SuperAPI AI Directory

skill-knowledge-extractor

General & other

View repository

Updated 10-16 09:44

About this skill

Knowledge Extractor

Transform tacit knowledge into reusable Skills, turning personal experience into a team asset.

Core Value: Whether the user describes it verbally, pastes a conversation record, or uploads a document, this Skill can help identify reusable patterns and generate a structured Skill draft—without relying on any scripts, Claude analyzes and completes it directly.

Applicable Scenarios

Scenario Input Example Output
Sales pitch accumulation Sales conversation records / verbally described closing methods Opening/objection handling/closing Skill
Expert experience extraction Expert's work documents, notes, verbal descriptions Checklist/decision tree Skill
Customer service template generation Customer service chat records / complaint handling cases Standard reply template Skill
Process standardization Operation step descriptions / verbally described work methods SOP Skill
Debugging/troubleshooting process Engineer's fault handling experience Troubleshooting steps Skill
Requirements review method Product manager's review experience Issue list/decision framework Skill

Workflow

Minimal mode (recommended starting point)

When the user directly describes experience or pastes records,without any scripts, Claude directly executes the following steps:

  1. Understand input: Identify the material type (verbal / conversation record / document)
  2. Extract patterns: Identify recurring steps, judgment logic, pitch structures, checkpoints
  3. Generate draft: Directly output the knowledge extraction report + Skill draft
  4. Confirm and iterate: Confirm accuracy with the user and adjust based on feedback

Scripts (scripts/extract_patterns.py) are an optional acceleration solution, suitable for batch processing large numbers of files. For single conversations or verbal scenarios, Claude can analyze directly.

Detailed Workflow

Step 1: Collect raw materials

If the user has not provided materials, guide them to provide any of the following:

  • Verbal experience (directly describe work methods, simplest)
  • Conversation records (chat screenshots, text records)
  • Work documents (notes, reports, emails)

For verbal scenarios, use guiding questions to help the user express clearly:

  • "What do you usually do first?"
  • "When encountering [common situation], how do you handle it?"
  • "Are there any particularly important points that others tend to overlook?"

Step 2: Identify reusable patterns

Identify the following types of patterns from the materials:

Pitch patterns(applicable to sales, customer service)

  • Opening patterns, problem diagnosis patterns, objection handling patterns, closing/wrap-up patterns

Checklist patterns(applicable to review, quality inspection, evaluation)

  • Must-check items, common problem points, risk warning items

Decision tree patterns(applicable to problem diagnosis, classification, troubleshooting)

  • Judgment conditions, branch paths, final conclusions

Template patterns(Applicable to documents, replies, reports)

  • Fixed structure, variable parameters, example filling

SOP mode(Applicable to process standardization)

  • Step sequence, key nodes, acceptance criteria

Step 3: Generate the Skill draft

Based on the extracted pattern, directly generate a Skill draft that conforms to the specification:

---
name: [skill-name]
description: [何时触发、适用场景的简洁描述]
---

# [Skill 名称]

[一句话说明核心价值]

## 使用场景

[描述何时使用这个 Skill]

## 核心模式

### 模式 1:[模式名称]
**触发条件**:[何时使用]
**执行步骤**:
1. [步骤1]
2. [步骤2]
**示例**:[具体示例]

### 模式 2:[模式名称]
...

Step 4: User confirmation and optimization

After showing the draft, ask:

  1. Does it accurately reflect their experience?
  2. Are there any important patterns that were missed?
  3. Do the examples need adjustment?

Iterate and optimize based on feedback.

Optional: Script batch processing

For scenarios that require analyzing a large number of files (such as 100+ conversation records), you can use a script to speed things up:

python scripts/extract_patterns.py --input <材料文件> --type <材料类型>

Material type:conversation(conversation records) / document(work documents) / description(oral descriptions)

Output format

Knowledge extraction report

知识提取报告
━━━━━━━━━━━━━━━━━━━━━━
来源:[材料来源描述]
分析数量:[分析的记录/文档数量]

发现的可复用模式:

1. **[模式名称]**(出现 X 次,占比 Y%)
   - 触发场景:[何时出现]
   - 核心要点:[关键内容]
   - 示例:[具体示例]

2. **[模式名称]**(出现 X 次,占比 Y%)
   ...

建议生成的 Skill:
- [skill-name-1]([功能描述])
- [skill-name-2]([功能描述])

知识覆盖度:
- 已提取:X 个模式
- 覆盖场景:Y%
- 建议补充:[需要更多材料的领域]

Skill draft

Generate a SKILL.md file that conforms to the specification, including:

  • Correct frontmatter (name, description)
  • Clear usage scenarios
  • Structured pattern descriptions
  • Specific examples

Special handling

Non-technical users

For non-technical users, use simpler language:

  • Avoid technical jargon
  • Use more concrete examples
  • Provide step-by-step guidance

Sensitive information

During the extraction process, note:

  • Desensitize customer information
  • Hide specific amounts
  • Anonymize names

Combining multiple people's experience

When multiple people's experience needs to be combined:

  1. Extract each person's patterns separately
  2. Identify common patterns (best practices)
  3. Mark differing patterns (personal style)
  4. Generate a comprehensive version of the Skill

Examples

Example 1: Sales script extraction

Input: 50 closing conversation records from top salesperson Zhang San

Output:

知识提取报告
━━━━━━━━━━━━━━━━━━━━━━
来源:销售部 张三 的 50 条成交对话记录

发现的可复用模式:

1. **痛点挖掘开场**(出现 45 次,占比 90%)
   - 触发场景:首次接触客户
   - 核心要点:先问痛点,再介绍产品
   - 示例:"您目前在 XX 方面遇到什么挑战?"

2. **价格异议处理**(出现 32 次,占比 64%)
   - 触发场景:客户说"太贵了"
   - 核心要点:先认同,再算账,最后对比
   - 示例:"您说得对,我们确实不是最便宜的。但您算过吗..."

3. **限时促成**(出现 28 次,占比 56%)
   - 触发场景:客户犹豫不决
   - 核心要点:限时优惠 + 案例背书
   - 示例:"这个月签约的话,可以享受..."

建议生成的 Skill:
- sales-opening-skill(开场白助手)
- objection-handler-skill(异议处理助手)
- closing-skill(促成助手)

Example 2: Contract review experience extraction

Input: Contract review notes from legal expert Li Si

Output:

知识提取报告
━━━━━━━━━━━━━━━━━━━━━━
来源:法务部 李四 的合同审查笔记(30 份)

发现的可复用模式:

1. **付款条款检查**(必查项)
   - 检查点:付款比例、付款条件、付款时间
   - 常见问题:预付比例过高、验收条件模糊
   - 建议条款:[标准条款模板]

2. **违约责任检查**(必查项)
   - 检查点:违约定义、赔偿上限、免责条款
   - 常见问题:赔偿无上限、免责范围过宽
   - 建议条款:[标准条款模板]

3. **知识产权检查**(技术合同必查)
   - 检查点:权属归属、使用范围、保密期限
   - 常见问题:权属不清、保密期过短
   - 建议条款:[标准条款模板]

建议生成的 Skill:
- contract-checklist-skill(合同检查清单)
- clause-library-skill(条款库)

Notes

  1. Prioritize the minimal mode: Oral description is the lowest-threshold input; directly help the user extract from the conversation without preparing any materials
  2. Material quality: The richer the input material, the more accurate the extracted patterns; but a single oral description is equally valuable
  3. Privacy protection: During the extraction process, customer information and sensitive data are automatically desensitized
  4. Continuous iteration: The Skill draft needs to be confirmed with the user before it is officially used
  5. Multi-person merging: When merging multiple people's experience, be careful to distinguish common patterns from personal style