1. 增加统计信息并上报逻辑，用于对账；2. 上报失败超过最大重试次数的消息记录到dead letter jsonl中

docs/config-list.md

@@ -194,8 +194,8 @@
 | `CSC_RAW_DUMP_BASE_URL` | string | - | 自定义上报服务端地址 |
 | `COSTRICT_RAW_DUMP_BASE_URL` | string | - | 兼容 opencode 的自定义上报地址 |
 | `COSTRICT_BASE_URL` | string | `https://zgsm.sangfor.com` | CoStrict 服务地址 |
-| `CSC_RAW_DUMP_LOCAL_MODE` | boolean | `false` | **本地留存模式**：数据只写入本地文件，不上报服务端 |
+| `CSC_RAW_DUMP_MODE` | boolean | `false` | **本地留存模式**：数据只写入本地文件，不上报服务端 |
-| `CSC_RAW_DUMP_LOCAL_DIR` | string | `~/.claude/raw-dump-local` | 本地留存目录 |
+| `CSC_RAW_DUMP_DIR` | string | `~/.claude/raw-dump` | 本地留存目录 |
 
 ### Bash/终端
 

src/services/rawDump/batchWorker.ts

@@ -16,6 +16,7 @@ import {
   uploadConversation,
   uploadSummary,
   uploadCommits,
+  uploadStatistics,
   authWithFallback,
 } from './worker.js'
 import {
@@ -29,8 +30,7 @@ import {
   MAX_ATTEMPTS,
   type QueueTask,
 } from './queue.js'
-import { readState, writeState } from './state.js'
+import { readState, writeState, appendDeadLetter } from './state.js'
-import type { RawDumpError } from './types.js'
 import { getSessionDirectory, loadSessionMessages } from './worker.js'
 import { getRepoInfo } from './git.js'
 import { createLogger } from './logger.js'
@@ -134,6 +134,25 @@ async function processTask(
   const authData = await authWithFallback()
   const repoInfo = await getCachedRepoInfo(task.directory)
 
+  if (task.messageID === '__statistics__' && task.statsData) {
+    await uploadStatistics(
+      {
+        sessionID: task.sessionID,
+        directory: task.directory,
+        sessionCount: task.statsData.sessionCount,
+        conversationCount: task.statsData.conversationCount,
+        upstreamTokens: task.statsData.upstreamTokens,
+        downstreamTokens: task.statsData.downstreamTokens,
+        startTime: task.statsData.startTime,
+        endTime: task.statsData.endTime,
+      },
+      authData,
+      state,
+    )
+    log.info('statistics task completed', { sessionID: task.sessionID })
+    return
+  }
+
   const conversationUploaded = await uploadConversation(
     {
       sessionID: task.sessionID,
@@ -156,7 +175,10 @@ async function processTask(
     repoInfo,
   })
 
-  log.info('task completed', { sessionID: task.sessionID, conversationUploaded })
+  log.info('task completed', {
+    sessionID: task.sessionID,
+    conversationUploaded,
+  })
 }
 
 /**
@@ -218,13 +240,16 @@ async function runBatch() {
           })
 
           if (task.attemptCount >= MAX_ATTEMPTS) {
-            // 彻底失败：记录错误，从队列移除
+            // 彻底失败：写入 dead letter，从队列移除
-            state.errors[key] = {
+            await appendDeadLetter({
-              message: errorMsg,
+              sessionID: task.sessionID,
-              count: task.attemptCount,
+              messageID: task.messageID,
-              lastAt: new Date().toISOString(),
+              directory: task.directory,
+              attemptCount: task.attemptCount,
+              error: errorMsg,
               endpoint: extractEndpointFromError(errorMsg),
-            } satisfies RawDumpError
+              failedAt: new Date().toISOString(),
+            })
             removeTask(key)
             log.warn('task permanently failed, removed from queue', {
               sessionID: task.sessionID,
@@ -232,12 +257,6 @@ async function runBatch() {
             })
           } else {
             // 失败但未超限：attemptCount++ 写回文件，下次重试
-            state.errors[key] = {
-              message: errorMsg,
-              count: task.attemptCount,
-              lastAt: new Date().toISOString(),
-              endpoint: extractEndpointFromError(errorMsg),
-            } satisfies RawDumpError
             await flushQueue()
           }
         }
@@ -279,15 +298,17 @@ export function startBatchWorker() {
   }
 
   // 启动时加载队列 + 随机抖动
-  loadQueue().then(() => {
+  loadQueue()
-    log.info('queue loaded', { count: getQueue().length })
+    .then(() => {
-    scheduleNext(Math.floor(Math.random() * 10_000))
+      log.info('queue loaded', { count: getQueue().length })
-  }).catch(err => {
+      scheduleNext(Math.floor(Math.random() * 10_000))
-    log.error('failed to load queue', {
+    })
-      error: err instanceof Error ? err.message : String(err),
+    .catch(err => {
+      log.error('failed to load queue', {
+        error: err instanceof Error ? err.message : String(err),
+      })
+      process.exit(1)
     })
-    process.exit(1)
-  })
 }
 
 const scriptPath = process.argv[1] || ''
@@ -296,4 +317,4 @@ if (
   scriptPath.endsWith('batchWorker.js')
 ) {
   startBatchWorker()
-}
+}

src/services/rawDump/index.ts

@@ -3,7 +3,10 @@
  * 队列模式：主进程只 enqueue，单 batch worker 顺序消费
  */
 
-import { getRawDumpMode, RAW_DUMP_MODE } from './localStorage.js'
+import {
+  getRawDumpMode,
+  RAW_DUMP_MODE,
+} from './localStorage.js'
 import { enqueue } from './queue.js'
 import { spawnBatchWorker } from './spawn.js'
 import { startBatchWorker } from './batchWorker.js'
@@ -103,3 +106,42 @@ export function reportSession(sessionID: string, directory: string): void {
   enqueue({ sessionID, messageID: '__summary__', directory })
   ensureBatchWorker()
 }
+
+export interface StatisticsData {
+  sessionCount: number
+  conversationCount: number
+  upstreamTokens: number
+  downstreamTokens: number
+  startTime: number
+  endTime: number
+}
+
+const lastReportStatsMap = new Map<string, number>()
+const STATS_DEBOUNCE_MS = 60_000 // 同一 session 1 分钟内不重复 enqueue
+
+/**
+ * 上报对账统计数据（session数、conversation数、token数）
+ * 通过特殊 messageID '__statistics__' 标识入队
+ */
+export function reportStatistics(
+  sessionID: string,
+  directory: string,
+  data: StatisticsData,
+): void {
+  if (!isEnabled()) return
+  const key = `${sessionID}:__statistics__`
+  const now = Date.now()
+  const last = lastReportStatsMap.get(key)
+  if (last && now - last < STATS_DEBOUNCE_MS) {
+    log.debug('reportStatistics debounced', { sessionID, lastMs: now - last })
+    return
+  }
+  lastReportStatsMap.set(key, now)
+  enqueue({
+    sessionID,
+    messageID: '__statistics__',
+    directory,
+    statsData: data,
+  })
+  ensureBatchWorker()
+}

src/services/rawDump/queue.ts

@@ -12,6 +12,7 @@
 import { promises as fs } from 'node:fs'
 import os from 'node:os'
 import path from 'path'
+import type { StatisticsData } from './index.js'
 
 const QUEUE_FILE = path.join(
   os.homedir(),
@@ -29,6 +30,7 @@ export interface QueueTask {
   directory: string
   enqueuedAt: number
   attemptCount: number
+  statsData?: StatisticsData
 }
 
 // ----------- 内存中的队列 -----------

src/services/rawDump/state.ts

@@ -1,21 +1,22 @@
 /**
  * Raw Dump 磁盘状态管理
- * 用于 conversation、summary、commits 的去重，以及错误聚合
+ * 用于 conversation、summary、commits 的去重
  * 通过文件锁保证多进程并发读写安全
  *
- * state.errors：按 sessionID:messageID 键控的错误聚合表
+ * 错误追踪（dead letter）：
- * - 每批次累加 count，记录最新错误消息和时间戳
+ * - 超过最大重试次数的任务追加写入 dead letter jsonl 文件
- * - 超过最大重试次数时由 batchWorker 写入 dead letter
+ * - 路径: ~/.claude/raw-dump/csc-dead-letter.jsonl
  */
 
 import { promises as fs, readFileSync, writeFileSync } from 'node:fs'
 import os from 'node:os'
 import path from 'node:path'
-import type { RawDumpState } from './types.js'
+import type { RawDumpState, DeadLetterEntry } from './types.js'
 
 const STATE_DIR = path.join(os.homedir(), '.claude', 'raw-dump')
 const STATE_FILE = path.join(STATE_DIR, 'csc-state.json')
 const STATE_LOCK_FILE = path.join(STATE_DIR, 'csc-state.lock')
+const DEAD_LETTER_FILE = path.join(STATE_DIR, 'csc-dead-letter.jsonl')
 
 /**
  * 创建空的 RawDumpState 对象
@@ -26,8 +27,6 @@ function createEmptyState(): RawDumpState {
     conversation: {},
     summary: {},
     commits: {},
-    errors: {},
-    // summary 值为 RFC3339 格式时间戳字符串，解析时无需转换
   }
 }
 
@@ -106,7 +105,6 @@ export async function readState(): Promise<RawDumpState> {
         conversation: parsed.conversation ?? {},
         summary: parsed.summary ?? {},
         commits: parsed.commits ?? {},
-        errors: parsed.errors ?? {},
       }
     } catch {
       return createEmptyState()
@@ -126,3 +124,23 @@ export async function writeState(state: RawDumpState): Promise<void> {
     await fs.writeFile(STATE_FILE, JSON.stringify(state, null, 2), 'utf-8')
   })
 }
+
+export async function readDeadLetter(): Promise<DeadLetterEntry[]> {
+  try {
+    const text = await fs.readFile(DEAD_LETTER_FILE, 'utf-8')
+    return text
+      .split('\n')
+      .filter(Boolean)
+      .map(line => JSON.parse(line) as DeadLetterEntry)
+  } catch {
+    return []
+  }
+}
+
+export async function appendDeadLetter(entry: DeadLetterEntry): Promise<void> {
+  await fs.mkdir(STATE_DIR, { recursive: true })
+  await fs.writeFile(DEAD_LETTER_FILE, JSON.stringify(entry) + '\n', {
+    flag: 'a',
+    encoding: 'utf-8',
+  })
+}

src/services/rawDump/types.ts

@@ -7,10 +7,9 @@
  * - summary：会话统计（按 session 去重，5 分钟内只上报一次）
  * - commits：提交记录（按 commit ID 去重）
  *
- * 错误追踪：
+ * 错误追踪（dead letter）：
- * - 上报失败的任务写入 failed queue，最多重试 MAX_RETRIES 次
+ * - 超过最大重试次数的任务追加写入 dead letter jsonl 文件
- * - 超过最大重试次数的任务移入 dead letter 文件
+ * - 路径: ~/.claude/raw-dump/csc-dead-letter.jsonl
- * - 所有错误聚合到 state.errors，按 sessionID:messageID 键控
  */
 
 export const RAW_DUMP_EVENT_ENV_KEY = '__CSC_RAW_DUMP_EVENT__'
@@ -25,14 +24,16 @@ export interface RawDumpState {
   conversation: Record<string, true>
   summary: Record<string, string> // RFC3339 时间戳，如 "2024-01-01T12:00:00.000Z"
   commits: Record<string, string>
-  errors: Record<string, RawDumpError>
 }
 
-export interface RawDumpError {
+export interface DeadLetterEntry {
-  message: string
+  sessionID: string
-  count: number
+  messageID: string
-  lastAt: string
+  directory: string
+  attemptCount: number
+  error: string
   endpoint?: string
+  failedAt: string
 }
 
 export interface JwtPayload {
@@ -107,3 +108,17 @@ export interface CommitPayload {
   subject: string
   parent_ids: string[]
 }
+
+export interface StatisticsPayload {
+  task_id: string
+  start_time: string
+  end_time: string
+  user_id: string
+  user_name: string
+  client_id: string
+  client_version: string
+  session_count: number
+  conversation_count: number
+  upstream_tokens: number
+  downstream_tokens: number
+}

src/services/rawDump/worker.ts

@@ -37,6 +37,7 @@ import type {
   CommitPayload,
   ConversationPayload,
   JwtPayload,
+  StatisticsPayload,
   SummaryPayload,
 } from './types.js'
 
@@ -1072,6 +1073,64 @@ export async function uploadCommits(
   return commits.length
 }
 
+const STATS_DEDUP_WINDOW_MS = 60 * 60 * 1000 // 同一 session 1 小时内只上报一次 statistics
+
+export async function uploadStatistics(
+  payload: {
+    sessionID: string
+    directory: string
+    sessionCount: number
+    conversationCount: number
+    upstreamTokens: number
+    downstreamTokens: number
+    startTime: number
+    endTime: number
+  },
+  authData: Awaited<ReturnType<typeof auth>>,
+  state: Awaited<ReturnType<typeof readState>>,
+): Promise<void> {
+  const key = `stats:${payload.sessionID}`
+  const lastReported = state.summary[key]
+  if (
+    lastReported &&
+    Date.now() - new Date(lastReported).getTime() < STATS_DEDUP_WINDOW_MS
+  ) {
+    log.debug('statistics skipped: reported recently', {
+      task_id: payload.sessionID,
+      lastReported,
+    })
+    return
+  }
+
+  const body: StatisticsPayload = {
+    task_id: payload.sessionID,
+    start_time: formatIso(payload.startTime),
+    end_time: formatIso(payload.endTime),
+    ...authData.user,
+    client_id: authData.clientId,
+    client_version: authData.version,
+    session_count: payload.sessionCount,
+    conversation_count: payload.conversationCount,
+    upstream_tokens: payload.upstreamTokens,
+    downstream_tokens: payload.downstreamTokens,
+  }
+
+  await postJson(
+    authData.baseUrl,
+    authData.headers,
+    '/raw-store/statistics',
+    body,
+  )
+  state.summary[key] = new Date().toISOString()
+  log.info('statistics uploaded', {
+    task_id: payload.sessionID,
+    session_count: payload.sessionCount,
+    conversation_count: payload.conversationCount,
+    upstream_tokens: payload.upstreamTokens,
+    downstream_tokens: payload.downstreamTokens,
+  })
+}
+
 /**
  * 从环境变量中解析 worker 入口传入的 payload
  * 抛出错误如果环境变量不存在或 JSON 解析失败