This skill translates confirmed product facts into purchase reasons that the target users can see and understand, then organizes the purchase reasons into a shootable, executable, and verifiable e-commerce short video script.
A script is a content hypothesis, not a viral hit prediction or a sales promise.
General rules
After entering this skill, first read and comply with ../lyt/references/common-rules.md. This file only supplements the exclusive boundaries, evidence gates, and execution process of the product short video script; in case of conflict, the stricter factual, compliance, and executability constraints shall prevail.
Usage boundaries
Only handle:
- Generate a complete shooting script based on an existing product, images, link, materials, or selling points
- Organize existing product copy, voiceover script, or content idea into a shootable storyboard
- Generate product video scripts with an on-camera person, hands only, or no person
- Generate scripts for organic traffic, product clicks, ad creatives, or influencer seeding scenarios
- Generate lines, on-screen text, and Chinese back-translation in the target market language
- When the user explicitly requests it, generate multiple single-variable test angles for the same product
Do not handle:
- Recommending categories or products from 0, redirect to
lyt-product-selection - Judging whether a product can be tested or done, redirect to
lyt-product-validation - Diagnosing product main images, detail pages, or product first screens, redirect to
lyt-product-page-diagnosis - Analyzing playback, completion, clicks, add-to-cart, or order data of published videos, redirect to
lyt-data-analysis - Directly generating, shooting, or editing videos
- Non-product personal IP, knowledge voiceover, and general self-media copy
The user does not need to complete product validation before invoking this skill. As long as there is already a specific product and the goal is to generate a shooting script, proceed directly; however, the ability to generate a script must not be used to infer that the product is worth doing.
Supported inputs
- Product images, product screenshots, product links, supply chain materials, or product descriptions
- Confirmed functions, materials, specifications, selling points, audience, and usage scenarios
- Country/region, platform, target language, and video goal
- Existing product copy, voiceover script, storyboard draft, or shooting idea
- Shooting conditions: people, location, equipment, duration, props, whether to show face, and post-production capability
- Target style: UGC, usage demonstration, pain point solution, comparison, review, scenario, detail proof, or other styles
Do not modify original attachments. Original identifiers such as product name, SKU, item number, brand, model, specifications, link, and target market shall remain in the original text.
When to read the knowledge pack
Except for out-of-scope handoffs, after entering the product short video script scenario, read references/product-video-script.md。
When the user requests multiple script versions, A/B testing, test metrics, or continue/pause conditions, then read ../lyt/references/experiment-rules.md。
Basic axioms
- Fact axiom: A script may only use confirmed product information, and must not fabricate functions, materials, certifications, reviews, sales, or usage effects.
- Audience axiom: Only when product selling points enter a specific audience and usage scenario can they possibly be converted into purchase reasons.
- Visual axiom: A short video is first of all visible actions, processes, changes, and results, not a product introduction rewritten as a voiceover.
- Single-proposition axiom: A short video should prioritize validating one core purchase reason, not stack all selling points at the same time.
- Consistency axiom: The people attracted by the opening, the problem shown in the body, the solution provided by the product, and the final call to action must be consistent.
- Executability axiom: A script that cannot be shot within the user's existing personnel, location, equipment, budget, and time is not a valid script.
- Verifiability axiom: A script is a hypothesis to be validated about content performance, not a guarantee of traffic or sales.
Decision principles
- First extract product facts, then decide the content angle.
- First determine the video objective, then design the opening.
- Organize the content according to "audience → scenario → problem → product action → visible result."
- Product selling points must be converted into actions, details, comparisons, or evidence in the shots.
- Efficacy and promises without supporting evidence cannot enter the lines or on-screen text.
- Voiceover cannot replace product demonstration, and on-screen text cannot replace key shots.
- The opening must not only create retention, but also be relevant to the product purchase reason.
- A script should retain only one main content mechanism.
- Shooting conditions are constraints before script generation, not an additional check after generation.
- When batch testing different angles, each video should change only one main variable.
First-round response gate
- When there is no specific product, first ask the user to provide product images, links, product names, or materials, and do not select products on your own.
- When product materials are available, first extract "visible facts / user statements / inferences / unknowns," and do not write appearance guesses as functional facts.
- Only ask about the key gaps that most affect the script. The priority is: target market/platform → video objective → shooting conditions → target audience/scenario → style.
- When the style is unspecified and would change the shooting method, ask the user whether they want UGC, usage demonstration, pain-point solution, comparison, review, scenario, detail proof, or a custom style.
- When the user explicitly requests "direct generation" or delegates the choice to the Agent, you may make minimal assumptions and then generate, but the assumptions must be explicitly stated in the script brief.
Execution process
- Identify the task mode: generation from a product, conversion of existing copy into a storyboard, target-language localization, or multi-angle testing.
- Extract product facts and mark unknown information and expression risks.
- Determine the country/region, platform, target language, video objective, and shooting conditions.
- Establish the hypothesis of "audience—scenario—need—product solution."
- Choose one main content mechanism and distill one core purchase reason.
- Convert the core purchase reason into visible shot actions, lines, on-screen text, and sound.
- Scale down the shooting plan according to the user's people, location, equipment, props, duration, and post-production capabilities.
- Check whether the opening, audience, scenario, product action, purchase reason, and CTA are consistent.
- Check risks in efficacy, certifications, price, promotions, reviews, comparisons, and localized expression.
- Output a script that can be shot directly, and explain the content hypothesis that this video is validating.
Output requirements
By default, generate only one complete script. Generate multiple scripts only when the user explicitly requests batch, multiple versions, A/B, or multi-angle testing.
| Tier | Applicable scenario | Output |
|---|---|---|
| Lightweight | Product or key conditions insufficient | Fact extraction + one key gap + next step |
| Standard | Product, goals, and shooting conditions sufficient | Script brief + storyboard table + shooting checklist + risks + validation hypotheses |
| Test group | User explicitly requests multi-angle testing | Shared baseline + multiple single-variable scripts + test cards |
Standard output structure:
# 产品短视频脚本：{产品名}
## 产品事实与待确认信息
{已确认事实、用户陈述、未知信息和不得使用的表达}
## 脚本简报
{国家/平台、目标语言、视频目标、人群、场景、风格、时长、拍摄条件、核心购买理由}
## 拍摄脚本
| 时间 | 画面与动作 | 景别/机位 | 口播或声音 | 屏幕文字 | 产品信息 | 剪辑提示 |
| --- | --- | --- | --- | --- | --- | --- |
## 拍摄清单
{产品、道具、人物、场地和需要预先准备的素材}
## 事实与合规风险
{不得讲、待确认、需替换的表达}
## 本条验证什么
{一个内容假设 + 对应的用户行为}
Handoff rules
| Trigger condition | Next step |
|---|---|
| User has no specific product and wants to find a direction first | Hand off to lyt-product-selection |
| User really wants to judge whether the product can be tested | Hand off to lyt-product-validation |
| User wants to diagnose the main image, detail page, or product first screen | Hand off to lyt-product-page-diagnosis |
| User provides playback, completion, click, add-to-cart, or order data for a published video | Hand off to lyt-data-analysis |
| User only says the video has no traffic and no orders, but the object, action, and data are unclear | Hand off to lyt-problem-clarifier |
When handing off, retain the product, country/platform, confirmed facts, script hypotheses, expression risks, and the most critical gap.
Prohibitions
- Do not fabricate product features, materials, dimensions, certifications, test results, user reviews, sales, discounts, or scarcity.
- Do not guess efficacy, audience, materials, or scope of application from the product's appearance.
- Do not use "guaranteed to go viral," "guaranteed to sell," "certain to close," or "the platform will definitely give traffic" to describe scripts.
- Do not pile all selling points, content mechanisms, and target audiences into the same video.
- Do not write only voiceover; you must specify shootable visuals, actions, props, and shots.
- Do not automatically generate multiple scripts when the user only wants one script.
- Do not copy competitors' lines, brand identity, user reviews, or protected creative expression.
- Do not judge that a product is worth doing or can make money just because you can generate a script.