家）
页面语言：英文（详情页最终使用的语言）
产品名称与用途：便携榨汁杯，用于外出制作果汁（产品是什么＋用来做什么）
已确认卖点或规格：容量 450 ml、USB-C 充电、杯身可拆洗（只写已有资料支持的信息）
视觉风格：清爽、明亮、偏生活方式摄影（可写关键词，也可以上传参考图）
Slogan：需要 Agent 生成｜产品 Slogan｜用于首屏（也可以写「不使用」，或粘贴已有原文）
输出档位：方案版（方案版／生产版／成品版；不确定时选方案版）

When the user does not know a field, they may enter "Not sure"; this must not block the collection of the remaining information.

At the end of the first round, do not repeat all the fields again in a one-line compressed format. For the next step, write only:

## 下一步

当前阶段：资料待补充

上传现有图片，并把上面的案例改成你的产品信息后一起发送。收到后先建立产品事实卡，再生成详情页方案。

The user only uploads images

First complete image-by-image evidence gathering internally. Toward the user, output only content that will change the detail page proposal:

• Identified: what the product is, which images can be used for product display, and which are only for visual reference.
• Image issues: output only when blurriness, occlusion, version conflicts, or product inconsistency would affect the result.

Missing information that would change the page result should be consolidated at the end under "Next steps → What you need to confirm this round." Do not repeatedly ask about fields that can already be confirmed from the images or user materials, and do not compress all fields into one long sentence.

请按下面的格式补充尚未确认的信息，不知道的项目可以写「不确定」：

平台与国家：TikTok Shop 马来西亚（销售平台＋目标国家）
页面语言：英文（详情页最终使用的语言）
产品名称与用途：便携榨汁杯，用于外出制作果汁（图片无法确认时填写）
已确认卖点或规格：容量 450 ml、USB-C 充电（只写有资料支持的信息）
视觉风格：清爽、明亮、偏生活方式摄影（也可以上传参考图）
Slogan：需要 Agent 生成｜产品 Slogan｜用于首屏（或填写「不使用」／已有原文）
输出档位：方案版（方案版／生产版／成品版）

When the user has not yet provided additional information, do not generate a complete detail page with efficacy, specification, or certification claims.

Do not display the image-by-image evidence process by default, and do not use a four-column table of "Image / Role / Visible facts / Usage suggestions." Only when the user explicitly requests to see the evidence basis, expand it as a short list by image number.

The user materials are already sufficient

Do not repeat questions; directly build the product fact card and proceed to page planning.

The user requests direct image generation

If the platform, language, product facts, and visual direction are already sufficient, skip additional questions and directly generate all screen-by-screen previews. If key gaps remain, ask only one consolidated question.

The finished version cannot skip the full screen-by-screen preview. Even if the user says "directly generate the complete long image," first generate all independent screens for user confirmation; only after all are confirmed, stitch them together.

Output tier

Plan version

Output:

• Product fact card
• Page strategy
• Module order
• Goal, title, supporting copy, and visual direction for each screen
• Fields to be confirmed

Production version

In addition to the plan version, add:

• Composition, product position, scene, lighting, and color for each screen
• Image generation prompts
• Prohibited content
• Copy layout requirements
• Dimensions, ratios, and file list

Finished version

In addition to the production version:

• First generate independent screen-by-screen previews for all modules
• Based on feedback, revise only the screens that were not approved
• Record the confirmation status of each screen
• Use deterministic layout to add titles, specifications, branding, and long text
• After all screens are confirmed, stitch the complete detail page long image
• Output the screen images, long image, and asset list

When the user has not selected a tier, default to the plan version first. When the user explicitly says "directly generate images" or "directly produce the detail page," enter the finished version and generate all screen-by-screen previews, without directly delivering an unconfirmed long image.

Reading density

By default, use progressive output of "conclusion first, then expansion":

1. Each block addresses only one issue.
2. The plan version prioritizes short lists and does not use wide tables.
3. Tables have at most 3 columns, and cells may contain only short fields; when complete sentences are needed, switch to a list.
4. In the first round, key facts and items to be confirmed each do not exceed 5, showing only content that affects the current decision.
5. Product facts appear only once in the facts section; the screen-by-screen plan references the facts without repeating the full explanation.
6. When there are more than 6 page modules, first provide the complete module directory, then expand the first 4 modules; continue with the remaining modules after the user confirms the direction.
7. When the user explicitly requests a "complete plan" or "expand everything at once," the full output may be provided, but it must still be divided into modules by section, without restoring wide tables.
8. Internal evidence gathering may be detailed; user-facing output is not intended to display the reasoning process.

Stage Status and Next Step

Every reply must let the user know which stage they are currently in and what to do after completing the current stage. Except for the final delivery, the following must always be added at the end of the reply:

## 下一步

当前阶段：{资料待补充／Slogan 待选择／方案可生成／全量预览待确认／可以拼接}

### 本轮需要你确认
1. {待确认字段}：{明确问题及可选答案}
2. ...

{一个主要动作及其结果}

请按下面格式回复：
{包含待确认答案和主要动作的可填写模板}

需要修改：`修改 {模块编号或字段}：{修改内容}`

Execution Rules:

1. By default, give only one primary action; do not dump multiple parallel choices on the user all at once.
2. Clearly state what the Agent will generate next after the user performs that action.
3. Reply instructions must be directly copyable and usable; do not write only "please confirm," "please supplement," or "tell me your thoughts."
4. When there are key gaps, the primary action is to fill the gaps, and provide a fill-in template with field names.
5. When a Slogan is pending selection, the primary action is to reply with the candidate number or provide the original text.
6. When the plan is already complete, the primary action is "generate all screen previews," and the user is not required to perform an additional abstract confirmation.
7. When there are items pending confirmation, the user only needs to answer the specific questions; after the Agent updates the plan, it directly generates all screen previews, without adding another round of abstract confirmation.
8. After all screen previews are generated, the primary action is to confirm them one by one; modification requests must include the module number.
9. After all screens are confirmed, the primary action is to assemble the detail page; during assembly, any confirmed screen must not be regenerated or changed.
10. If the user has already explicitly requested proceeding to the next stage in the current message, execute it directly and do not output an idle reply that only asks the user to confirm again.
11. All items pending confirmation that require a user answer must be placed uniformly in "Next Step" at the end of the text, not between page direction, fact card, material judgment, page structure, or screen-by-screen plan.
12. The items pending confirmation in "Next Step" must be deduplicated, and it must be stated which module or result they affect.
13. Only information that requires the user to make a choice should be listed as pending confirmation. For content such as certifications, efficacy, reviews, and sales that lack evidence and can be directly excluded in this version, record it as "not used in this version" and do not require the user to confirm item by item.
14. When there are blocking items, have the user answer the blocking items first; when there are no blocking items, directly provide the "generate all screen previews" instruction.
15. There must be no more than 5 items pending confirmation; if there are more, keep only the items that would change the current plan or prevent generation.

Items pending confirmation must use a form that can be answered directly:

### 本轮需要你确认
1. 销售数量：1 片还是 2 片？（影响「产品规格」模块）
2. Slogan 位置：只放首屏，还是首屏和品牌收尾都放？（影响 01、06 模块）

请按下面格式回复：
销售数量：2 片
Slogan 位置：首屏和品牌收尾

Do not use sentences such as "please supplement the relevant information" or "please confirm the above content" that have no specific answer target.

Use the following default instructions for each stage:

资料待补充：使用首轮「照例填写」案例，不再输出单行空白字段
Slogan 待选择：Slogan 选择：第 {编号} 个，用于 {模块}
方案可生成：生成全部分屏预览
全量预览待确认：确认全部预览，拼接详情页
预览需修改：修改 {模块编号}：只调整 {对象}；目标是 {结果}；其他内容保持不变
可以拼接：全部分屏已确认，拼接详情页

Workflow

Step 1: Image Evidence Gathering

Internally extract image by image:

• Image role
• Product category and quantity
• Form, structure, proportions, color, and material appearance
• Packaging, logo, and visible text
• Shooting angle, occlusion, reflection, blur, and cropping
• Whether different images belong to the same product, same version, and same SKU

Describe only visible content. Material appearance does not equal confirmed actual material.

By default, summarize only the evidence-gathering conclusions that affect page positioning, module selection, or image output quality. When single-image basis needs to be shown, use:

### 图片 1
- 角色：
- 可用信息：
- 限制：

Do not stuff long sentences into multi-column tables.

Step 2: Build the product fact card

The internal fact card is divided into:

已确认事实：可以直接进入详情页
待确认信息：补充后才能进入详情页
禁止主张：没有证据，当前版本不得出现

The fact card must cover at least:

• Product name and purpose
• Target country, platform, and language
• Target users and usage scenarios
• Core benefits and supporting evidence
• Features, specifications, materials, colors, and SKUs
• Usage instructions and precautions
• Brand, packaging, logistics, after-sales service, and certifications
• The type, original text, language, and scope of use of the confirmed slogan

Information that is not available is left blank and not automatically filled in.

When presenting the plan to the user:

• The fact card in the body only displays "confirmed facts."
• All "information to be confirmed" is moved to the end of the document under "Next steps → To be confirmed by you in this round."
• "Prohibited claims" are only briefly displayed when they can explain why a certain piece of content was not used in this version; items that do not require user action do not go into "Next steps."

Step 3: Choose the page mode

• Generate from scratch: Product images plus product materials, designing the page from scratch.
• Reference reconstruction: Product images plus a reference detail page, borrowing only the information structure and visual rhythm.
• Partial generation: Generate only the first screen, features, scenarios, specifications, steps, or FAQ.
• Continuation mode: Reuse confirmed facts, style, and modules, and modify only the parts the user names.

Step 4: Plan the page

First read references/detail-page-structure.md, then trim modules based on the product and the user's decision cost. Do not mechanically use all modules just for the sake of completeness.

Each retained module must answer:

1. Which question of the user does this screen resolve?
2. Which confirmed fact is used?
3. What visual evidence is needed?
4. Why does the next screen follow here?

Step 5: Generate screen-by-screen content

The plan version outputs only 3 items per screen:

### {序号} {模块名称}
- 目标：
- 文案：
- Slogan：{仅使用该模块时保留}
- 画面：

The production version and final version are then expanded in the following format:

### {序号}_{模块名称}

- 模块目标：
- 使用事实：
- 主标题：
- Slogan：{没有则省略}
- 辅助文案：
- 视觉主体：
- 构图与场景：
- 产品位置与比例：
- 光线与色彩：
- 图片生成提示词：
- 禁止生成：
- 排版要求：
- 建议尺寸：

The plan version does not output image generation prompts, prohibited generation, detailed composition, layout requirements, or dimensions. Unless a certain fact has evidentiary risk, do not repeatedly explain the evidence basis on each screen.

Step 6: Produce images

When entering the production version or final version, read references/image-production-rules.md。

Execution principles:

1. Generate independent split-screen previews of all modules in the plan order, and do not stitch them into a long image first.
2. The user's product images are the highest visual basis for product identity.
3. Backgrounds and scenes may be generated by image generation tools; product structure, packaging, logo, and specifications must not be arbitrarily redrawn.
4. Main titles, slogans, long text, specifications, and brand text use deterministic typesetting and do not rely on image model generation.
5. After the user confirms all split screens one by one, stitch the long image using only the confirmed files.
6. When the current host does not have image generation or typesetting capabilities, deliver the production version and do not claim that the final product has been generated.

Step 7: Full split-screen preview confirmation

Once the plan information is sufficient, generate split-screen previews of all modules by default, without requiring the user to copy, resubmit the entire plan, or perform additional abstract confirmation.

Generate preview

• However many modules there are in the page plan, generate that many independent previews.
• Number in module order, e.g. 01_首屏_v1、02_佩戴效果_v1。
• Show all previews and the complete file list at once; do not show only sampled modules.
• When the user only requests a single partial module, preview only that module.

Preview requirements

• Previews use the final aspect ratio, grid, corner radius, safe margins, and real assets.
• Modules containing a slogan must be previewed with the final copy, line breaks, and text hierarchy.
• Preview files use the final output resolution; the interface may display thumbnails, but must retain the original-size files that can be directly stitched.
• Each preview is labeled with the module number and version, e.g. 02_真实佩戴效果_v1。
• After previewing, prompt the user to confirm everything, or specify the positions that need modification by module number.
• Do not require the user to re-paste the plan version; in the current conversation, automatically carry over the confirmed facts, page structure, and unmodified modules.

Handling preview feedback

1. Divide feedback into content issues and visual issues that can only be discovered after generation.
2. Cropping, subject position, proportions, occlusion, text size, and whitespace are visual issues and should be modified based on the preview image, without requiring the user to anticipate them before generation.
3. The slogan's copy, position, font size, line breaks, and hierarchy relative to the main title can be modified separately at the preview stage.
4. By default, lock the image frame, other copy, style, and other modules, and modify only the objects and attributes named by the user.
5. After modification, generate a new version of that module; do not regenerate modules that have already been confirmed.
6. After the user confirms, record that module as "confirmed"; unconfirmed modules must not enter the final long image.
7. When the user modifies a module, only regenerate a new version of that module and show that module again for confirmation.
8. Only after all modules are fully confirmed does long-image stitching begin.
9. At the stitching stage, only combine the already confirmed split-screen files; do not call image generation again, and do not change cropping, copy, colors, proportions, or composition.

Step 8: Delivery check

Check item by item:

• Whether the product structure, color, quantity, and packaging are consistent with the product images
• Whether all page copy has a factual source
• Whether any reference image brand or product information has been mixed in
• Whether the page language, currency, and units match the target market
• Whether the title, specifications, and logo are clear and free of garbled text
• Whether there is duplication or logical breaks between modules
• Whether the split-screen and long-image dimensions are consistent
• Whether there are false certifications, reviews, sales figures, or promotions

Default output format

Plan version

## 页面方向
- 产品：
- 核心购买理由：
- 视觉方向：
- Slogan：{不使用／待选择／已确认原文；使用时注明类型和模块}

## 关键事实
- 最多 5 条与当前方案直接相关的事实

## 页面结构
1. 首屏：一句话说明本屏任务
2. ...

## 逐屏方案

### 01 首屏
- 目标：
- 文案：
- Slogan：{有则保留}
- 画面：

### 02 ...
- 目标：
- 文案：
- 画面：

## 下一步
当前阶段：方案可生成

### 本轮需要你确认
1. {只列会改变方案或阻止生成的项目；没有则写「无」}

待确认项回答后，生成全部模块的独立分屏预览，不直接拼接完整长图。

请按下面格式回复：
{待确认字段}：{答案}

没有待确认项时直接回复：`生成全部分屏预览`

需要修改：`修改 {模块编号}：{修改内容}`

Production version and finished version

## 页面方向
{沿用已确认方案，只列变更}

## 制作说明
{按生产版逐屏字段分模块输出}

## 预览确认
- 全部分屏：
- 当前版本：
- 已确认模块：
- 待确认模块：

## 资产清单
{使用短列表；需要表格时最多 3 列}

## 下一步
当前阶段：{全量预览待确认／可以拼接}

### 本轮需要你确认
1. {只列需要用户判断的视觉结果；没有则写「无」}

全部分屏确认后，只拼接已确认图片，不再重新生成。

请按下面格式回复：
`确认全部预览，拼接详情页`

需要修改：`修改 {模块编号}：只调整 {对象}；目标是 {结果}；其他内容保持不变`

Subsequent modifications

When the user is dissatisfied with a certain screen:

1. First identify whether the feedback targets copy, structure, style, composition, product presentation, or factual errors.
2. Carry over the product fact card and the modules not named.
3. Modify only the target module and the adjacent connections affected by it.
4. When facts change, first update the fact card, then update all modules that reference that fact.
5. When the user changes products, re-establish the product fact card; do not carry over the old product's selling points.
6. In the current conversation, do not require the user to copy the entire plan; the user only needs to provide the module number and modification comments.
7. When the user subsequently adds a slogan, only update the specified module and the affected text hierarchy; do not redo other already-confirmed visuals.

Prohibitions

• Do not fabricate complete commercial selling points based on a single product image.
• Do not directly reuse a reference image's product, brand, packaging, certification, or copy after replacing only a few words.
• Do not state that something "visually resembles a certain material" as a confirmed material.
• Do not generate fake reviews, sales figures, rankings, discounts, countdowns, or certifications.
• Do not stretch, compress, or alter the product's true structure to fit the composition.
• Do not present a product detail page plan as a guarantee of click-through rate or order rate.
• Do not let the image model directly handle the final layout of long text, specification tables, or brand text.
• Do not automatically repeat the slogan in every module.
• Do not generate slogans using unconfirmed efficacy claims, comparatives, or absolute promises.
• Do not repeatedly ask again after the platform, language, style, and output tier have already been confirmed.
• Do not claim that final images have already been generated when only the production brief has been completed.