segmentation

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Segmentation Skill

细分Skill

Write the audience description and segment-specific prompts for AI segmentation. These prompts feed any segmentation step that routes prospects into mutually exclusive audience segments.

编写用于AI细分的受众描述和细分专属提示词。这些提示词会供所有将潜在客户划分到互斥受众群的细分步骤使用。

Output Contract

输出规范

segmentation.md

must follow a fixed structure so downstream campaign tooling can parse it reliably. Hard rules this skill must honor:

```
## Default Segment
```
H2 (or legacy
```
## General Audience
```
) - its body is the general audience description for the Default / catch-all segment.
```
## Segments
```
H2 wraps all audience-segment H3s.
Each audience segment is
```
### Segment N: {Name}
```
(or
```
### {Name}
```
). Authors may write
```
Segment N:
```
prefix for organization; tooling strips it before slugifying.
The canonical name slug (
```
re.sub(r"[^a-z0-9]+", "-", name.lower()).strip("-")
```
) MUST match the segment folder's slug. Mismatches warn and leave the segment with an empty description.
Always create the
```
0 - default/
```
folder alongside the numbered segments. Audience segments are
```
1 - {slug}/
```
,
```
2 - {slug}/
```
, etc.
Optional
```
## Segmentation Mode
```
H2 selects
```
ai
```
(default) or
```
percentage
```
. In
```
percentage
```
mode each
```
### Segment N
```
carries a
```
- **Weight**: N
```
line (int 1-100) instead of criteria, and so does
```
## Default Segment
```
(the Default is one of the test arms - use it as the baseline/control); weights of ALL arms including the Default must sum to exactly 100, and tooling records a percentage weight per arm. See the Segmentation Modes section below.

segmentation.md

必须遵循固定结构，以便下游营销活动工具能够可靠解析。本Skill必须遵守以下硬性规则：

```
## Default Segment
```
二级标题（或旧版的
```
## General Audience
```
）——其正文为默认/兜底细分群的通用受众描述。
```
## Segments
```
二级标题包裹所有受众细分的三级标题。
每个受众细分群的标题格式为
```
### Segment N: {Name}
```
（或
```
### {Name}
```
）。创作者可添加
```
Segment N:
```
前缀用于整理；工具在生成slug前会移除该前缀。
标准名称slug（格式为
```
re.sub(r"[^a-z0-9]+", "-", name.lower()).strip("-")
```
）必须与细分群文件夹的slug一致。若不匹配会触发警告，且该细分群的描述将为空。
始终在带编号的细分群旁创建
```
0 - default/
```
文件夹。受众细分群文件夹格式为
```
1 - {slug}/
```
、
```
2 - {slug}/
```
等。
可选的
```
## Segmentation Mode
```
二级标题用于选择
```
ai
```
（默认）或
```
percentage
```
模式。在
```
percentage
```
模式下，每个
```
### Segment N
```
下会有一行
```
- **Weight**: N
```
（1-100的整数）而非细分标准，
```
## Default Segment
```
同理（默认群是测试组之一——用作基线/对照组）；包括默认群在内的所有测试组的权重之和必须恰好为100，工具会记录每个组的百分比权重。详见下文的细分模式部分。

Required Context

所需上下文

Read
plan.md
from the plan folder - Segment definitions and angles
Read
qualification.md
from the plan folder (if exists) - Audience description to reuse, criteria to NOT repeat
Read
search-strategy.md
from the plan folder - Search filters already applied. Never restate any of these dimensions in segment criteria.
Read
{workspace}/research.md
- Client overview, ICP details (
```
{workspace}
```
= the client campaign workspace, default
```
./cold-email/{slug}/
```
under the current directory)

读取计划文件夹中的
plan.md
—— 细分群定义与切入角度
读取计划文件夹中的
qualification.md
（若存在） —— 可复用的受众描述，以及不可重复使用的标准
读取计划文件夹中的
search-strategy.md
—— 已应用的搜索筛选条件。绝对不要在细分标准中重复提及这些维度。
读取
{workspace}/research.md
—— 客户概况、ICP详情（
```
{workspace}
```
为客户营销活动工作区，默认为当前目录下的
```
./cold-email/{slug}/
```
）

Segmentation Modes

细分模式

A campaign is in exactly one mode (set by

## Segmentation Mode

, default

ai

ai
(sub-ICP segmentation) - the normal mode. Contacts are assigned by AI to the segment whose criteria they match. Each
```
### Segment N
```
H3 holds detectable criteria. This is what the rest of this skill describes.
percentage
(sequence A/B test) - contacts are distributed randomly across the arms by weight, and each arm runs a totally different email sequence (authored in its own folder). Used to test entire sequences against each other. There are no criteria - each
```
### Segment N
```
carries only a
```
- **Weight**: N
```
line, and so does
```
## Default Segment
```
- the Default is one of the test arms, used as the baseline/control variant whenever the test has one. The weights of ALL arms - numbered plus the Default - must sum to exactly 100.

The two modes are mutually exclusive - never mix percentage arms with AI-criteria segments in one campaign. When dispatched by

campaign-planning

, follow the mode declared in

plan.md

一个营销活动只能使用一种模式（通过

## Segmentation Mode

设置，默认为

ai

）：

ai
（ICP下细分） —— 常规模式。AI会根据联系人匹配的标准将其分配到对应细分群。每个
```
### Segment N
```
三级标题下包含可识别的细分标准。本Skill的其余内容均围绕该模式展开。
percentage
（邮件序列A/B测试） —— 联系人按权重随机分配到各个测试组，每个组运行完全不同的邮件序列（在各自的文件夹中编写）。用于对比测试完整的邮件序列。该模式下没有细分标准——每个
```
### Segment N
```
下仅包含一行
```
- **Weight**: N
```
，
```
## Default Segment
```
同理——默认群是测试组之一，当测试需要对照组时，将其用作基线/对照版本。所有测试组（带编号的组+默认群）的权重之和必须恰好为100。

两种模式互斥——切勿在同一个营销活动中混合使用百分比权重测试组和AI标准细分群。当由

campaign-planning

调度时，遵循

plan.md

中声明的模式。

Core Rules

核心规则

Write short, trust the AI - The segmentation AI is intelligent. Describe each segment in a sentence or two of plain language and let it route. Don't hand it exhaustive title lists or keyword tables - a clear picture of the persona is enough.
Never repeat the search filters - the search strategy / filters already bound the list (headcount, geography, titles, seniority, industries). Don't restate any of it in the audience description or segment criteria. Segments split the audience by something the filters didn't already pin down - usually persona, role focus, or company type - never by re-stating a filter dimension.
Do NOT repeat qualification criteria - If qualification.md exists, reuse only its audience description. Don't duplicate any of its dis/qualification criteria.
Segments must be mutually exclusive - (AI mode) Each contact fits exactly one segment. Keep the line that distinguishes them crisp.
Cover the full audience - (AI mode) Every qualified contact should land somewhere; the Default segment catches whatever doesn't match a numbered one.
Default segment = the general audience description - The
```
## Default Segment
```
body is NOT catch-all criteria. Write it as the short, general audience description (persona + org type only, no filters). The segmentation AI surfaces it as the campaign's overall audience description, so keep it broad and brief - no headcount, geography, value-prop, or "doesn't fit the above" language.
Skill scope is description + criteria only - Write the audience description and segment-specific criteria. A surrounding system prompt (if any) is tool-side; do not invent a full prompt template here.
Percentage mode carries no criteria - In
```
percentage
```
mode, skip all criteria. Write only the audience description (still used by qualification) and a
```
- **Weight**: N
```
per arm. Assignment is random, so mutual-exclusivity/coverage rules don't apply.
Reasoning should be citable - (AI mode) Write criteria around concrete, observable signals the model can actually cite (role focus, company type, stage) rather than vague vibes - that makes routing reasons legible and auditable when a tool stores them.

内容简洁，信任AI —— 细分AI具备智能性。用一两句平实的语言描述每个细分群，让AI完成分配即可。无需提供详尽的职位列表或关键词表——清晰的用户画像就足够了。
切勿重复搜索筛选条件 —— 搜索策略/筛选条件已经限定了联系人范围（员工规模、地域、职位、职级、行业）。不要在受众描述或细分标准中重复提及这些内容。细分应基于筛选条件未覆盖的维度划分受众——通常是用户画像、角色侧重点或公司类型——绝对不要重复使用筛选维度。
不要重复资质判定标准 —— 若存在qualification.md，仅复用其受众描述即可。不要复制其任何资质合格/不合格标准。
细分群必须互斥 ——（AI模式下）每个联系人仅属于一个细分群。要明确划分不同细分群的界限。
覆盖全部受众 ——（AI模式下）每个符合资质的联系人都应被分配到某个细分群；默认群会收纳所有不匹配带编号细分群的联系人。
默认群=通用受众描述 ——
```
## Default Segment
```
的正文不是兜底标准。应将其编写为简短的通用受众描述（仅包含用户画像+组织类型，不含筛选条件）。细分AI会将其作为营销活动的整体受众描述展示，因此要保持宽泛简洁——不要提及员工规模、地域、价值主张或“不符合以上条件”之类的表述。
本Skill仅负责描述与标准 —— 只需编写受众描述和细分专属标准。（若有）系统提示词由工具端提供；不要在此处创建完整的提示词模板。
百分比模式无细分标准 —— 在
```
percentage
```
模式下，跳过所有标准编写。只需撰写受众描述（仍用于资质判定）和每个测试组的
```
- **Weight**: N
```
。分配是随机的，因此不适用互斥/覆盖规则。
判定依据需可溯源 ——（AI模式下）应围绕模型可实际引用的具体、可观测的信号（如角色侧重点、公司类型、发展阶段）编写标准，而非模糊的感觉——这样工具存储分配结果时，分配原因会清晰可查、可审计。

Workflow

工作流程

Step 1: Load Context

步骤1：加载上下文

Read plan.md for segment definitions
Read qualification.md (if exists) for audience description and criteria to avoid repeating
Read client context and research

读取plan.md获取细分群定义
读取qualification.md（若存在）获取受众描述，以及需避免重复的标准
读取客户背景与调研资料

Step 2: Write Segmentation Prompts

步骤2：编写细分提示词

Audience Description:

Reuse from qualification.md if it exists; otherwise one short sentence.
No filter dimensions (headcount, geography, titles, industries).

Segment Criteria:

One short section per segment from plan.md - a sentence or two on who belongs, the persona not a keyword list.
Detectable from a LinkedIn profile + company data. Make the distinction crisp, then trust the AI to handle the edges.

受众描述：

若存在qualification.md则复用其中的内容；否则写一句简短的描述。
不包含筛选维度（员工规模、地域、职位、行业）。

细分标准：

plan.md中的每个细分群对应一个简短的部分——用一两句话说明该群的受众特征，是用户画像而非关键词列表。
标准需可通过LinkedIn个人资料+公司数据识别。明确划分界限，其余边界情况交由AI处理即可。

Step 3: Write segmentation.md

步骤3：编写segmentation.md

AI mode (default):

markdown

undefined

AI模式（默认）：

markdown

undefined

Segmentation Prompt

Audience Description

[Reuse from qualification.md or write new - short description of target audience]

Segments

Segment 1: [Name from plan.md]

[Criteria for assigning contacts to this segment - what makes them belong here]

Segment 2: [Name from plan.md]

[Criteria for this segment]

Segment 3: [Name from plan.md]

[Criteria for this segment]

Default Segment

[The general audience description: a short, plain statement of who the whole audience is - persona + org type only, e.g. "Financial advisors and wealth managers at independent RIAs and advisory practices". Very short and general: NO criteria, filters, headcount, geography, or "doesn't fit the above" language. The segmentation AI surfaces this as the campaign's general audience description. Maps to the local

0 - default/

folder.]


**Percentage mode (sequence A/B test):**

```markdown

0 - default/

folder.]


**百分比模式（邮件序列A/B测试）：**

```markdown

Segmentation Prompt

Segmentation Mode

percentage

Audience Description

[Reuse from qualification.md - applies via qualification; same as AI mode]

Segments

Segment 1: [Arm name from plan.md]

Weight: 33

Weight: 33

Segment 2: [Arm name from plan.md]

Weight: 33

Weight: 33

Default Segment

Weight: 34

[The baseline / control arm. Put the control sequence here - the one the other arms are measured against. The Default is a real A/B arm and gets its weighted share of traffic.]


The Default is **one of your A/B arms** in percentage mode - use it as the baseline / control variant (the one you measure the other arms against). It is always active, so it can't be a weightless catch-all; rather than waste it, give it the control sequence. Like any arm it carries its own `- **Weight**:`, and the weights of ALL arms - numbered **plus** the Default - must be integers summing to exactly 100. (Omit the Default's weight line and tooling can back-fill the remainder `100 - sum(arms)` instead, so you can also list arms summing to ≤99 and let the Default absorb the rest.) Each arm name still has to slugify to its `N - {slug}/` folder. No criteria - the arm's full email sequence lives in its own segment folder.

Weight: 34

[The baseline / control arm. Put the control sequence here - the one the other arms are measured against. The Default is a real A/B arm and gets its weighted share of traffic.]


在百分比模式下，默认群是**A/B测试组的一员**——将其用作基线/对照版本（即用来衡量其他测试组表现的参照组）。它始终处于激活状态，因此不能是无权重的兜底群；与其浪费这个位置，不如将对照邮件序列放在这里。和其他测试组一样，它也有自己的`- **Weight**:`配置，所有测试组（带编号的组**加上**默认群）的权重必须是整数，且总和恰好为100。（如果省略默认群的权重行，工具会自动填充`100 - 其他组权重之和`的剩余值，因此你也可以只列出权重总和≤99的测试组，让默认群吸收剩余权重。）每个测试组的名称仍需生成对应的slug，匹配`N - {slug}/`文件夹。该模式下无细分标准——测试组的完整邮件序列存放在各自的细分群文件夹中。

Step 4: Create Segment Folders

步骤4：创建细分群文件夹

Always scaffold a

0 - default/

folder in addition to the numbered audience segments. The default folder holds the catch-all / general-audience copy for the Default segment. Its numeric prefix

sorts it first and marks it as the default segment folder.

{plan_folder}/0 - default/
{plan_folder}/1 - {segment-slug}/
{plan_folder}/2 - {segment-slug}/
{plan_folder}/3 - {segment-slug}/

Slug format for audience segments: lowercase, hyphens, no special characters. Derived from segment name. The default folder is always named

0 - default

(literal, no variation).

除了带编号的受众细分群之外，始终要创建一个

0 - default/

文件夹。默认文件夹存放默认群的兜底/通用受众文案。其数字前缀

会让它排在最前面，同时标记它为默认细分群文件夹。

{plan_folder}/0 - default/
{plan_folder}/1 - {segment-slug}/
{plan_folder}/2 - {segment-slug}/
{plan_folder}/3 - {segment-slug}/

受众细分群的slug格式：小写、连字符分隔、无特殊字符。由细分群名称派生而来。默认文件夹始终命名为

0 - default

（固定名称，无变体）。

Output Handling

输出处理

Workflow mode (plan folder provided):

工作流模式（提供计划文件夹时）：

Write
```
segmentation.md
```
to the plan folder root
Create segment subfolders
Confirm: "Segmentation saved. Created segment folders: [list]"

将
```
segmentation.md
```
写入计划文件夹根目录
创建细分群子文件夹
确认信息：“细分已保存。已创建细分群文件夹：[列表]”

Standalone mode:

独立模式：

Return the output inline to the user

直接向用户返回输出内容

Segment Mapping

细分映射关系

How

segmentation.md

maps to local folders and routing behavior (tool-neutral):

AI mode:

Audience description /
```
## Default Segment
```
body -> general audience description used for campaign-wide routing context; local folder
```
0 - default/
```
.
Each numbered
```
### Segment N
```
-> matching
```
N - {slug}/
```
folder and an AI-criteria segment. Assignment uses the written criteria.
Default remains the catch-all when a contact does not match a numbered segment; still write its body as the short general audience description, not "everyone else" criteria.

Percentage mode (sequence A/B test):

Each numbered arm and the Default are percentage-weight arms. No AI criteria prompts - contacts distribute by weight only.
Default is a real arm (baseline/control when you have one), not a weightless catch-all. All arm weights including Default must sum to exactly 100.
Audience description still applies via qualification.
Mode is fixed for the campaign plan - do not mix AI-criteria segments with percentage arms in one plan.

segmentation.md

与本地文件夹及路由行为的映射关系（与工具无关）：

AI模式：

受众描述 /
```
## Default Segment
```
正文 -> 用于营销活动全量路由上下文的通用受众描述；对应本地文件夹
```
0 - default/
```
。
每个带编号的
```
### Segment N
```
-> 对应匹配的
```
N - {slug}/
```
文件夹，以及一个AI标准细分群。分配依据编写的标准执行。
当联系人不匹配任何带编号的细分群时，默认群作为兜底；仍需将其正文编写为简短的通用受众描述，而非“其他所有人”之类的标准。

百分比模式（邮件序列A/B测试）：

每个带编号的测试组和默认群都是百分比权重测试组。无AI标准提示词——联系人仅按权重分配。
默认群是正式的测试组（若需要对照则作为基线/对照组），而非无权重的兜底群。包括默认群在内的所有测试组权重之和必须恰好为100。
受众描述仍通过资质判定环节生效。
营销活动计划的模式是固定的——切勿在同一个计划中混合使用AI标准细分群和百分比权重测试组。