# 业务路由推荐 — 时序图

> 核心流程：员工发送非IT消息 → Dify识别 → 后端发送名片 → 前端渲染 → 点击跳转企微

```mermaid
sequenceDiagram
    autonumber
    participant E as 员工(H5)
    participant H5 as H5前端
    participant BE as 后端API<br/>(h5.py)
    participant TASK as 后台AI任务<br/>(h5_ai_task.py)
    participant RS as 路由服务<br/>(routing_service.py)
    participant DIFY as Dify API<br/>(统一意图识别)
    participant DB as PostgreSQL
    participant WS as WebSocket<br/>(ws_manager)
    participant AG as 坐席端
    participant WECOM as 企微

    %% ===== 第1阶段：员工发送消息 =====
    rect rgb(240, 248, 255)
        E->>H5: 发送非IT消息 "打印机坏了"
        H5->>BE: POST /h5/conversations/current/messages
        BE->>DB: 存储员工消息(msg_type=text)
        BE->>WS: broadcast(new_message → 坐席)
        BE-->>H5: 即时返回(user_message, ai_reply=null)
        BE->>TASK: asyncio.create_task(process_h5_ai_reply)
    end

    %% ===== 第2阶段：后台任务 - 路由检测 =====
    rect rgb(255, 248, 240)
        Note over TASK: ① BYOD关键词检查(未命中，跳过)
        
        TASK->>RS: 路由关键词预过滤("打印机")
        RS-->>TASK: 命中 → 需调Dify
        
        TASK->>RS: detect_routing_intent(text, employee_id)
        RS->>DIFY: POST /v1/chat-messages<br/>{query: "打印机坏了"}
        DIFY-->>RS: {intent_type: "non_it_routing",<br/>business_category: "行政",<br/>routing_confidence: 0.85}
        RS-->>TASK: 路由结果(非IT, 行政, 0.85)
        
        Note over TASK: routing_confidence ≥ 0.7 → 触发名片推荐
    end

    %% ===== 第3阶段：查询联系人 + 发送名片 =====
    rect rgb(240, 255, 240)
        TASK->>RS: get_contact_by_category("行政")
        RS->>DB: SELECT * FROM business_contacts<br/>WHERE category='行政' AND is_active=true LIMIT 1
        DB-->>RS: 返回联系人(王芳, wecom_userid=WangFang)
        RS-->>TASK: BusinessContact对象
        
        TASK->>RS: send_contact_card(db, conversation, employee_id, contact, reason)
        
        %% 消息1: 路由说明文本
        RS->>DB: 存Message(ai, text, "打印机问题属于行政...")
        RS->>WS: broadcast_to_employees([employee_id], ai_reply)
        WS-->>H5: ai_reply(路由文本)
        RS->>WS: broadcast(new_message → 坐席)
        WS-->>AG: new_message(路由文本)
        
        %% 消息2: 名片卡片
        RS->>DB: 存Message(ai, contact_card, extra_data={contact, reason})
        RS->>WS: broadcast_to_employees([employee_id], ai_reply)
        WS-->>H5: ai_reply(msg_type=contact_card, extra_data)
        RS->>WS: broadcast(new_message → 坐席)
        WS-->>AG: new_message(msg_type=contact_card, extra_data)
        
        %% 消息3: 系统提示
        RS->>DB: 存Message(system, "以上为AI自动推荐...")
        RS->>WS: broadcast_to_employees([employee_id], ai_reply)
        WS-->>H5: ai_reply(系统提示)
        RS->>WS: broadcast(new_message → 坐席)
        WS-->>AG: new_message(系统提示)
        
        %% P1: 记录路由事件
        RS->>DB: INSERT routing_event(P1统计)
    end

    %% ===== 第4阶段：前端渲染 =====
    rect rgb(255, 245, 238)
        H5->>H5: MessageBubble渲染<br/>路由文本(AI气泡)
        H5->>H5: MessageBubble渲染<br/>ContactCard名片卡片
        H5->>E: 显示名片(头像/姓名/部门/负责业务/联系TA)
        H5->>H5: MessageBubble渲染<br/>系统消息(居中灰字)
        
        AG->>AG: MessageBubble渲染<br/>名片卡片(坐席端, 无联系TA按钮)
    end

    %% ===== 第5阶段：点击跳转 =====
    rect rgb(248, 240, 255)
        E->>H5: 点击「联系TA」
        H5->>H5: 读取 extra_data.contact.wecom_userid
        alt userid 非空 && SDK就绪
            H5->>WECOM: wx.invoke('openEnterpriseChat',<br/>{userids: 'WangFang', chatTitle: '王芳'})
            WECOM-->>E: 打开企微单聊窗口
        else userid为空 || SDK未就绪
            H5->>E: Toast("暂时无法发起聊天，请联系管理员")
        end
        
        Note over E,BE: 当前IT服务台会话不自动结束<br/>员工可返回继续咨询
    end
```

## 降级兜底流程

```mermaid
sequenceDiagram
    autonumber
    participant TASK as 后台AI任务
    participant RS as 路由服务
    participant DIFY as Dify API
    participant DB as PostgreSQL

    TASK->>RS: 路由关键词预过滤("工牌丢了")
    RS-->>TASK: 命中("工牌" → 人力资源)
    
    TASK->>RS: detect_routing_intent(text)
    RS->>DIFY: 调用统一意图识别
    
    alt Dify 调用成功
        DIFY-->>RS: {intent_type: "non_it_routing", business_category: "人力资源", routing_confidence: 0.88}
        RS-->>TASK: 使用Dify结果
    else Dify 调用失败/超时
        DIFY--xRS: 异常/超时
        RS->>RS: 降级兜底：ROUTING_KEYWORD_TO_CATEGORY["工牌"] = "人力资源"
        RS-->>TASK: 降级结果(business_category="人力资源", routing_confidence=0.6, source="fallback")
        Note over RS: confidence < 0.7 → 不触发名片推荐
        TASK->>TASK: 继续走正常AI回复流程
    end
```

## 坐席手动发名片流程（P1）

```mermaid
sequenceDiagram
    autonumber
    participant AG as 坐席端
    participant BE as 后端API<br/>(routing.py)
    participant RS as 路由服务
    participant DB as PostgreSQL
    participant WS as WebSocket
    participant H5 as H5前端

    AG->>AG: 点击「发名片」按钮
    AG->>BE: GET /routing/contacts?business_category=财务
    BE->>RS: get_contacts(category="财务")
    RS->>DB: SELECT * FROM business_contacts WHERE category='财务'
    DB-->>RS: 联系人列表
    RS-->>BE: 联系人列表
    BE-->>AG: 返回联系人列表
    
    AG->>AG: 坐席选择联系人(李明)
    AG->>BE: POST /conversations/{id}/send-contact-card<br/>{contact_id: 5, reason: "员工咨询报销问题"}
    
    BE->>RS: send_contact_card(db, conversation, agent_id, contact, reason)
    RS->>DB: 查询 BusinessContact by id=5
    DB-->>RS: 返回联系人(李明)
    
    RS->>DB: 创建 contact_card 消息<br/>(sender_type=ai, msg_type=contact_card,<br/>extra_data={contact, reason})
    RS->>WS: broadcast_to_employees([employee_id], ai_reply)
    WS-->>H5: ai_reply(contact_card)
    RS->>WS: broadcast(new_message → 坐席)
    WS-->>AG: new_message(contact_card)
    
    BE-->>AG: 返回成功
    H5->>H5: 渲染名片卡片
```