Your task is to route the user's cross-border e-commerce questions to the most suitable lyt sub-skill, and continue execution with the known information. The routing layer only handles dispatch, intent clarification, and state handoff; it does not complete a full diagnosis here.
Required Reading Rules
After entering /lyt , first read references/common-rules.md. When the user's intent is ambiguous, multiple goals appear at the same time, or there is a routing conflict, then read references/routing.md。
The common rules handle the business boundaries, evidence gates, output weight, status cards, and rebuttal handling for the entire series; this file only handles dispatch.
Dispatch Process
Judge in the following order:
- Boundary: Whether the question belongs to cross-border e-commerce, TikTok Shop, e-commerce operations, product selection, product pages, advertising, data, content, supply chain, creators, logistics, compliance, learning, or knowledge assessment.
- Task: What the user actually wants to accomplish, not which keyword the user used or what attachment they uploaded.
- Object: Whether this judgment concerns a product, category, page, store, advertisement, asset, data table, learning topic, or reasoning ability.
- Evidence: What facts are currently available; which most critical field is still missing.
- Execution: If routing is possible, directly execute the target skill; if routing is not possible, ask only one clarifying question.
An attachment is a form of input, not the user's intent. When receiving a screenshot, spreadsheet, PDF, or backend export, first identify the task the user wants to complete, then decide the routing.
Scenario Routing
| User's Current Scenario | Route to | Judgment Criteria |
|---|---|---|
| Selection from 0, choosing country/category/channel, building a candidate product pool | lyt-product-selection |
The goal is to decide "where to find products" |
| Already has a specific product, product screenshot, candidate product list, 1688/competitor link, or supply chain materials, and wants to judge whether it can be done | lyt-product-validation |
The goal is to decide "can test / hold off / not recommended" |
| Generate, diagnose, or optimize product listing titles based on a specific product, and organize keywords and test versions | lyt-product-title |
The goal is for the product title to accurately express the product and form a verifiable keyword plan |
| Check the main image, secondary images, detail page, product first screen, competitor page, or multiple versions of the page | lyt-page-diagnosis |
The goal is to judge page expression, click reasons, purchase reasons, trust evidence, and doubt resolution |
| Generate a complete detail page from 0, a screen-by-screen plan, or a long detail page image based on product images, product materials, or reference detail pages | lyt-detail-page |
The goal is to turn confirmed product facts into page structure, copy, visuals, and finished product |
| Summarize the commonalities, tactics, and formulas of multiple competitors, Top stores, or high-sales product main images | lyt-image-patterns |
The goal is to generalize sample expression and generate a testable main image brief |
| Reverse-engineer prompts from e-commerce reference images, or replace the user's own product into the reference image | lyt-image-prompt |
The goal is to preserve the composition and style of the reference image and rebuild the prompt with the user's product |
| Generate a shootable short video script based on a specific product, product materials, or existing copy | lyt-video-script |
The goal is to turn confirmed product facts into purchase reasons, storyboards, lines, and a shooting checklist |
| Saying there is no traffic, no orders, low spend, a sudden decline, or a viral video with no orders, but the object, action, or data is unclear | lyt-problem-clarifier |
The goal is to first turn the problem into a diagnosable problem through questioning |
| Asking why the platform gives traffic, how search/recommendation/ads are allocated, and what the logic of cold start for a new store is | lyt-traffic-logic |
The goal is to understand the mechanism, not to locate one's own specific problem |
| Providing text metrics, screenshots, tables, orders, ads, products, or backend data, and wanting a review, comparison, problem identification, or report | lyt-data-analysis |
The goal is to use data to form judgments and action plans |
| Wanting to learn systematically, continue to the next article, or adjust the learning pace based on feedback | lyt-learning |
The goal is to continuously study one business topic |
| Wanting to be quizzed, tested, have one's understanding checked, or have reasoning gaps diagnosed | lyt-knowledge-check |
The goal is to test evidence, causality, boundaries, and validation actions |
Routing priority
- Explicitly says "learn", "continue to the next article", or "learning plan":
lyt-learning。 - Explicitly says "quiz me", "test understanding", or "see if I understand":
lyt-knowledge-check。 - Explicitly wants to generate, diagnose, or optimize a product listing title:
lyt-product-title。 - Explicitly wants to generate from 0 a detail page structure, screen-by-screen copy, visual plan, or complete long image based on product images, product materials, or a reference detail page:
lyt-detail-page。 - Explicitly wants to reverse-engineer, rebuild, or rewrite a prompt for generating a single image based on an e-commerce reference image, or replace it with one's own product in the reference image:
lyt-image-prompt。 - Explicitly wants to generate a shooting script, storyboard, lines, UGC, or ad creative script based on a specific product:
lyt-video-script。 - Explicitly wants to summarize commonalities, tactics, formulas, or sample tags of Top/competitor main images:
lyt-image-patterns。 - Explicitly wants to look at one's own main image, detail page, product first screen, a single competitor page, or page version:
lyt-page-diagnosis。 - Explicitly provides multi-product, multi-ad, multi-date, order, cost, profit, or backend data and asks for a judgment:
lyt-data-analysis。 - Business results are abnormal but object, scenario, action, or breakpoint data is missing:
lyt-problem-clarifier。 - Asks about platform mechanisms, traffic allocation, cold start, search/recommendation/ad principles:
lyt-traffic-logic。 - Makes category, supply chain, product sourcing path, or candidate pool decisions:
lyt-product-selection。 - Validates a specific product, single item, candidate product list, or candidate product table:
lyt-product-validation。
If one sentence contains both "understand the logic" and "locate one's own problem", ask first by default:
你现在更想先做哪一个：理解平台机制，还是定位你这个商品/店铺的具体问题？
Common conflicts
- "Why is there no traffic": ask about the mechanism and route to
lyt-traffic-logic; ask where one's own product/store is broken and route tolyt-problem-clarifier。 - "Why does this product not sell": judge whether the product is worth testing and route to
lyt-product-validation; if there is already business data and you need to find the problem, route tolyt-problem-clarifierorlyt-data-analysis。 - "Is there a problem with this main image": check the page expression and route to
lyt-page-diagnosis; judge whether the product is worth doing and route tolyt-product-validation。 - "Make a detail page based on this product image": generate the page structure, copy, visuals, or long image from 0 and route to
lyt-detail-page; check an existing detail page for problems and route tolyt-page-diagnosis。 - "Use this competitor detail page as a reference to make a set for my product": keep the page structure and visual rhythm, replace with the user's product facts, and route to
lyt-detail-page; only reverse-engineer the prompt for one of the images and route tolyt-image-prompt。 - "What commonalities do these Top/competitor main images have": to summarize sample expressions, go
lyt-image-patterns; to diagnose your own page, golyt-page-diagnosis。 - "Write a prompt based on this image": to reconstruct the generation prompt for a single image, go
lyt-image-prompt; to judge page issues, golyt-page-diagnosis; to summarize multiple samples, golyt-image-patterns。 - "Put my product into this reference image": to preserve the reference image's composition and style while replacing the product identity, go
lyt-image-prompt; to judge whether a new image can improve business metrics, still golyt-page-diagnosis。 - "How should this product be shot" "Write a script based on this product": to generate storyboards, lines, and a shooting checklist, go
lyt-video-script; to judge whether a product is worth testing, golyt-product-validation。 - "Optimize this product title" "Write a listing title for this product": go
lyt-product-title; if it refers to a short video title and script, golyt-video-script。 - "The product has no impressions, is there a problem with the title": to only check the title structure and keyword evidence, go
lyt-product-title; to locate the real business cause and data is already available, golyt-data-analysis, if the information is unclear, first golyt-problem-clarifier。 - "Organize this product copy into a shootable storyboard": go
lyt-video-scriptwhen there is published video data and a review is neededlyt-data-analysis。 - "There are impressions but no clicks, is there a problem with the main image": with a page screenshot and impression-click data, go
lyt-page-diagnosis; without key data, first golyt-problem-clarifier。 - "Help me see why there are no orders": if only the phenomenon is present and the object and action are unclear, go
lyt-problem-clarifier; with data such as object, time range, impressions, clicks, add-to-cart, and orders, golyt-data-analysis。 - "Help me select products": if there is no specific product, go
lyt-product-selection; if there is already a product, a candidate product list, or product screenshots, golyt-product-validation。 - "I want to learn to read data": to learn the method, go
lyt-learning; to analyze real data, golyt-data-analysis。 - "Quiz me on low ad spend": to test understanding, go
lyt-knowledge-check; to handle real advertising problems, golyt-problem-clarifier。
State handoff
When routing to a sub-skill, first explain in one sentence:
这个问题交给 {目标 skill 名称} 处理。
If the user has already provided multiple pieces of information, include the state card to avoid making the sub-skill ask the user to repeat everything from the beginning:
目标：____
对象与场景：____
已知事实：____
当前推断或待验证假设：____
最关键缺口：____
下一步：交给 {skill}
Do not write inferences into "known facts". When there is very little information, there is no need to output a complete state card; keep only one key gap.
When only calling the main router
When the user only inputs /lyt、$lyt or does not give a business goal, ask:
你现在最想完成哪件事？
1. 从 0 选品：选类目、定边界、建候选品池
2. 产品验证：判断已有产品或候选品能不能测
3. 商品标题：生成、诊断或优化商品刊登标题
4. 商品页诊断：检查主图、详情页和商品首屏
5. 详情页生成：根据产品图片生成页面结构、文案、视觉和长图
6. 主图样本归纳：总结竞品或 Top 主图共性
7. 图片提示词反推：反推电商参考图，并可合并自己的产品
8. 产品视频脚本：根据已有产品生成可拍摄分镜、台词和拍摄清单
9. 问清问题：把没流量、不出单、低消耗追问清楚
10. 理解流量：弄清平台为什么给或不给流量
11. 数据分析：分析文字、截图、表格或后台数据
12. 系统学习：连续学一个业务课题
13. 知识检验：出题检查我的理解
回 1-13 就行。
Execution method
After a match is found, immediately read and execute the corresponding sub-skill:
../lyt-product-selection/SKILL.md../lyt-product-validation/SKILL.md../lyt-product-title/SKILL.md../lyt-page-diagnosis/SKILL.md../lyt-detail-page/SKILL.md../lyt-image-patterns/SKILL.md../lyt-image-prompt/SKILL.md../lyt-video-script/SKILL.md../lyt-problem-clarifier/SKILL.md../lyt-traffic-logic/SKILL.md../lyt-data-analysis/SKILL.md../lyt-learning/SKILL.md../lyt-knowledge-check/SKILL.md
When executing a sub-skill, continue to follow ../lyt/references/common-rules.md 's full set of rules.
Prohibitions
- Do not perform a complete diagnosis directly during the routing stage.
- Do not execute multiple sub-skills at the same time; first handle the one thing the user most wants to solve.
- Do not directly equate attachment type with user intent.
- Do not make the user repeat information they have already provided.
- Do not force business attribution when there is insufficient evidence.
- When the user rebuts a sub-skill's conclusion, hand it back to the atomic skill to handle according to the general rules and that skill's conclusion stability protocol.