菜单
从访谈录音/文本中萃取结构化知识。TRIGGER when: 用户需要"萃取"、"提取知识"、"访谈分析"、 "知识萃取报告"、从DOCX/TXT提取专家经验。支持单文件和多文件综合萃取，自动检测输入类型。 输出严格按 extraction-schema.json 的 JSON 结构。 SKIP: 生成 HTML/PPTX/DOCX 交付物（那是 zhongwen-content 的职责）。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
字体风格自适应提示生成器 - 输入任意文字，自动生成专业字体设计提示词并生成图片
Automatically add "Save as PNG" screenshot functionality to HTML files. Use this skill whenever the user mentions: adding screenshot functionality to HTML, converting HTML to images, saving web pages as pictures, creating downloadable screenshots from HTML, adding export-to-image buttons, generating PNG from HTML content, or when they need to capture long scrolling HTML pages as complete images. Also trigger when users mention html2canvas, full-page screenshots, or need to share HTML content as image files.
Use this skill whenever the user needs to extract knowledge and experiences from organizations, including AI implementation, digital transformation, organizational change, business innovation, and best practices documentation. Typical triggers: - "萃取XX经验" (Extract XX experience) - "萃取知识" (Extract knowledge) - "设计访谈提纲" (Design interview outline) - "萃取报告" (Extraction report) - "AI萃取" (AI extraction) - "知识萃取" (Knowledge extraction) - "经验萃取" (Experience extraction) - "最佳实践" (Best practices) - "案例总结" (Case study summary) - Any task involving extracting, summarizing, or documenting experiences from organizations This skill automatically handles: 1. Auto-detection of input type (single file vs batch folder) 2. Zero-confirmation pipeline: text extraction → Agent萃取 → Agent渲染 → 验证 3. Pipeline status tracking with visual summary 4. Interview outline generation for 4 types of interviewees 5. Extraction report generation using extraction-schema.json + report-template.html ## Pipeline Sub-Agents (called automatically) -
Use this skill whenever the user wants to convert any file format to Markdown. Supports PDF, DOCX, PPTX, XLSX, audio, video, images, and more. This is Microsoft's markitdown tool that converts various document formats to clean Markdown text.
创建多页面导航入口，整合多个独立 HTML 文件为统一的导航中心
Use this skill whenever the user needs to extract knowledge and experiences from Ping An (平安) related organizations, including Ping An Life Insurance (平安人寿), Ping An Zhiniao (平安知鸟), Ping An Bank (平安银行), Ping An Good Doctor (平安好医生), Ping An Technology (平安科技), and other Ping An subsidiaries. Typical triggers: - "萃取平安XX经验" (Extract Ping An XX experience) - "萃取平安知识" (Extract Ping An knowledge) - "设计访谈提纲" (Design interview outline) - "萃取报告" (Extraction report) - "平安AI萃取" (Ping An AI extraction) - "平安知识萃取" (Ping An knowledge extraction) - "平安经验萃取" (Ping An experience extraction) - Any task involving extracting, summarizing, or documenting experiences from Ping An organizations This skill automatically handles: 1. Batch document conversion (DOCX/XLSX/PPTX → Markdown) for fast processing 2. Parallel chunk-based processing (9-part methodology for large projects) 3. Interview outline generation for 4 types of interviewees (strategic, business, IT product, execution) 4. Extraction report generation using 4 frameworks
| name | knowledge-interviewer |
| description | 从访谈录音/文本中萃取结构化知识。TRIGGER when: 用户需要"萃取"、"提取知识"、"访谈分析"、 "知识萃取报告"、从DOCX/TXT提取专家经验。支持单文件和多文件综合萃取，自动检测输入类型。 输出严格按 extraction-schema.json 的 JSON 结构。 SKIP: 生成 HTML/PPTX/DOCX 交付物（那是 zhongwen-content 的职责）。 |
从领域专家访谈/调研文档中提取结构化知识，输出 JSON 格式的萃取结果。
支持两种模式（自动检测）：
knowledge_gaps，不编造metadata.source 列出所有来源文件检查用户指定的输入路径：
如果是 .docx 文件，需先用 python-docx 提取文本（同 Phase 1 逻辑）。
Read: C:\Users\xjtul\.claude\projects\C--Users-xjtul\templates\extraction-schema.json
理解每个字段的含义和层级。输出必须严格匹配 schema 的 required 字段和类型。
单文件模式：读取指定的单个文件。
综合模式：
对照 schema 的 10 个顶层字段（title/metadata/overview/core_products/methodology_framework/core_philosophy/best_practices/competitive_insights/future_directions/key_quotes/knowledge_gaps）逐一填充。
| 字段 | 填充要点 |
|---|---|
title | {平台/产品名}知识萃取报告 |
metadata | 日期、来源（综合模式列出所有文件）、受访者、素材类型、质量说明 |
overview | 200-300字全景概述，含定位、产品矩阵、突破方向 |
core_products | 每个产品5字段(product_name/tag/positioning/maturity/key_highlights)，子模块用 sub_modules |
methodology_framework | 萃取出的方法论，每项含 stages(3-5步)/description/application |
core_philosophy | 核心理念列表，每项含 name/description/quote_ref |
best_practices | STAR2框架(Situation-Task-Action-Result-Reflection) |
competitive_insights | 竞争维度/我方优势/行业现状/关键差异 |
future_directions | 3-6项未来方向，标注优先级(高/中/低) |
key_quotes | 8-15条，含 text/speaker/context/topic_tag |
knowledge_gaps | 5-10项未覆盖知识点，标注优先级和补充来源 |
综合模式特殊处理：
metadata.source：列出所有源文件名metadata.material_type：标注为"多文档综合分析"overview：覆盖所有文档的共同主题core_products：去重合并，同一产品在多文档中出现的合并为一个条目best_practices：案例可来自不同文档，标注来源key_quotes：可来自不同文档，speaker 标注来源文件名knowledge_gaps：综合判断，跨文档知识缺口将填充完成的 JSON 写入用户指定（或默认）的输出路径。
文件命名规则：
{源文件名}_萃取报告.json{文件夹名}_综合萃取报告.json使用默认模型（Opus），适合分析型任务（非结构化文本→结构化知识）。
knowledge-interviewer 只负责"文本→JSON"这一步。 由主对话 orchestrator 调用 zhongwen-content（Sonnet 模型）完成"JSON→HTML"渲染。 两者不互相调用。