legal-qa-extractor

Legal Q&A Extraction Skill

Overview

Extract valuablelegal Q&A pairsfrom lawyer-client communication records (client questions + lawyer answers) to generate structured knowledge base content. Suitable for transforming client consultations into reusable legal knowledge for later organization into articles, cases, or learning materials.

Core Value: Extract Q&A content that can be reused by other clients, rather than the specific case details of a single client.

Core Principles

Client Information Protection

Absolutely prohibitedto include the following in Q&A content:

❌ Client's real name
❌ Enterprise/institution name
❌ Specific address, contact information
❌ Any information that can identify an individual or enterprise

Anonymization Standards：

✅ Client name → Use aliases such as "the party," "the client," "Party A/Party B," etc.
✅ Enterprise name → Use aliases such as "a certain company," "a certain enterprise," "the enterprise," etc.
✅ Specific information → Blur into "a certain place," "a certain time," "relevant personnel," etc.
✅ Client's original words → Quote after anonymization: "the client asked..." rather than "President Zhang said..."

Value Judgment of Q&A Pairs

Extracted Q&A should meet at least one of the following conditions:

Universality: Other clients may also encounter the same type of issue
Typicality: Represents common concerns in a certain type of legal scenario
Educational Value: The lawyer's answer has legal popularization value or guiding significance
Reusability: Can stand alone as an article or serve as a knowledge base entry

Workflow

Step 1: Analyze Input Content

Obtain the communication record document and the user's specific requirements, and understand:

The communication content and context in the document
Which aspects the user wants to focus on
Which Q&A have reuse value

Step 2: Identify Valuable Q&A Pairs

Comprehensively review the Q&A content in the document, and identify value according to the following dimensions:

Q&A Identification Dimensions

Direct Q&A - The client clearly raises a question, and the lawyer provides an answer
Follow-up Response - The client asks a follow-up question, and the lawyer further explains
Risk Reminder - Risk points proactively reminded by the lawyer and response suggestions
Operational Guidance - Q&A involving processes, steps, and precautions
Common Concerns - The same type of issue that other clients may also encounter

Q&A Extraction Methods

Semantic Analysis: Understand the client's true question and the lawyer's core viewpoint
Logical Summarization: Organize scattered dialogue into independent Q&A pairs
Value Assessment: Determine whether the Q&A has universal applicability
Standardized Expression: Convert colloquial expressions into standardized question-and-answer format

Step 3: Generate structured report

Save the output file tothe same directory as the original document, file naming:{原文档名}_法律问答提取_{YYYYMMDD}.md

Output formatSee details in output-template.md。

Each question-and-answer pair includes the following parts:

Question: the question raised by the client (standardized wording)
Lawyer's answer: the lawyer's core viewpoints and suggestions (300-500 words)
Question background(optional): the context in which the question arose (200-300 words)
Value labels: universality/typicality/educational value/reusability
Applicable scenarios: under what circumstances readers would care about this question and answer

Step 4: Question-and-answer classification and association

Provide the following analysis dimensions:

Question-and-answer feature analysis: number of key question-and-answer items, number of regular question-and-answer items, value distribution
Domain classification: classify by legal domain or business scenario
Association analysis: which question-and-answer items are related and can be combined into an article

Quality standards

Independence: each question-and-answer pair can be understood independently, without relying on the full conversation context
Universality: the question-and-answer content also has reference value for other clients
Completeness: both the question and the answer are clear and complete, without being taken out of context
Strict anonymization: strictly follow client information protection rules
Reusability: the extracted question-and-answer items can be directly used in articles, knowledge bases, or learning materials

Input requirements

This skill accepts the following types of input:

Document path: the path of the communication record document provided by the user
Pasted content: communication record content pasted directly by the user
User requirements: what types of question-and-answer items the user wants to focus on

Configuration and archiving

User configuration

This skill supports customizing scanning behavior through config/.env files:

# 复制模板
cp config/.env.example config/.env

# 编辑配置
QA_EXTRACTOR_SCAN_PATHS=~/Documents/legal-consultations
QA_EXTRACTOR_SCAN_DEPTH=3
QA_EXTRACTOR_FILE_TYPES=md,txt

Scan records

The skill maintains archive/scan-log.json to record scanning status:

{
  "version": "1.1.0",
  "last_full_scan": "2026-05-10T23:00:00+08:00",
  "watch_paths": [
    {
      "path": "~/Documents/legal-consultations",
      "last_scanned": "2026-05-10T23:00:00+08:00",
      "status": "active"
    }
  ],
  "projects": {
    "潜在项目_A": {
      "source_files": ["path/to/file1.md", "path/to/file2.md"],
      "last_extracted": "2026-05-10",
      "qa_count": 8,
      "output_file": "archive/extractions/2026-05-10_潜在项目_A_qa.md",
      "status": "active"
    }
  },
  "files": {
    "path/to/file.md": {
      "content_hash": "sha256:abc123...",
      "last_modified": "2026-05-09",
      "last_extracted": "2026-05-10",
      "qa_count": 5,
      "status": "active"
    }
  },
  "archived": {
    "path/to/old_project": {
      "archived_at": "2026-05-01",
      "reason": "moved_by_user"
    }
  }
}

Field descriptions:

Field	Purpose
`files.*.content_hash`	File content fingerprint; when it changes, re-extraction is triggered
`files.*.status`	`active` / `archived`, skip scanning after archiving
`projects.*.source_files`	Multiple communication records for the same project, merged output
`projects.*.status`	`active` / `archived`, locked after the project is archived
`archived.*`	Record paths that have been moved away to prevent duplicate scanning

Incremental scanning logic:

Folder type determination (used to identify archived and active projects):

Naming pattern	Type	Processing method
Pure numbers / date range (e.g., `24`、`25 01-06`、`26 01-06`）	Archive container	Deep scan internal subfolders
Date + name (e.g., `260323 张美金...`、`260404 顾忆芬...`）	Direct project	Directly scan this folder
Other	Project	Directly scan this folder

Iterate over all direct subfolders under watch_paths
Determine the type of each folder (archive container / direct project)
Recursively scan archive containers; scan content files for direct projects
Compute for each file content_hash
If a file is not in scan-log, or its hash does not match the record → trigger extraction
Trigger extraction only if all files in the project have changed
After extraction is complete, merge multiple records by project and output a single QA file

Extraction result archiving

The extracted question-answer pairs are saved in archive/extractions/ directory, aggregated by project:

archive/
├── scan-log.json              # 扫描状态记录
└── extractions/
    ├── 2026-05-10_潜在项目_A_qa.md   # 项目A的多轮咨询合并
    ├── 2026-05-10_潜在项目_B_qa.md   # 项目B的单次记录
    └── 2026-05-11_潜在项目_C_qa.md   # 项目C（后续增量新增）

Multi-round consultation merge logic:

All .md/.txt files within the same project folder are treated as multiple rounds of communication for the same project
During extraction, integrate all question-answer pairs in file order (filename or modification time)
Deduplication: for multiple occurrences of the same question, keep the latest answer
Output filename:{日期}_{项目名}_qa.md

Note：archive/ The directory is not included in Git version control.

Applicable scenarios

Organize customer consultation records into reusable knowledge
Extract content marketing material from conversations
Build a legal question-and-answer knowledge base
Prepare material for legal popularization articles
Create a list of frequently asked customer questions
Regularly scan project folders and incrementally extract new content