Your task is to organize confirmed product facts and sourced keywords into a product title that can be listed, understood, and tested. A title is not a container for stacking keywords, nor can it replace the product, price, main image, and fulfillment themselves.
General rules
After entering this skill, first read and comply with ../lyt/references/common-rules.md. When title tests need to be designed, then read ../lyt/references/experiment-rules.md。
The keyword evidence rules for this skill are in references/title-evidence.md; the testing rules are in references/title-testing.md。
Usage boundaries
Only handles:
- Generating a product listing title from scratch based on a specific product and confirmed information
- Diagnosing missing information, ambiguity, repetition, keyword stuffing, and factual risks in existing product titles
- Optimizing titles based on real search terms, advertising keywords, competitor titles, or backend data
- Generating a small number of clearly differentiated test versions for the same product
- Organizing titles for multiple SKUs in bulk while preserving specification and variant differences
Does not handle:
- Short video titles, cover titles, content topic selection, or social media titles
- Product main image, secondary image, and detail page diagnosis, transfer to
lyt-page-diagnosis - Determining what to sell from scratch, transfer to
lyt-product-selection - Determining whether a specific product is worth testing, transfer to
lyt-product-validation - Full business attribution with existing impressions, clicks, add-to-cart, and order data, transfer to
lyt-data-analysis - Fabricating search volume, trending search trends, platform internal weights, or competitor conversion data
Three working modes
| Mode | User goal | Main deliverables |
|---|---|---|
| New title | Has a specific product but no usable title | Factual baseline title or data-driven title |
| Title diagnosis | Has an existing title and wants to know the problems and how to fix them | Original title diagnosis + revised version |
| Bulk optimization | Has multiple products or SKUs | Unified fields + preserved differences + bulk title table |
First-round input
First reuse the information the user has already provided, and only fill gaps that affect title facts, language, and compliance. When information is insufficient, provide the following template that can be filled in directly; do not cram all fields into one sentence:
平台和国家：TikTok Shop 马来西亚
标题语言：英文
产品名称与用途：手持衣物去毛器，用于去除衣服表面毛球
已确认属性或规格：USB-C 充电、3 档、可拆卸集毛盒
品牌：无品牌
现有标题：有就粘贴，没有可留空
真实关键词：后台搜索词、广告词或平台联想词；没有可留空
竞品标题：有就粘贴，没有可留空
禁止使用的词：没有可留空
The minimum executable input is: platform and country, title language, specific product name and use, and at least one verifiable product fact. When these are missing, ask follow-up questions first and do not generate a seemingly complete title.
Conclusion strength
Factual baseline title
When the user only has product facts and no real keyword data:
- Can generate a clearly structured title that can be listed.
- Keywords only use generic product names, user-provided terms, and descriptions directly supported by facts.
- Clearly state "This is a factual baseline title and does not mean search optimization has been completed."
- Agent-inferred synonyms must be marked as "keywords pending verification" and must not be silently written into the final title.
Data-driven titles
When the user provides backend search terms, ad keywords, platform suggested terms, or verifiable keyword data:
- Filter terms by source, relevance, purchase intent, and factual match.
- State where each core term comes from.
- You may generate keyword priority and differentiated test versions.
- Do not write search volume, click-through rate, or conversion rate correlations as title causality.
Competitor titles can only provide expression clues. A competitor's use of a term does not mean the term is effective, nor that the user's product can use it.
Execution process
1. Confirm the title object
First confirm that this is a product listing title, not a short video title, main image copy, or detail page module title. Also confirm the platform, country, and language.
2. Build a product fact card
Only record content that can be confirmed from user materials, visible information in images, or reliable sources:
产品是什么：____
主要用途：____
已确认属性：____
规格或数量：____
适用对象或场景：____
品牌与型号：____
禁止主张：____
Visible in images does not equal proven functionality. Do not infer material, efficacy, certification, target audience, or performance based on appearance.
3. Check platform restrictions
Title length, prohibited and restricted terms, brand rules, and category requirements may change. Prioritize the user's current backend prompts or the platform's latest official rules; when unable to verify, mark as "pending confirmation" and do not treat a remembered fixed character count as the current rule.
4. Diagnose existing titles
When there is an original title, check it from the following positions:
- Whether the core product term is missing, too late, or ambiguous
- Whether attributes, specifications, quantity, and applicable scenarios are consistent with the facts
- Whether synonyms are repeated, terms are mechanically stacked, or languages are mixed
- Whether it contains unprovable efficacy, certifications, absolutes, or brand terms
- Whether variant titles make different SKUs indistinguishable
Diagnosis only points out specific issues that affect understanding, search matching, or compliance, and does not use abstract scoring.
5. Build a keyword pool
Record according to references/title-evidence.md : keyword, type, source, matching evidence with the product, risk, and whether it can enter the title. Prioritize keeping the core product term, followed by highly relevant attributes, specifications, scenarios, and brand information.
6. Generate candidate titles
By default, output at most 3:
- Safe version: States most clearly what the product is, with the lowest factual risk.
- Keyword version: Provided only when there is real keyword evidence, highlighting the search expression with the highest relevance.
- Differentiated version: Highlights one provable attribute, specification, or use scenario, used to test different purchase intents.
When there is no real keyword evidence, output only the safe version and necessary structural differentiation versions; do not create near-synonym rewrites just to make up 3.
Titles do not have a fixed formula that is universal across all platforms. By default, organize according to "core product term → key attribute or use → specification/quantity → scenario/audience → brand/model", then adjust according to platform language conventions and the user's backend rules.
7. Perform title quality check
Check item by item:
- Whether each term can be supported by product facts or keyword evidence
- Whether users can quickly determine what the product is, what it does, and what they are buying
- Whether there is repetition, stacking, truncation, mixed languages, or contradictory specifications
- Whether it contains infringement, prohibited or restricted terms, absolute efficacy claims, or unverified claims
- Whether it complies with the length and format requirements already confirmed for the current platform
- Whether multiple candidates truly change only one main direction of expression
8. Design tests
When the user wants to verify effectiveness, read references/title-testing.md and ../lyt/references/experiment-rules.md. Prioritize single-variable changes, and record the version, time, traffic source, and metric definition.
Without a before-and-after comparison, exposure, clicks, or order changes cannot be attributed to the title; title effectiveness also works together with the main image, price, reviews, category, inventory, and traffic audience.
Output format
Lightweight version
## 推荐标题
{标题}
标题类型：事实基线 / 数据驱动
## 为什么这样写
- 核心产品词：{词与来源}
- 保留信息：{属性、规格或场景}
- 删除或未采用：{词与原因}
## 下一步
{需要补的关键词数据，或具体测试动作}
Standard version
# 商品标题优化：{产品}
## 当前判断
{这是事实基线标题，还是数据驱动标题；当前证据能支持到哪里}
## 原标题诊断
| 位置 | 具体问题 | 依据 | 处理 |
| --- | --- | --- | --- |
## 关键词证据
| 关键词 | 类型 | 来源 | 事实匹配 | 处理 |
| --- | --- | --- | --- | --- |
## 候选标题
### 1. 稳妥版
{标题}
### 2. 关键词版或差异版
{标题}
## 风险与待确认
{平台规则、品牌、功效、规格或翻译风险}
## 下一步
{确认标题 / 补充真实关键词 / 开始单变量测试}
Batch tasks use table output: product or SKU, original title, recommended title, main changes, keyword basis, items to be confirmed. Do not repeat lengthy explanations for each SKU.
Next-step interaction
After outputting the title, the end of the text must tell the user what to do next, and provide a short format that can be replied to directly:
下一步：
1. 采用：回复「采用 1」
2. 修改：回复「修改 1：把 ×× 改成 ××」
3. 补关键词：粘贴后台搜索词或广告关键词
4. 做测试：回复「测试 1 和 2」
Show only the options actually available in this round. If there are still facts to be confirmed, put them together at the end under "Next steps"; do not bury them in the middle.
Handoff rules
| User's next step | Handoff |
|---|---|
| Determine whether this product is worth testing | lyt-product-validation |
| Diagnose the main image, detail page, or product first screen | lyt-page-diagnosis |
| Generate a detail page from 0 based on the product | lyt-detail-page |
| Have complete operational data and want to determine the effect after changing the title | lyt-data-analysis |
| Want to generate short-video titles, storyboards, or lines | lyt-video-script |
During handoff, retain the product fact card, current title, keyword sources, confirmed platform rules, and hypotheses to be verified.
Prohibited items
- Do not claim that a certain title "will definitely go viral," "will definitely increase traffic," or "is what the platform likes most."
- Do not fabricate search volume, popularity, competition, click-through rate, or conversion rate.
- Do not directly treat words in competitor titles as valid keywords.
- Do not write into the title any materials, effects, certifications, specifications, audiences, or brands that the user has not confirmed.
- Do not use the number of keywords as a substitute for title quality, and do not mechanically fill up the character limit.
- Do not output a large number of synonymous titles at once and increase the burden of choice.
- Do not expand a product title task into a short-video title or content title task.