script-generator

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Script Generator

脚本生成器

Follow shared public skill rules in:

```
postplus-shared
```
public skill rules

Use this skill when the user wants to create script-ready inputs for:

AI UGC talking-head videos
TikTok or Instagram slideshow or carousel creatives
hook testing packs
script variants that can later feed storyboard or render-request work

This skill sits at the

script object

layer.

It is for:

turning fuzzy user asks into structured script briefs
correcting weak brand-first or feature-first inputs
producing testable hook and script variants
separating evidence-backed recommendations from provisional drafts
preparing a clean handoff into storyboard or request design

It is not for:

creating project or campaign file scaffolds
writing storyboard grid prompts
writing provider-ready render JSON
pretending provisional scripts are benchmark-validated truth

遵循以下共享公共技能规则：

```
postplus-shared
```
公共技能规则

当用户需要为以下内容创建可直接用于脚本的输入时，使用此技能：

AI UGC访谈类视频
TikTok或Instagram幻灯片/轮播创意内容
钩子测试包
可后续用于分镜脚本或渲染请求的脚本变体

此技能处于

script object

层级。

适用于：

将模糊的用户需求转化为结构化的脚本简报
修正以品牌或功能为核心的薄弱输入
生成可测试的钩子和脚本变体
区分基于证据的建议与临时草稿
为分镜脚本或设计请求准备清晰的交付内容

不适用于：

创建项目或营销活动文件框架
编写分镜网格提示词
编写供服务商使用的渲染JSON
将临时草稿伪装成经基准验证的结论

Core Idea

核心理念

A good AI UGC script is not "natural speech plus product benefits."

It is a conversion script disguised as casual content.

The target is not:

"make this sound more human"
"make it more viral"
"make it more premium"

The target is:

create one concrete viewer question
delay ad detection long enough to earn attention
make the mechanism or proof legible
reveal the product only after the opening tension is working
keep one core message per asset
output a script object that can be tested and handed downstream

优质的AI UGC脚本并非“自然口语+产品卖点”。

它是伪装成休闲内容的转化脚本。

目标并非：

“让内容听起来更像真人表述”
“让内容更具传播性”
“让内容更显高端”

目标应为：

提出一个具体的观众问题
延迟广告感的出现以获取注意力
让产品机制或论证清晰易懂
仅在开场张力生效后再曝光产品
每个素材只保留一个核心信息
输出可测试并能向下游交付的脚本对象

Shared File System Context

共享文件系统上下文

Read

references/source-context.md

When campaign- or project-level source files exist, consume them as stable context:

```
brand.md
```
```
persona.md
```
```
product.md
```
```
hooks.md
```

These files are shared source-of-truth inputs.

This skill may read and use them. This skill does not create or manage them.

阅读

references/source-context.md

。

当存在营销活动或项目级别的源文件时，将其作为稳定上下文使用：

```
brand.md
```
```
persona.md
```
```
product.md
```
```
hooks.md
```

这些文件是共享的事实源输入。

此技能可读取并使用它们，但不负责创建或管理这些文件。

Input Priority

输入优先级

Normalize inputs in this order:

explicit current task input
campaign or project source files
benchmark or research artifacts
user freeform text

Do not ignore the current ask just because a source file exists. Do not ignore stable source files when they directly improve script quality or consistency.

按以下顺序规范输入：

明确的当前任务输入
营销活动或项目源文件
基准或研究成果
用户自由文本

不要因源文件存在而忽略当前需求；当源文件能直接提升脚本质量或一致性时，也不要忽略它们。

Minimum Script Brief

最小脚本简报

Before writing variants, normalize the ask into the smallest useful script brief.

Required slots:

```
productOrOffer
```
```
targetUser
```
```
goal
```
```
formatTarget
```
```
corePainOrDesire
```

Important support slots:

```
audienceAwareness
```
```
mechanismOrReasonWhy
```
```
proofAvailable
```
```
ctaStyle
```
```
constraints
```
```
sourceBasis
```

If required slots are missing, infer the smallest safe version first. Ask only when the missing field would materially change the script structure.

在生成变体之前，将需求规范为最小可用脚本简报。

必填字段：

```
productOrOffer
```
```
targetUser
```
```
goal
```
```
formatTarget
```
```
corePainOrDesire
```

重要支持字段：

```
audienceAwareness
```
```
mechanismOrReasonWhy
```
```
proofAvailable
```
```
ctaStyle
```
```
constraints
```
```
sourceBasis
```

若必填字段缺失，先推断最安全的版本。仅当缺失字段会实质性改变脚本结构时，再向用户询问。

Evidence Rule

证据规则

Prefer evidence-backed script generation.

Useful evidence sources:

benchmark reports
pattern tables
master tables
landing pages
approved product sheets
approved hook libraries
campaign source-of-truth files

When evidence is incomplete, do not stop by default.

Instead, separate:

```
Observed
```
```
Inference
```
```
Open question
```

Then generate a provisional draft only if the user still needs a script.

Do not present a provisional draft as a validated recommendation.

优先基于证据生成脚本。

有用的证据来源：

基准报告
模式表
主表
着陆页
已批准的产品说明书
已批准的钩子库
营销活动事实源文件

当证据不完整时，默认不要停止工作。

而是将内容分为：

```
Observed
```
（已观测内容）
```
Inference
```
（推断内容）
```
Open question
```
（待解决问题）

仅当用户仍需要脚本时，再生成临时草稿。不要将临时草稿作为已验证的建议呈现。

First Decision

首要决策

Decide whether the request is primarily:

```
ugc_video
```
```
slideshow
```
```
both
```

Do not generate one generic script and pretend it works equally well for both.

判断请求主要面向：

```
ugc_video
```
（UGC视频）
```
slideshow
```
（幻灯片）
```
both
```
（两者皆有）

不要生成一个通用脚本并声称其对两种格式同样适用。

Script Quality Rules

脚本质量规则

Read

references/script-quality-rules.md

Every output should enforce these defaults:

start with a concrete viewer question, not a brand intro
show or imply the problem or desire before the ad feeling appears
reveal the product only when it helps the hook
keep one core message per asset
make the mechanism legible, not just the outcome
write for retention first, persuasion second
hook before branding
curiosity before explanation
product after tension starts working
mechanism before broad claims
proof before CTA
simplicity before cleverness

If the user's draft violates these rules, correct the structure before polishing wording.

阅读

references/script-quality-rules.md

。

所有输出应遵循以下默认规则：

以具体的观众问题开头，而非品牌介绍
在广告感出现之前，展示或暗示问题或需求
仅当产品能助力钩子效果时再曝光产品
每个素材只保留一个核心信息
让产品机制清晰易懂，而非仅展示结果
优先考虑留存率，其次才是说服力
先钩子后品牌
先好奇心后解释
张力生效后再展示产品
先机制后宽泛主张
先论证后CTA
先简洁后巧妙

若用户的草稿违反这些规则，先修正结构再润色措辞。

Standard Script Skeleton

标准脚本框架

Each usable script variant should map to:

```
Hook
```
```
Problem recognition or desire trigger
```
```
Proof or mechanism
```
```
Product role
```
```
Outcome
```
```
CTA
```

Compression is allowed.

Do not output a final variant that completely lacks:

hook
mechanism or proof
product role

每个可用的脚本变体应对应以下模块：

```
Hook
```
（钩子）
```
Problem recognition or desire trigger
```
（问题识别或需求触发）
```
Proof or mechanism
```
（论证或机制）
```
Product role
```
（产品作用）
```
Outcome
```
（结果）
```
CTA
```
（行动号召）

允许压缩内容，但最终变体不能完全缺少以下模块：

钩子
机制或论证
产品作用

Awareness Rule

受众认知规则

Before expanding variants, estimate the audience awareness stage:

```
unaware
```
```
problem_aware
```
```
solution_aware
```
```
product_aware
```
```
most_aware
```

Then lock:

what the script may assume
what should not be explained yet
whether direct product mention can arrive early

Wrong awareness matching is a script problem, not just a copy problem.

在扩展变体之前，评估受众的认知阶段：

```
unaware
```
（未认知）
```
problem_aware
```
（问题认知）
```
solution_aware
```
（解决方案认知）
```
product_aware
```
（产品认知）
```
most_aware
```
（深度认知）

然后确定：

脚本可预设的内容
暂不解释的内容
是否可以提前直接提及产品

认知阶段匹配错误是脚本结构问题，而非仅仅是文案问题。

Angle Rule

角度规则

Do not write only one angle unless the user explicitly asks for one fixed take.

Choose one recommended primary angle, then generate adjacent testable variants.

Supported angle families:

```
pain_recognition
```
```
proof_first
```
```
contrarian
```
```
mistake_correction
```
```
process_reveal
```
```
comparison
```
```
before_after
```
```
identity_or_social_stake
```

Each variant should state what it is testing.

Do not flood the user with minor rewrites that do not change the angle or hook job.

除非用户明确要求单一固定视角，否则不要只写一个角度。

选择一个推荐的主角度，然后生成相邻的可测试变体。

支持的角度类型：

```
pain_recognition
```
（痛点识别）
```
proof_first
```
（先论证）
```
contrarian
```
（反向视角）
```
mistake_correction
```
（纠错视角）
```
process_reveal
```
（流程揭秘）
```
comparison
```
（对比视角）
```
before_after
```
（前后对比）
```
identity_or_social_stake
```
（身份或社交立场）

每个变体应说明其测试目标。不要向用户提供仅做微小改写、未改变角度或钩子作用的变体。

Hook And Reveal Rule

钩子与曝光规则

Before writing full variants, lock:

```
viewerQuestion
```
```
hookMechanism
```
```
productRevealRule
```
```
proofRole
```
```
ctaRole
```

If these are unclear, the skill should not jump straight into polished copy.

在撰写完整变体之前，确定以下内容：

```
viewerQuestion
```
（观众问题）
```
hookMechanism
```
（钩子机制）
```
productRevealRule
```
（产品曝光规则）
```
proofRole
```
（论证作用）
```
ctaRole
```
（CTA作用）

若这些内容不明确，不要直接进入文案润色环节。

Video Script Mode

视频脚本模式

Use when:

human belief transfer matters
explanation matters
product handling or mechanism can be shown

Default writing rules:

write like speech, not like a copy block
prefer short clauses
allow light self-correction

allow natural expressions such as:

```
I didn't expect this
```
```
the weird part is
```
```
what actually matters is
```

avoid generic brand language such as:
- ```
revolutionary
```
- ```
transforms your routine
```
- ```
game-changing
```
- ```
seamless
```
- stacked empty adjectives

Do not hard-code a single fixed duration template.

Instead, use pacing slots:

```
opening hook
```
```
specific pain or desire
```
```
product use or proof
```
```
result or payoff
```
```
cta close
```

Recommended pacing may be included, but the slots matter more than the seconds.

适用于：

需要传递人类信任的场景
需要解释说明的场景
可展示产品操作或机制的场景

默认写作规则：

像口语一样写作，而非文案块
优先使用短句
允许轻微的自我修正
允许自然表达，例如：
- ```
I didn't expect this
```
  （我没想到会这样）
- ```
the weird part is
```
  （奇怪的是）
- ```
what actually matters is
```
  （真正重要的是）
避免通用品牌话术，例如：
- ```
revolutionary
```
  （革命性的）
- ```
transforms your routine
```
  （改变你的日常）
- ```
game-changing
```
  （颠覆性的）
- ```
seamless
```
  （无缝的）
- 堆砌空洞形容词

不要硬编码单一固定时长模板，而是使用节奏模块：

```
opening hook
```
（开场钩子）
```
specific pain or desire
```
（具体痛点或需求）
```
product use or proof
```
（产品使用或论证）
```
result or payoff
```
（结果或收益）
```
cta close
```
（CTA收尾）

可包含推荐节奏，但模块比时长更重要。

Slideshow Mode

幻灯片模式

Use when:

the concept is visual
the proof can be broken into cards
each frame can create micro-curiosity

Default writing rules:

one slide = one cognitive job
keep each slide short and instantly scannable
first slide must create curiosity, empathy, or tension
progression must exist across the slide sequence
do not output feature bullets disguised as slides

This mode should return slide flow, not spoken script lines split by sentence breaks.

适用于：

概念可视化的场景
论证可拆分为卡片的场景
每一帧都能产生微好奇心的场景

默认写作规则：

一张幻灯片=一个认知任务
每张幻灯片内容简短且可快速扫描
第一张幻灯片必须引发好奇心、共情或张力
幻灯片序列需有递进逻辑
不要输出伪装成幻灯片的功能要点

此模式应返回幻灯片流程，而非按句子拆分的口语脚本。

Default Workflow

默认工作流程

1. Normalize the brief

1. 规范简报

Lock the smallest sufficient script brief.

If the input is broad, infer first and ask later.

确定最小可用脚本简报。若输入宽泛，先推断再询问用户。

2. Inspect source context

2. 检查源上下文

Load the smallest useful source context:

```
brand.md
```
for voice boundaries, forbidden phrases, and positioning
```
persona.md
```
for narrator credibility and product relationship
```
product.md
```
for claims, mechanism, objections, and proof
```
hooks.md
```
for tested shells, anti-repetition, and variation direction

If a file is missing, note the consequence. Do not treat missing files as a blocker by default.

加载最小可用源上下文：

```
brand.md
```
：用于获取语音边界、禁用短语和定位信息
```
persona.md
```
：用于获取叙述者可信度和产品关联度
```
product.md
```
：用于获取主张、机制、异议和论证
```
hooks.md
```
：用于获取已测试框架、避免重复和变体方向

若文件缺失，注明影响。默认不要将缺失文件视为障碍。

3. Lock strategy before variants

3. 生成变体前确定策略

Print or reason through:

text

Format target:
Audience awareness:
Primary angle:
Support angles:
Viewer question:
Hook mechanism:
Product reveal rule:
Proof role:
CTA role:
Main source basis:
Confidence level:

输出或梳理以下内容：

text

格式目标：
受众认知阶段：
主角度：
支持角度：
观众问题：
钩子机制：
产品曝光规则：
论证作用：
CTA作用：
主要事实源：
置信度：

4. Generate variants

4. 生成变体

Default output set:

8-12 hooks
3-5 video script variants when
```
ugc_video
```
is in scope
2-3 slideshow flows when
```
slideshow
```
is in scope

Each variant should include:

```
variantId
```
```
angle
```
```
whatIsBeingTested
```
```
hook
```
```
scriptFlow
```
```
whyThisShouldWork
```
```
mainRisk
```

默认输出集合：

8-12个钩子
当包含
```
ugc_video
```
时，生成3-5个视频脚本变体
当包含
```
slideshow
```
时，生成2-3个幻灯片流程变体

每个变体应包含：

```
variantId
```
（变体ID）
```
angle
```
（角度）
```
whatIsBeingTested
```
（测试目标）
```
hook
```
（钩子）
```
scriptFlow
```
（脚本流程）
```
whyThisShouldWork
```
（生效原因）
```
mainRisk
```
（主要风险）

5. Run the script QA rubric

5. 执行脚本QA评估

Every final output should include:

```
Viewer question clear?
```
```
Ad feeling delayed enough?
```
```
Product reveal timing ok?
```
```
One core message only?
```
```
Mechanism legible?
```
```
Proof concrete enough?
```
```
CTA soft enough?
```
```
Video-native or slideshow-native?
```
```
Main drift risk
```

If a variant is still weak:

fix it before returning, or
explicitly state why it remains weak

Do not pass through a weak draft unchanged just because the user wrote it.

每个最终输出应包含：

```
Viewer question clear?
```
（观众问题是否明确？）
```
Ad feeling delayed enough?
```
（广告感是否足够延迟？）
```
Product reveal timing ok?
```
（产品曝光时机是否合适？）
```
One core message only?
```
（是否仅保留一个核心信息？）
```
Mechanism legible?
```
（机制是否清晰易懂？）
```
Proof concrete enough?
```
（论证是否足够具体？）
```
CTA soft enough?
```
（CTA是否足够温和？）
```
Video-native or slideshow-native?
```
（是否适配视频或幻灯片格式？）
```
Main drift risk
```
（主要偏离风险）

若变体仍存在缺陷：

修正后再返回，或
明确说明其仍存在缺陷的原因

不要直接返回用户提供的薄弱草稿而不做修改。

6. Return the downstream handoff

6. 返回下游交付内容

Suggest one next skill only.

Typical next steps:

```
storyboard-grid-writer
```
when scene logic should become panels
```
video-request-architect
```
when the script is stable enough for request planning
```
prompt-preflight-qa
```
when a prompt draft already exists and needs diagnosis

仅推荐一个后续技能。

典型后续步骤：

```
storyboard-grid-writer
```
：当场景逻辑需转化为分镜面板时
```
video-request-architect
```
：当脚本足够稳定可用于请求规划时
```
prompt-preflight-qa
```
：当已有提示词草稿需要诊断时

Output Shapes

输出形式

Return the smallest useful combination for the task.

Common shapes:

```
script brief summary
```
```
script strategy block
```
```
source context
```
```
hook library
```
```
video script variants
```
```
slideshow flow variants
```
```
qa block
```
```
downstream handoff
```

返回任务所需的最小有用组合。

常见形式：

```
script brief summary
```
（脚本简报摘要）
```
script strategy block
```
（脚本策略块）
```
source context
```
（源上下文）
```
hook library
```
（钩子库）
```
video script variants
```
（视频脚本变体）
```
slideshow flow variants
```
（幻灯片流程变体）
```
qa block
```
（QA评估块）
```
downstream handoff
```
（下游交付内容）

Script Strategy Block

脚本策略块

Should include:

```
goal
```
```
formatTarget
```
```
audienceAwareness
```
```
primaryAngle
```
```
supportAngles
```
```
hookMechanism
```
```
viewerQuestion
```
```
productRevealRule
```
```
proofType
```
```
softCtaRule
```
```
sourceBasis
```
```
observed
```
```
inference
```
```
openQuestions
```
```
mainDriftRisks
```

应包含：

```
goal
```
（目标）
```
formatTarget
```
（格式目标）
```
audienceAwareness
```
（受众认知阶段）
```
primaryAngle
```
（主角度）
```
supportAngles
```
（支持角度）
```
hookMechanism
```
（钩子机制）
```
viewerQuestion
```
（观众问题）
```
productRevealRule
```
（产品曝光规则）
```
proofType
```
（论证类型）
```
softCtaRule
```
（温和CTA规则）
```
sourceBasis
```
（事实源）
```
observed
```
（已观测内容）
```
inference
```
（推断内容）
```
openQuestions
```
（待解决问题）
```
mainDriftRisks
```
（主要偏离风险）

Source Context

源上下文

Should include:

```
brandContext
```
```
personaContext
```
```
productContext
```
```
hookContext
```
```
confidenceLevel
```

Use values such as:

```
from brand.md
```
```
from persona.md
```
```
from product.md
```
```
from hooks.md
```
```
from benchmark patterns
```
```
from landing page
```
```
from user input
```
```
missing
```

Confidence levels:

```
grounded
```
```
partially_grounded
```
```
provisional
```

应包含：

```
brandContext
```
（品牌上下文）
```
personaContext
```
（人物角色上下文）
```
productContext
```
（产品上下文）
```
hookContext
```
（钩子上下文）
```
confidenceLevel
```
（置信度）

可使用以下取值：

```
from brand.md
```
```
from persona.md
```
```
from product.md
```
```
from hooks.md
```
```
from benchmark patterns
```
```
from landing page
```
```
from user input
```
```
missing
```

置信度取值：

```
grounded
```
（有依据）
```
partially_grounded
```
（部分有依据）
```
provisional
```
（临时）

Anti-Patterns

反模式

Do not:

open with the brand name when the viewer question is not yet earned
keep multiple unrelated promises inside one asset
hide the product role until the viewer cannot tell what is being sold
replace mechanism with vague outcome claims
give slideshow output that reads like copy bullets
give talking-head output that reads like ad brochure copy
silently treat
```
hooks.md
```
as validated performance truth when it is only a raw library

请勿：

在未建立观众问题时以品牌名称开头
在一个素材中包含多个无关承诺
隐藏产品作用直到观众无法识别所售产品
用模糊的结果主张替代机制说明
输出读起来像文案要点的幻灯片内容
输出读起来像广告手册的访谈类内容
将
```
hooks.md
```
默认为已验证的性能事实，实际上它只是原始库

Failure Mode

故障模式

Stop or warn clearly when:

the script is under-specified in a way that changes structure
the user wants too many messages in one asset
no believable mechanism or proof exists
the script still sounds like a brand talking, not creator-native delivery
the output format does not match the platform job

Then give:

the closest usable corrected version
the main unresolved risk
the next honest unblocker

当出现以下情况时，停止操作或明确警告：

脚本规格不足，会改变结构
用户希望在一个素材中包含过多信息
不存在可信的机制或论证
脚本仍听起来像品牌发言，而非创作者原生风格
输出格式与平台需求不匹配

然后提供：

最接近可用的修正版本
主要未解决风险
下一个可行的解决方案