enrichment-design

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Data Points Builder

Bridge the gap between research hypotheses and table enrichment. Define WHAT to research about each company before running enrichment.

搭建研究假设与表格补全之间的桥梁。在执行补全前，定义需要针对每家公司调研的内容。

When to Use

使用场景

After
```
market-research
```
has produced a hypothesis set
Before
```
list-enrichment
```
— this skill designs the columns, that skill runs them
When the user says "what should we research about these companies?"

在
```
market-research
```
生成假设集之后
在执行
```
list-enrichment
```
之前——本skill负责设计列，后者负责运行补全
当用户询问“我们应该针对这些公司调研哪些内容？”时

Two Modes

两种模式

Mode 1: Segmentation

模式1：细分模式

Goal: Design columns that score or confirm hypothesis fit per company.

Input: Hypothesis set (from

market-research

or context file)

Process:

Read the hypothesis set
For each hypothesis, propose 1-2 columns that would confirm or deny fit
Discuss with user — refine, add, remove
Output final
```
column_configs
```

Example: If hypothesis is "Database blind spot — 80-90% of targets invisible to standard tools":

Column: "Data Infrastructure Maturity" (select: ["No CRM", "Basic CRM", "Full stack"])
Column: "Digital Footprint Score" (grade: 1-5)

目标： 设计用于为每家公司打分或验证假设匹配度的列。

输入： 假设集（来自

market-research

或上下文文件）

流程：

读取假设集
针对每个假设，提出1-2个可验证假设是否成立的列
与用户讨论——优化、添加或移除列
输出最终的
```
column_configs
```

示例： 若假设为“数据库盲区——80-90%的目标对象无法被标准工具识别”：

列："Data Infrastructure Maturity"（选项：["无CRM", "基础CRM", "全栈"]）
列："Digital Footprint Score"（评分：1-5）

Mode 2: Personalization

模式2：个性化模式

Goal: Design columns that capture company-specific hooks for email personalization.

Input: Target list + what the user wants to personalize on

Process:

Ask what hooks matter for this campaign (leadership quotes, recent launches, hiring signals, tech stack, etc.)
Propose 2-4 columns with prompts
Discuss with user — refine
Output final
```
column_configs
```

Example: For personalization hooks:

Column: "Recent Product Launch" (text: describe any product launched in last 6 months)
Column: "Leadership Public Statement" (text: find a public quote from CEO/CTO about [topic])

目标： 设计用于捕获公司专属切入点的列，以支持邮件个性化。

输入： 目标列表 + 用户希望用于个性化的维度

流程：

询问用户本次营销活动关注哪些切入点（如领导层言论、近期产品发布、招聘信号、技术栈等）
提出2-4个带提示词的列
与用户讨论——优化列设计
输出最终的
```
column_configs
```

示例： 针对个性化切入点：

列："Recent Product Launch"（文本：描述过去6个月内发布的任何产品）
列："Leadership Public Statement"（文本：查找CEO/CTO关于[主题]的公开言论）

Interactive Column Design

交互式列设计

Do NOT just generate columns silently. Walk through this with the user:

Step 1: Present the framework

Show the user the two modes and ask which applies (or both).

Step 2: Propose initial columns

Based on hypotheses or user input, propose 3-5 columns. For each, show:

Column: [name]
Type: [output_format]
Agent: [research_pro | llm]
Prompt: [the actual prompt text]
Why: [what this tells us for segmentation/personalization]

Step 3: Refine together

Ask:

"Any columns to add?"
"Any to remove or merge?"
"Should any prompts be more specific?"

Step 4: Confirm column budget

Guidance:

3-5 columns is the sweet spot
6-7 is acceptable if each serves a clear purpose
8+ adds noise — push back and suggest merging

Step 5: Output column_configs

Generate the final column configs as a JSON array ready for

list-enrichment

json

[
  {
    "kind": "agent",
    "name": "Column Display Name",
    "key": "column_key_snake_case",
    "value": {
      "agent_type": "research_pro",
      "prompt": "Research prompt using {input} for domain...",
      "output_format": "text"
    }
  }
]

请勿直接静默生成列。需与用户逐步完成设计：

步骤1：介绍框架

向用户展示两种模式，询问适用哪种（或同时适用）。

步骤2：提出初始列方案

基于假设或用户输入，提出3-5个列。每个列需展示：

Column: [名称]
Type: [输出格式]
Agent: [research_pro | llm]
Prompt: [实际提示文本]
Why: [该列对细分/个性化的作用]

步骤3：共同优化

询问用户：

“是否需要添加列？”
“是否需要移除或合并列？”
“是否需要优化提示词使其更具体？”

步骤4：确认列数量范围

参考指南：

3-5列是最优数量
6-7列可接受，但每列需有明确用途
8列以上会增加干扰——建议合并列

步骤5：输出column_configs

生成可直接用于

list-enrichment

的最终列配置JSON数组：

json

[
  {
    "kind": "agent",
    "name": "Column Display Name",
    "key": "column_key_snake_case",
    "value": {
      "agent_type": "research_pro",
      "prompt": "Research prompt using {input} for domain...",
      "output_format": "text"
    }
  }
]

Column Design Guidelines

列设计指南

Agent Type Selection

Agent类型选择

Data point type	Agent type	Why
Factual data from the web (funding, launches, news)	`research_pro`	Needs web research
Classification from company profile	`llm`	Profile data is enough
Nuanced judgment (maturity, fit score)	`research_reasoning`	Needs chain-of-thought
People/org structure	`linkedin`	LinkedIn-specific

数据点类型	Agent类型	原因
网络中的事实数据（融资、产品发布、新闻）	`research_pro`	需要网络调研
基于公司简介的分类	`llm`	仅需简介数据即可
精细化判断（成熟度、匹配度评分）	`research_reasoning`	需要思维链推理
人员/组织架构	`linkedin`	需LinkedIn专属数据

Output Format Selection

输出格式选择

Data point type	Format	When
Free-form research	`text`	Open-ended questions
Score/rating	`grade`	1-5 scale assessments
Category	`select`	Mutually exclusive buckets
Multiple tags	`multiselect`	Non-exclusive tags
Structured data	`json`	Multiple related fields
Yes/no with evidence	`json`	`{"match": bool, "evidence": str}`

数据点类型	格式	适用场景
自由形式调研内容	`text`	开放式问题
评分/评级	`grade`	1-5分制评估
分类	`select`	互斥选项
多标签	`multiselect`	非互斥标签
结构化数据	`json`	多个相关字段
带依据的是/否判断	`json`	`{"match": bool, "evidence": str}`

Prompt Writing Tips

提示词撰写技巧

Always include
```
{input}
```
for the company domain
Be specific about output format in the prompt itself
Include fallback: "If not found, return N/A" or "If unclear, return 'Unknown'"
For
```
select
```
/
```
multiselect
```
: list the labels in the prompt too
For hypothesis scoring: reference the specific hypothesis in the prompt
Keep prompts under 200 words

始终包含
```
{input}
```
以指代公司域名
在提示词中明确指定输出格式
包含 fallback 逻辑：“若未找到，返回N/A”或“若不确定，返回'Unknown'”
对于
```
select
```
/
```
multiselect
```
：在提示词中列出选项标签
对于假设评分：在提示词中引用具体假设
提示词长度控制在200词以内

Reference Library

参考库

See references/data-point-library.md for ~20 pre-built column configs organized by use case.

查看references/data-point-library.md获取按使用场景分类的约20个预构建列配置。

Output Handoff

输出交接

After column design is complete:

Present the final
```
column_configs
```
JSON to the user
Tell the user: "These configs are ready for
```
list-enrichment
```
. Run that skill with your table ID and these columns."
If the user wants to run immediately, hand off to
```
list-enrichment
```
workflow

列设计完成后：

向用户展示最终的
```
column_configs
```
JSON
告知用户：“这些配置已准备好用于
```
list-enrichment
```
。请使用您的表格ID和这些列运行该skill。”
若用户希望立即运行，将工作流交接至
```
list-enrichment
```