feat: 新增微信图片文字索引与问问微信图片检索能力

tests/unit/ask-wechat-service.test.ts

@@ -377,7 +377,7 @@ describe('AskWechatService — 桌面问问微信主路径', () => {
     expect(result.diagnostics.outcome).toBe('invalid_question')
   })
 
-  it('生产日志只记录形态字段，不含回答内容', async () => {
+  it('本地日志必须包含问题与模型回答原文，方便回放排查', async () => {
     const logs: AskWechatLogRecord[] = []
     const { provider } = providerFactory([answer('BOBO 最近在准备搬家，提到了房租和押金。')])
     const service = new AskWechatService(new QueryAgentService(provider, vi.fn()), {
@@ -388,17 +388,53 @@ describe('AskWechatService — 桌面问问微信主路径', () => {
     await service.ask(request('BOBO 最近在忙什么'))
 
     expect(logs).toHaveLength(1)
-    expect(Object.keys(logs[0].details ?? {}).sort()).toEqual([
-      'entry',
-      'model',
-      'modelCallCount',
-      'outcome',
-      'provider',
-      'toolCallCount',
-      'tools',
-      'totalMs'
-    ])
-    expect(JSON.stringify(logs)).not.toContain('准备搬家')
+    // 形态字段仍然一个不少（排查时既要知道"问了什么"，也要知道"跑了什么工具、多久"）。
+    for (const key of ['entry', 'model', 'modelCallCount', 'outcome', 'provider', 'toolCallCount', 'tools', 'totalMs']) {
+      expect(logs[0].details).toHaveProperty(key)
+    }
+    // 措辞回归：真机上曾出现"图片已识别出文字却答'没有取得 OCR 文字'"，
+    // 当时日志里只有工具名与次数，无法判断是索引没建还是链路没接上。
+    // 现在问答原文进入**本机**日志（不上传、不进遥测），可以直接回放。
+    expect(logs[0].details?.question).toBe('BOBO 最近在忙什么')
+    expect(String(logs[0].details?.answer)).toContain('准备搬家')
+  })
+
+  it('图片 OCR 的两条结构化事实进入诊断（不重复正文）', async () => {
+    const logs: AskWechatLogRecord[] = []
+    const { provider } = providerFactory([answer('那张图里有价格文字。')])
+    const runtime = new QueryAgentService(provider, vi.fn())
+    // 直接在 Runtime 结果里放一条带图片 OCR 的 trace，验证聚合口径。
+    vi.spyOn(runtime, 'run').mockResolvedValue({
+      question: '图里写了什么',
+      provider: 'Fixture',
+      model: 'fixture',
+      modelCallCount: 1,
+      toolCallCount: 1,
+      toolTotalMs: 5,
+      totalMs: 10,
+      modelDurationsMs: [],
+      modelDiagnostics: [],
+      answer: '那张图里有价格文字。',
+      traces: [
+        {
+          toolName: 'query_messages',
+          input: {},
+          durationMs: 5,
+          status: 'completed',
+          imageOcrTextCount: 3,
+          imageOcrCoverageState: 'partial'
+        }
+      ]
+    } as never)
+    const service = new AskWechatService(runtime, {
+      entry: 'desktop',
+      log: (record) => logs.push(record)
+    })
+
+    await service.ask(request('图里写了什么'))
+
+    expect(logs[0].details?.imageOcrTextCount).toBe(3)
+    expect(logs[0].details?.imageOcrCoverageState).toBe('partial')
   })
 
   it('forgetConversation 清掉指定会话的澄清上下文', async () => {

tests/unit/query-agent-image-ocr-coverage.test.ts

@@ -0,0 +1,137 @@
+/**
+ * §6 / §7 / §18：图片文字索引覆盖度在 Query Agent 这一层的语义。
+ *
+ * 这里能确定性验证的是"覆盖度**真的进到了模型上下文**"，而不是只做了个 UI 数字：
+ * - 系统提示词把 imageOcrCoverage 定义成**独立于文字索引**的维度，并禁止凭零结果下"没有"；
+ * - tool result 透传里真的带着 imageOcrCoverage（模型能看见比例与结论句）。
+ *
+ * 真模型最终怎么说话不在本文件断言范围内（那需要真实模型与真实数据）。
+ */
+import { describe, expect, it, vi } from 'vitest'
+import {
+  QueryAgentService,
+  type QueryAgentProvider
+} from '../../src/main/services/query-agent-service'
+
+function provider(
+  responses: Array<Awaited<ReturnType<QueryAgentProvider['chatWithTools']>>>,
+  configured = true
+): QueryAgentProvider {
+  return {
+    getRuntimeConfig: () => ({
+      configured,
+      providerName: 'Fixture Provider',
+      model: 'fixture-model',
+      modelName: 'Fixture Model'
+    }),
+    chatWithTools: vi.fn(async () => responses.shift() || { success: true, data: 'done' })
+  }
+}
+
+/** 与 `buildImageOcrCoverage` 在 partial 时产出的结论句保持一致。 */
+const PARTIAL_SUMMARY =
+  '图片文字索引只完成 30%（30 / 100 条图片消息），当前图片搜索结果可能不完整。涉及图片、截图、海报里的文字的问题，当前不能因为没搜到就回答"没有"。'
+
+function searchOnce(execute: ReturnType<typeof vi.fn>, finalAnswer: string): QueryAgentProvider {
+  return provider([
+    {
+      success: true,
+      toolCalls: [
+        {
+          id: 'call-1',
+          name: 'search_messages',
+          arguments: JSON.stringify({
+            target: { query: '技术交流群' },
+            timeRange: { kind: 'all' },
+            queries: ['ChatGPT 价格']
+          })
+        }
+      ]
+    },
+    { success: true, data: finalAnswer }
+  ])
+}
+
+describe('Query Agent 的图片文字索引覆盖度语义', () => {
+  it('系统提示词把图片覆盖度声明为独立维度，并禁止凭零结果断言"没有"', async () => {
+    const execute = vi.fn(async () => ({ status: 'completed', evidenceCount: 0, evidence: [] }))
+    const configured = searchOnce(execute, 'ok')
+    await new QueryAgentService(configured, execute).run('图片文字索引相关的问题')
+
+    const systemPrompt = String(vi.mocked(configured.chatWithTools).mock.calls[0]?.[0]?.[0]?.content)
+    expect(systemPrompt).toContain('imageOcrCoverage')
+    expect(systemPrompt).toContain('独立于文字索引')
+    expect(systemPrompt).toContain('not_built')
+    // 零结果诚实性必须写死在提示词里，不能指望模型自己想到。
+    expect(systemPrompt).toContain('绝不能')
+  })
+
+  it('partial 覆盖度随 tool result 进入模型上下文，而不是只留在 UI 里', async () => {
+    const execute = vi.fn(async () => ({
+      status: 'completed',
+      coverage: { state: 'partial' },
+      evidenceCount: 0,
+      evidence: [],
+      imageOcrCoverage: {
+        state: 'partial',
+        totalImageMessages: 100,
+        processed: 30,
+        indexed: 28,
+        empty: 2,
+        missing: 0,
+        failed: 0,
+        pending: 70,
+        countedAtLabel: '09-15 20:13',
+        summary: PARTIAL_SUMMARY
+      }
+    }))
+    const configured = searchOnce(execute, '图片文字索引只完成 30%，暂时无法确认。')
+    await new QueryAgentService(configured, execute).run(
+      '技术交流群之前是不是发过 ChatGPT 价格的图片？'
+    )
+
+    const toolMessage = vi
+      .mocked(configured.chatWithTools)
+      .mock.calls[1]?.[0].find((message) => message.role === 'tool')
+    const presented = JSON.parse(String(toolMessage?.content)) as Record<string, any>
+
+    expect(presented.imageOcrCoverage).toMatchObject({
+      state: 'partial',
+      totalImageMessages: 100,
+      processed: 30,
+      indexed: 28,
+      empty: 2,
+      pending: 70
+    })
+    expect(presented.imageOcrCoverage.summary).toContain('不能因为没搜到就回答')
+  })
+
+  it('complete 覆盖度不下发零结果约束（避免模型机械附加警告）', async () => {
+    const execute = vi.fn(async () => ({
+      status: 'completed',
+      coverage: { state: 'complete' },
+      evidenceCount: 1,
+      evidence: [{ messageRef: 'opaque', timestamp: 1, sender: '张三', sourceKind: 'image', text: 'ChatGPT Plus $20' }],
+      imageOcrCoverage: {
+        state: 'complete',
+        totalImageMessages: 100,
+        processed: 100,
+        indexed: 90,
+        empty: 10,
+        missing: 0,
+        failed: 0,
+        pending: 0,
+        summary: '图片文字索引已覆盖所统计的全部 100 条图片消息。'
+      }
+    }))
+    const configured = searchOnce(execute, '找到了。')
+    await new QueryAgentService(configured, execute).run('技术交流群发过 ChatGPT 价格的图片吗')
+
+    const toolMessage = vi
+      .mocked(configured.chatWithTools)
+      .mock.calls[1]?.[0].find((message) => message.role === 'tool')
+    const presented = JSON.parse(String(toolMessage?.content)) as Record<string, any>
+    expect(presented.imageOcrCoverage.state).toBe('complete')
+    expect(presented.imageOcrCoverage.summary).not.toContain('不能因为没搜到就回答')
+  })
+})

tests/unit/wcdb-image-message-count.test.ts

@@ -0,0 +1,165 @@
+/**
+ * 图片消息计数 / 增量的**标识符与列名**回归测试。
+ *
+ * 真机反馈「检测到图片消息为 0 → 无法统计（未找到该会话的消息表）」，根因有两个，
+ * 都在这里钉死：
+ *
+ * 1. **标识符混淆**：`contact.md5` 是 `md5(wxid)` 的哈希，而原生接口
+ *    （`wcdbGetMessageTableStats` / `wcdbGetMessages`）要的是**原始 username**。
+ *    把 md5 直接当 username 传，原生侧匹配不到任何表 —— 既有读消息路径一直有做这层转换
+ *    （`wechat-db.chatMd5ToUsername`、`chat-service` 的 `getUsernameByMd5`），统计路径漏了。
+ * 2. **类型列名硬编码**：不同微信版本的列名不统一（项目读行时用的是别名列表），
+ *    把 `local_type` 写进 WHERE 会在列名不同的库上抛错。
+ */
+import { createHash } from 'node:crypto'
+import { describe, expect, it, vi } from 'vitest'
+import { Wcdb4Client } from '../../src/main/wcdb4-client'
+
+const USERNAME = 'wxid_real_user'
+const ROOM = '12345678@chatroom'
+const MD5_OF_USERNAME = createHash('md5').update(USERNAME).digest('hex')
+const MD5_OF_ROOM = createHash('md5').update(ROOM).digest('hex')
+
+type ClientOptions = {
+  sessions?: { username: string }[]
+  chatTables?: { name: string; db_number: string }[] | null
+  columns?: string[]
+  count?: number
+  maxLocalId?: number
+  identifierLog?: string[]
+}
+
+function makeClient(options: ClientOptions): Wcdb4Client {
+  const identifierLog = options.identifierLog ?? []
+  return Object.assign(Object.create(Wcdb4Client.prototype), {
+    cachedSessions: (options.sessions ?? []).map((session) => ({ ...session })),
+    cachedChatTables: options.chatTables ?? null,
+    messageTypeColumnCache: new Map<string, string | null>(),
+    messageTypeColumnCandidates: [
+      'local_type',
+      'localType',
+      'msg_type',
+      'msgType',
+      'message_type',
+      'messageType',
+      'type',
+      'WCDB_CT_local_type'
+    ],
+    // 断言点：走到消息表查找时用的标识符必须是 username，不是 md5。
+    listMessageStoresAsync: vi.fn(async (identifier: string) => {
+      identifierLog.push(identifier)
+      return [{ tableName: 'Msg_fixture', dbPath: 'C:/fixture/message_0.db' }]
+    }),
+    // 聚合查询的返回（真实实现由 pickValue 解析）。
+    callJsonAsync: vi.fn(async () => [
+      { image_count: options.count ?? 0, image_max_local_id: options.maxLocalId ?? 0 }
+    ]),
+    readMessageColumns: vi.fn(() =>
+      (options.columns ?? []).map((name) => ({ name, declaration: 'INTEGER' }))
+    ),
+    wcdbGetMessageTableStats: vi.fn(async () => 0),
+    wcdbExecQuery: vi.fn(async () => 0)
+  }) as Wcdb4Client
+}
+
+describe('图片消息统计：会话标识符必须是 username 而不是 md5', () => {
+  it('从 session 能把 md5 反解成 username，并把它交给消息表查找', async () => {
+    const identifierLog: string[] = []
+    const client = makeClient({
+      sessions: [{ username: USERNAME }],
+      columns: ['local_type'],
+      count: 7,
+      identifierLog
+    })
+
+    const result = await client.countImageMessagesAsync(MD5_OF_USERNAME)
+
+    expect(result).toEqual({ count: 7, typeColumn: 'local_type' })
+    // ★ 核心回归断言：绝不能把 md5 当 username 传下去。
+    expect(identifierLog).toEqual([USERNAME])
+    expect(identifierLog).not.toContain(MD5_OF_USERNAME)
+  })
+
+  it('群聊同样走 md5 → username', async () => {
+    const identifierLog: string[] = []
+    const client = makeClient({
+      sessions: [{ username: ROOM }],
+      columns: ['local_type'],
+      count: 2,
+      identifierLog
+    })
+
+    await client.countImageMessagesAsync(MD5_OF_ROOM)
+    expect(identifierLog).toEqual([ROOM])
+  })
+
+  it('不在 session 列表、只以 Chat_<md5> 表存在的会话也能反解', async () => {
+    const identifierLog: string[] = []
+    const client = makeClient({
+      sessions: [],
+      chatTables: [{ name: `Chat_${MD5_OF_ROOM}`, db_number: ROOM }],
+      columns: ['local_type'],
+      count: 3,
+      identifierLog
+    })
+
+    const result = await client.countImageMessagesAsync(MD5_OF_ROOM)
+    expect(result.count).toBe(3)
+    expect(identifierLog).toEqual([ROOM])
+  })
+
+  it('增量水位使用同一个标识符（否则水位永远拿不到、退化成每轮重扫）', async () => {
+    const identifierLog: string[] = []
+    const client = makeClient({
+      sessions: [{ username: USERNAME }],
+      columns: ['local_type'],
+      count: 7,
+      maxLocalId: 42,
+      identifierLog
+    })
+
+    const watermark = await client.imageConversationWatermarkAsync(MD5_OF_USERNAME)
+    expect(watermark).toEqual({ count: 7, maxLocalId: 42 })
+    expect(identifierLog).toEqual([USERNAME])
+  })
+})
+
+describe('图片消息统计：类型列名随版本变化', () => {
+  it('列名不是 local_type 时照样能统计，并把真实列名透出', async () => {
+    const client = makeClient({
+      sessions: [{ username: USERNAME }],
+      columns: ['msg_type'],
+      count: 5
+    })
+
+    const result = await client.countImageMessagesAsync(MD5_OF_USERNAME)
+    expect(result).toEqual({ count: 5, typeColumn: 'msg_type' })
+  })
+
+  it('探测不到类型列时明确失败，而不是静默返回 0', async () => {
+    const client = makeClient({ sessions: [{ username: USERNAME }], columns: [] })
+
+    const result = await client.countImageMessagesAsync(MD5_OF_USERNAME)
+    // "数不出来"必须是 null + 原因，不能是 0。
+    expect(result.count).toBeNull()
+    expect(result.error).toBe('消息表缺少可识别的消息类型列')
+  })
+
+  it('一张表都没匹配到时给出可诊断的原因', async () => {
+    const client = makeClient({ sessions: [{ username: USERNAME }], columns: ['local_type'] })
+    Object.assign(client, { listMessageStoresAsync: vi.fn(async () => []) })
+
+    const result = await client.countImageMessagesAsync(MD5_OF_USERNAME)
+    expect(result.count).toBeNull()
+    expect(result.error).toBe('未找到该会话的消息表')
+  })
+
+  it('原生统计通道不可用时说明是环境问题，与 0 张区分开', async () => {
+    const client = makeClient({ sessions: [{ username: USERNAME }], columns: ['local_type'] })
+    Object.assign(client, { wcdbExecQuery: null })
+
+    const result = await client.countImageMessagesAsync(MD5_OF_USERNAME)
+    expect(result.count).toBeNull()
+    expect(result.error).toBe('当前数据服务不支持消息表统计')
+  })
+})