chore: sync VERSION file with release v1.1.232 [skip ci]

fix：修复codex统计token问题

CLAUDE.md

@@ -60,6 +60,7 @@ Claude Relay Service 是一个多平台 AI API 中转服务，支持 **Claude (
 
 - **apiKeyService.js**: API Key管理，验证、限流、使用统计、成本计算
 - **userService.js**: 用户管理系统，支持用户注册、登录、API Key管理
+- **userMessageQueueService.js**: 用户消息串行队列，防止同账户并发用户消息触发限流
 - **pricingService.js**: 定价服务，模型价格管理和成本计算
 - **costInitService.js**: 成本数据初始化服务
 - **webhookService.js**: Webhook通知服务
@@ -185,6 +186,10 @@ npm run service:stop          # 停止服务
 - `CLAUDE_OVERLOAD_HANDLING_MINUTES`: Claude 529错误处理持续时间（分钟，0表示禁用）
 - `STICKY_SESSION_TTL_HOURS`: 粘性会话TTL（小时，默认1）
 - `STICKY_SESSION_RENEWAL_THRESHOLD_MINUTES`: 粘性会话续期阈值（分钟，默认0）
+- `USER_MESSAGE_QUEUE_ENABLED`: 启用用户消息串行队列（默认false）
+- `USER_MESSAGE_QUEUE_DELAY_MS`: 用户消息请求间隔（毫秒，默认200）
+- `USER_MESSAGE_QUEUE_TIMEOUT_MS`: 队列等待超时（毫秒，默认5000，锁持有时间短无需长等待）
+- `USER_MESSAGE_QUEUE_LOCK_TTL_MS`: 锁TTL（毫秒，默认5000，请求发送后立即释放无需长TTL）
 - `METRICS_WINDOW`: 实时指标统计窗口（分钟，1-60，默认5）
 - `MAX_API_KEYS_PER_USER`: 每用户最大API Key数量（默认1）
 - `ALLOW_USER_DELETE_API_KEYS`: 允许用户删除自己的API Keys（默认false）
@@ -337,6 +342,7 @@ npm run setup  # 自动生成密钥并创建管理员账户
 11. **速率限制未清理**: rateLimitCleanupService每5分钟自动清理过期限流状态
 12. **成本统计不准确**: 运行 `npm run init:costs` 初始化成本数据，检查pricingService是否正确加载模型价格
 13. **缓存命中率低**: 查看缓存监控统计，调整LRU缓存大小配置
+14. **用户消息队列超时**: 优化后锁持有时间已从分钟级降到毫秒级（请求发送后立即释放），默认 `USER_MESSAGE_QUEUE_TIMEOUT_MS=5000` 已足够。如仍有超时，检查网络延迟或禁用此功能（`USER_MESSAGE_QUEUE_ENABLED=false`）
 
 ### 调试工具
 
@@ -510,6 +516,9 @@ npm run setup  # 自动生成密钥并创建管理员账户
   - `concurrency:{accountId}` - Redis Sorted Set实现的并发计数
 - **Webhook配置**:
   - `webhook_config:{id}` - Webhook配置
+- **用户消息队列**:
+  - `user_msg_queue_lock:{accountId}` - 用户消息队列锁（当前持有者requestId）
+  - `user_msg_queue_last:{accountId}` - 上次请求完成时间戳（用于延迟计算）
 - **系统信息**:
   - `system_info` - 系统状态缓存
   - `model_pricing` - 模型价格数据（pricingService）

VERSION

@@ -1 +1 @@
-1.1.226
+1.1.232

config/config.example.js

@@ -203,6 +203,15 @@ const config = {
   development: {
     debug: process.env.DEBUG === 'true',
     hotReload: process.env.HOT_RELOAD === 'true'
+  },
+
+  // 📬 用户消息队列配置
+  // 优化说明：锁在请求发送成功后立即释放（而非请求完成后），因为 Claude API 限流基于请求发送时刻计算
+  userMessageQueue: {
+    enabled: process.env.USER_MESSAGE_QUEUE_ENABLED === 'true', // 默认关闭
+    delayMs: parseInt(process.env.USER_MESSAGE_QUEUE_DELAY_MS) || 200, // 请求间隔（毫秒）
+    timeoutMs: parseInt(process.env.USER_MESSAGE_QUEUE_TIMEOUT_MS) || 5000, // 队列等待超时（毫秒），锁持有时间短，无需长等待
+    lockTtlMs: parseInt(process.env.USER_MESSAGE_QUEUE_LOCK_TTL_MS) || 5000 // 锁TTL（毫秒），5秒足以覆盖请求发送
   }
 }
 

src/app.js

@@ -625,6 +625,14 @@ class Application {
     }, 60000) // 每分钟执行一次
 
     logger.info('🔢 Concurrency cleanup task started (running every 1 minute)')
+
+    // 📬 启动用户消息队列服务
+    const userMessageQueueService = require('./services/userMessageQueueService')
+    // 先清理服务重启后残留的锁，防止旧锁阻塞新请求
+    userMessageQueueService.cleanupStaleLocks().then(() => {
+      // 然后启动定时清理任务
+      userMessageQueueService.startCleanupTask()
+    })
   }
 
   setupGracefulShutdown() {
@@ -661,6 +669,15 @@ class Application {
             logger.error('❌ Error stopping rate limit cleanup service:', error)
           }
 
+          // 停止用户消息队列清理服务
+          try {
+            const userMessageQueueService = require('./services/userMessageQueueService')
+            userMessageQueueService.stopCleanupTask()
+            logger.info('📬 User message queue service stopped')
+          } catch (error) {
+            logger.error('❌ Error stopping user message queue service:', error)
+          }
+
           // 停止费用排序索引服务
           try {
             const costRankService = require('./services/costRankService')

src/middleware/auth.js

@@ -6,6 +6,8 @@ const logger = require('../utils/logger')
 const redis = require('../models/redis')
 // const { RateLimiterRedis } = require('rate-limiter-flexible') // 暂时未使用
 const ClientValidator = require('../validators/clientValidator')
+const ClaudeCodeValidator = require('../validators/clients/claudeCodeValidator')
+const claudeRelayConfigService = require('../services/claudeRelayConfigService')
 
 const FALLBACK_CONCURRENCY_CONFIG = {
   leaseSeconds: 300,
@@ -201,6 +203,53 @@ const authenticateApiKey = async (req, res, next) => {
       )
     }
 
+    // 🔒 检查全局 Claude Code 限制（与 API Key 级别是 OR 逻辑）
+    // 仅对 Claude 服务端点生效 (/api/v1/messages 和 /claude/v1/messages)
+    if (!skipKeyRestrictions) {
+      const normalizedPath = (req.originalUrl || req.path || '').toLowerCase()
+      const isClaudeMessagesEndpoint =
+        normalizedPath.includes('/v1/messages') &&
+        (normalizedPath.startsWith('/api') || normalizedPath.startsWith('/claude'))
+
+      if (isClaudeMessagesEndpoint) {
+        try {
+          const globalClaudeCodeOnly = await claudeRelayConfigService.isClaudeCodeOnlyEnabled()
+
+          // API Key 级别的 Claude Code 限制
+          const keyClaudeCodeOnly =
+            validation.keyData.enableClientRestriction &&
+            Array.isArray(validation.keyData.allowedClients) &&
+            validation.keyData.allowedClients.length === 1 &&
+            validation.keyData.allowedClients.includes('claude_code')
+
+          // OR 逻辑：全局开启 或 API Key 级别限制为仅 claude_code
+          if (globalClaudeCodeOnly || keyClaudeCodeOnly) {
+            const isClaudeCode = ClaudeCodeValidator.validate(req)
+
+            if (!isClaudeCode) {
+              const clientIP = req.ip || req.connection?.remoteAddress || 'unknown'
+              logger.api(
+                `❌ Claude Code client validation failed (global: ${globalClaudeCodeOnly}, key: ${keyClaudeCodeOnly}) from ${clientIP}`
+              )
+              return res.status(403).json({
+                error: {
+                  type: 'client_validation_error',
+                  message: 'This endpoint only accepts requests from Claude Code CLI'
+                }
+              })
+            }
+
+            logger.api(
+              `✅ Claude Code client validated (global: ${globalClaudeCodeOnly}, key: ${keyClaudeCodeOnly})`
+            )
+          }
+        } catch (error) {
+          logger.error('❌ Error checking Claude Code restriction:', error)
+          // 配置服务出错时不阻断请求
+        }
+      }
+    }
+
     // 检查并发限制
     const concurrencyLimit = validation.keyData.concurrencyLimit || 0
     if (!skipKeyRestrictions && concurrencyLimit > 0) {

src/models/redis.js

@@ -2556,4 +2556,217 @@ redisClient.getDateStringInTimezone = getDateStringInTimezone
 redisClient.getHourInTimezone = getHourInTimezone
 redisClient.getWeekStringInTimezone = getWeekStringInTimezone
 
+// ============== 用户消息队列相关方法 ==============
+
+/**
+ * 尝试获取用户消息队列锁
+ * 使用 Lua 脚本保证原子性
+ * @param {string} accountId - 账户ID
+ * @param {string} requestId - 请求ID
+ * @param {number} lockTtlMs - 锁 TTL（毫秒）
+ * @param {number} delayMs - 请求间隔（毫秒）
+ * @returns {Promise<{acquired: boolean, waitMs: number}>}
+ *   - acquired: 是否成功获取锁
+ *   - waitMs: 需要等待的毫秒数（-1表示被占用需等待，>=0表示需要延迟的毫秒数）
+ */
+redisClient.acquireUserMessageLock = async function (accountId, requestId, lockTtlMs, delayMs) {
+  const lockKey = `user_msg_queue_lock:${accountId}`
+  const lastTimeKey = `user_msg_queue_last:${accountId}`
+
+  const script = `
+    local lockKey = KEYS[1]
+    local lastTimeKey = KEYS[2]
+    local requestId = ARGV[1]
+    local lockTtl = tonumber(ARGV[2])
+    local delayMs = tonumber(ARGV[3])
+
+    -- 检查锁是否空闲
+    local currentLock = redis.call('GET', lockKey)
+    if currentLock == false then
+      -- 检查是否需要延迟
+      local lastTime = redis.call('GET', lastTimeKey)
+      local now = redis.call('TIME')
+      local nowMs = tonumber(now[1]) * 1000 + math.floor(tonumber(now[2]) / 1000)
+
+      if lastTime then
+        local elapsed = nowMs - tonumber(lastTime)
+        if elapsed < delayMs then
+          -- 需要等待的毫秒数
+          return {0, delayMs - elapsed}
+        end
+      end
+
+      -- 获取锁
+      redis.call('SET', lockKey, requestId, 'PX', lockTtl)
+      return {1, 0}
+    end
+
+    -- 锁被占用，返回等待
+    return {0, -1}
+  `
+
+  try {
+    const result = await this.client.eval(
+      script,
+      2,
+      lockKey,
+      lastTimeKey,
+      requestId,
+      lockTtlMs,
+      delayMs
+    )
+    return {
+      acquired: result[0] === 1,
+      waitMs: result[1]
+    }
+  } catch (error) {
+    logger.error(`Failed to acquire user message lock for account ${accountId}:`, error)
+    // 返回 redisError 标记，让上层能区分 Redis 故障和正常锁占用
+    return { acquired: false, waitMs: -1, redisError: true, errorMessage: error.message }
+  }
+}
+
+/**
+ * 释放用户消息队列锁并记录完成时间
+ * @param {string} accountId - 账户ID
+ * @param {string} requestId - 请求ID
+ * @returns {Promise<boolean>} 是否成功释放
+ */
+redisClient.releaseUserMessageLock = async function (accountId, requestId) {
+  const lockKey = `user_msg_queue_lock:${accountId}`
+  const lastTimeKey = `user_msg_queue_last:${accountId}`
+
+  const script = `
+    local lockKey = KEYS[1]
+    local lastTimeKey = KEYS[2]
+    local requestId = ARGV[1]
+
+    -- 验证锁持有者
+    local currentLock = redis.call('GET', lockKey)
+    if currentLock == requestId then
+      -- 记录完成时间
+      local now = redis.call('TIME')
+      local nowMs = tonumber(now[1]) * 1000 + math.floor(tonumber(now[2]) / 1000)
+      redis.call('SET', lastTimeKey, nowMs, 'EX', 60)  -- 60秒后过期
+
+      -- 删除锁
+      redis.call('DEL', lockKey)
+      return 1
+    end
+    return 0
+  `
+
+  try {
+    const result = await this.client.eval(script, 2, lockKey, lastTimeKey, requestId)
+    return result === 1
+  } catch (error) {
+    logger.error(`Failed to release user message lock for account ${accountId}:`, error)
+    return false
+  }
+}
+
+/**
+ * 强制释放用户消息队列锁（用于清理孤儿锁）
+ * @param {string} accountId - 账户ID
+ * @returns {Promise<boolean>} 是否成功释放
+ */
+redisClient.forceReleaseUserMessageLock = async function (accountId) {
+  const lockKey = `user_msg_queue_lock:${accountId}`
+
+  try {
+    await this.client.del(lockKey)
+    return true
+  } catch (error) {
+    logger.error(`Failed to force release user message lock for account ${accountId}:`, error)
+    return false
+  }
+}
+
+/**
+ * 获取用户消息队列统计信息（用于调试）
+ * @param {string} accountId - 账户ID
+ * @returns {Promise<Object>} 队列统计
+ */
+redisClient.getUserMessageQueueStats = async function (accountId) {
+  const lockKey = `user_msg_queue_lock:${accountId}`
+  const lastTimeKey = `user_msg_queue_last:${accountId}`
+
+  try {
+    const [lockHolder, lastTime, lockTtl] = await Promise.all([
+      this.client.get(lockKey),
+      this.client.get(lastTimeKey),
+      this.client.pttl(lockKey)
+    ])
+
+    return {
+      accountId,
+      isLocked: !!lockHolder,
+      lockHolder,
+      lockTtlMs: lockTtl > 0 ? lockTtl : 0,
+      lockTtlRaw: lockTtl, // 原始 PTTL 值：>0 有TTL，-1 无过期时间，-2 键不存在
+      lastCompletedAt: lastTime ? new Date(parseInt(lastTime)).toISOString() : null
+    }
+  } catch (error) {
+    logger.error(`Failed to get user message queue stats for account ${accountId}:`, error)
+    return {
+      accountId,
+      isLocked: false,
+      lockHolder: null,
+      lockTtlMs: 0,
+      lockTtlRaw: -2,
+      lastCompletedAt: null
+    }
+  }
+}
+
+/**
+ * 扫描所有用户消息队列锁（用于清理任务）
+ * @returns {Promise<string[]>} 账户ID列表
+ */
+redisClient.scanUserMessageQueueLocks = async function () {
+  const accountIds = []
+  let cursor = '0'
+  let iterations = 0
+  const MAX_ITERATIONS = 1000 // 防止无限循环
+
+  try {
+    do {
+      const [newCursor, keys] = await this.client.scan(
+        cursor,
+        'MATCH',
+        'user_msg_queue_lock:*',
+        'COUNT',
+        100
+      )
+      cursor = newCursor
+      iterations++
+
+      for (const key of keys) {
+        const accountId = key.replace('user_msg_queue_lock:', '')
+        accountIds.push(accountId)
+      }
+
+      // 防止无限循环
+      if (iterations >= MAX_ITERATIONS) {
+        logger.warn(
+          `📬 User message queue: SCAN reached max iterations (${MAX_ITERATIONS}), stopping early`,
+          { foundLocks: accountIds.length }
+        )
+        break
+      }
+    } while (cursor !== '0')
+
+    if (accountIds.length > 0) {
+      logger.debug(
+        `📬 User message queue: scanned ${accountIds.length} lock(s) in ${iterations} iteration(s)`
+      )
+    }
+
+    return accountIds
+  } catch (error) {
+    logger.error('Failed to scan user message queue locks:', error)
+    return []
+  }
+}
+
 module.exports = redisClient

src/routes/admin/apiKeys.js

@@ -919,6 +919,62 @@ async function calculateKeyStats(keyId, timeRange, startDate, endDate) {
   // 去重（避免日数据和月数据重复计算）
   const uniqueKeys = [...new Set(allKeys)]
 
+  // 获取实时限制数据（窗口数据不受时间范围筛选影响，始终获取当前窗口状态）
+  let dailyCost = 0
+  let currentWindowCost = 0
+  let windowRemainingSeconds = null
+  let windowStartTime = null
+  let windowEndTime = null
+  let allTimeCost = 0
+
+  try {
+    // 先获取 API Key 配置，判断是否需要查询限制相关数据
+    const apiKey = await redis.getApiKey(keyId)
+    const rateLimitWindow = parseInt(apiKey?.rateLimitWindow) || 0
+    const dailyCostLimit = parseFloat(apiKey?.dailyCostLimit) || 0
+    const totalCostLimit = parseFloat(apiKey?.totalCostLimit) || 0
+
+    // 只在启用了每日费用限制时查询
+    if (dailyCostLimit > 0) {
+      dailyCost = await redis.getDailyCost(keyId)
+    }
+
+    // 只在启用了总费用限制时查询
+    if (totalCostLimit > 0) {
+      const totalCostKey = `usage:cost:total:${keyId}`
+      allTimeCost = parseFloat((await client.get(totalCostKey)) || '0')
+    }
+
+    // 只在启用了窗口限制时查询窗口数据
+    if (rateLimitWindow > 0) {
+      const costCountKey = `rate_limit:cost:${keyId}`
+      const windowStartKey = `rate_limit:window_start:${keyId}`
+
+      currentWindowCost = parseFloat((await client.get(costCountKey)) || '0')
+
+      // 获取窗口开始时间和计算剩余时间
+      const windowStart = await client.get(windowStartKey)
+      if (windowStart) {
+        const now = Date.now()
+        windowStartTime = parseInt(windowStart)
+        const windowDuration = rateLimitWindow * 60 * 1000 // 转换为毫秒
+        windowEndTime = windowStartTime + windowDuration
+
+        // 如果窗口还有效
+        if (now < windowEndTime) {
+          windowRemainingSeconds = Math.max(0, Math.floor((windowEndTime - now) / 1000))
+        } else {
+          // 窗口已过期
+          windowRemainingSeconds = 0
+          currentWindowCost = 0
+        }
+      }
+    }
+  } catch (error) {
+    logger.warn(`⚠️ 获取实时限制数据失败 (key: ${keyId}):`, error.message)
+  }
+
+  // 如果没有使用数据，返回零值但包含窗口数据
   if (uniqueKeys.length === 0) {
     return {
       requests: 0,
@@ -928,7 +984,14 @@ async function calculateKeyStats(keyId, timeRange, startDate, endDate) {
       cacheCreateTokens: 0,
       cacheReadTokens: 0,
       cost: 0,
-      formattedCost: '$0.00'
+      formattedCost: '$0.00',
+      // 实时限制数据（始终返回，不受时间范围影响）
+      dailyCost,
+      currentWindowCost,
+      windowRemainingSeconds,
+      windowStartTime,
+      windowEndTime,
+      allTimeCost
     }
   }
 
@@ -1029,55 +1092,6 @@ async function calculateKeyStats(keyId, timeRange, startDate, endDate) {
 
   const tokens = inputTokens + outputTokens + cacheCreateTokens + cacheReadTokens
 
-  // 获取实时限制数据
-  let dailyCost = 0
-  let currentWindowCost = 0
-  let windowRemainingSeconds = null
-  let windowStartTime = null
-  let windowEndTime = null
-  let allTimeCost = 0
-
-  try {
-    // 获取当日费用
-    dailyCost = await redis.getDailyCost(keyId)
-
-    // 获取历史总费用（用于总费用限制进度条，不受时间范围影响）
-    const totalCostKey = `usage:cost:total:${keyId}`
-    allTimeCost = parseFloat((await client.get(totalCostKey)) || '0')
-
-    // 获取 API Key 配置信息以判断是否需要窗口数据
-    const apiKey = await redis.getApiKey(keyId)
-    // 显式转换为整数，与 apiStats.js 保持一致，避免字符串比较问题
-    const rateLimitWindow = parseInt(apiKey?.rateLimitWindow) || 0
-
-    if (rateLimitWindow > 0) {
-      const costCountKey = `rate_limit:cost:${keyId}`
-      const windowStartKey = `rate_limit:window_start:${keyId}`
-
-      currentWindowCost = parseFloat((await client.get(costCountKey)) || '0')
-
-      // 获取窗口开始时间和计算剩余时间
-      const windowStart = await client.get(windowStartKey)
-      if (windowStart) {
-        const now = Date.now()
-        windowStartTime = parseInt(windowStart)
-        const windowDuration = rateLimitWindow * 60 * 1000 // 转换为毫秒
-        windowEndTime = windowStartTime + windowDuration
-
-        // 如果窗口还有效
-        if (now < windowEndTime) {
-          windowRemainingSeconds = Math.max(0, Math.floor((windowEndTime - now) / 1000))
-        } else {
-          // 窗口已过期
-          windowRemainingSeconds = 0
-          currentWindowCost = 0
-        }
-      }
-    }
-  } catch (error) {
-    logger.warn(`⚠️ 获取实时限制数据失败 (key: ${keyId}):`, error.message)
-  }
-
   return {
     requests: totalRequests,
     tokens,

src/routes/admin/claudeRelayConfig.js

@@ -0,0 +1,175 @@
+/**
+ * Claude 转发配置 API 路由
+ * 管理全局 Claude Code 限制和会话绑定配置
+ */
+
+const express = require('express')
+const { authenticateAdmin } = require('../../middleware/auth')
+const claudeRelayConfigService = require('../../services/claudeRelayConfigService')
+const logger = require('../../utils/logger')
+
+const router = express.Router()
+
+/**
+ * GET /admin/claude-relay-config
+ * 获取 Claude 转发配置
+ */
+router.get('/claude-relay-config', authenticateAdmin, async (req, res) => {
+  try {
+    const config = await claudeRelayConfigService.getConfig()
+    return res.json({
+      success: true,
+      config
+    })
+  } catch (error) {
+    logger.error('❌ Failed to get Claude relay config:', error)
+    return res.status(500).json({
+      error: 'Failed to get configuration',
+      message: error.message
+    })
+  }
+})
+
+/**
+ * PUT /admin/claude-relay-config
+ * 更新 Claude 转发配置
+ */
+router.put('/claude-relay-config', authenticateAdmin, async (req, res) => {
+  try {
+    const {
+      claudeCodeOnlyEnabled,
+      globalSessionBindingEnabled,
+      sessionBindingErrorMessage,
+      sessionBindingTtlDays,
+      userMessageQueueEnabled,
+      userMessageQueueDelayMs,
+      userMessageQueueTimeoutMs
+    } = req.body
+
+    // 验证输入
+    if (claudeCodeOnlyEnabled !== undefined && typeof claudeCodeOnlyEnabled !== 'boolean') {
+      return res.status(400).json({ error: 'claudeCodeOnlyEnabled must be a boolean' })
+    }
+
+    if (
+      globalSessionBindingEnabled !== undefined &&
+      typeof globalSessionBindingEnabled !== 'boolean'
+    ) {
+      return res.status(400).json({ error: 'globalSessionBindingEnabled must be a boolean' })
+    }
+
+    if (sessionBindingErrorMessage !== undefined) {
+      if (typeof sessionBindingErrorMessage !== 'string') {
+        return res.status(400).json({ error: 'sessionBindingErrorMessage must be a string' })
+      }
+      if (sessionBindingErrorMessage.length > 500) {
+        return res
+          .status(400)
+          .json({ error: 'sessionBindingErrorMessage must be less than 500 characters' })
+      }
+    }
+
+    if (sessionBindingTtlDays !== undefined) {
+      if (
+        typeof sessionBindingTtlDays !== 'number' ||
+        sessionBindingTtlDays < 1 ||
+        sessionBindingTtlDays > 365
+      ) {
+        return res
+          .status(400)
+          .json({ error: 'sessionBindingTtlDays must be a number between 1 and 365' })
+      }
+    }
+
+    // 验证用户消息队列配置
+    if (userMessageQueueEnabled !== undefined && typeof userMessageQueueEnabled !== 'boolean') {
+      return res.status(400).json({ error: 'userMessageQueueEnabled must be a boolean' })
+    }
+
+    if (userMessageQueueDelayMs !== undefined) {
+      if (
+        typeof userMessageQueueDelayMs !== 'number' ||
+        userMessageQueueDelayMs < 0 ||
+        userMessageQueueDelayMs > 10000
+      ) {
+        return res
+          .status(400)
+          .json({ error: 'userMessageQueueDelayMs must be a number between 0 and 10000' })
+      }
+    }
+
+    if (userMessageQueueTimeoutMs !== undefined) {
+      if (
+        typeof userMessageQueueTimeoutMs !== 'number' ||
+        userMessageQueueTimeoutMs < 1000 ||
+        userMessageQueueTimeoutMs > 300000
+      ) {
+        return res
+          .status(400)
+          .json({ error: 'userMessageQueueTimeoutMs must be a number between 1000 and 300000' })
+      }
+    }
+
+    const updateData = {}
+    if (claudeCodeOnlyEnabled !== undefined) {
+      updateData.claudeCodeOnlyEnabled = claudeCodeOnlyEnabled
+    }
+    if (globalSessionBindingEnabled !== undefined) {
+      updateData.globalSessionBindingEnabled = globalSessionBindingEnabled
+    }
+    if (sessionBindingErrorMessage !== undefined) {
+      updateData.sessionBindingErrorMessage = sessionBindingErrorMessage
+    }
+    if (sessionBindingTtlDays !== undefined) {
+      updateData.sessionBindingTtlDays = sessionBindingTtlDays
+    }
+    if (userMessageQueueEnabled !== undefined) {
+      updateData.userMessageQueueEnabled = userMessageQueueEnabled
+    }
+    if (userMessageQueueDelayMs !== undefined) {
+      updateData.userMessageQueueDelayMs = userMessageQueueDelayMs
+    }
+    if (userMessageQueueTimeoutMs !== undefined) {
+      updateData.userMessageQueueTimeoutMs = userMessageQueueTimeoutMs
+    }
+
+    const updatedConfig = await claudeRelayConfigService.updateConfig(
+      updateData,
+      req.admin?.username || 'unknown'
+    )
+
+    return res.json({
+      success: true,
+      message: 'Configuration updated successfully',
+      config: updatedConfig
+    })
+  } catch (error) {
+    logger.error('❌ Failed to update Claude relay config:', error)
+    return res.status(500).json({
+      error: 'Failed to update configuration',
+      message: error.message
+    })
+  }
+})
+
+/**
+ * GET /admin/claude-relay-config/session-bindings
+ * 获取会话绑定统计
+ */
+router.get('/claude-relay-config/session-bindings', authenticateAdmin, async (req, res) => {
+  try {
+    const stats = await claudeRelayConfigService.getSessionBindingStats()
+    return res.json({
+      success: true,
+      data: stats
+    })
+  } catch (error) {
+    logger.error('❌ Failed to get session binding stats:', error)
+    return res.status(500).json({
+      error: 'Failed to get session binding statistics',
+      message: error.message
+    })
+  }
+})
+
+module.exports = router

src/routes/admin/concurrency.js

@@ -7,12 +7,13 @@ const express = require('express')
 const router = express.Router()
 const redis = require('../../models/redis')
 const logger = require('../../utils/logger')
+const { authenticateAdmin } = require('../../middleware/auth')
 
 /**
  * GET /admin/concurrency
  * 获取所有并发状态
  */
-router.get('/concurrency', async (req, res) => {
+router.get('/concurrency', authenticateAdmin, async (req, res) => {
   try {
     const status = await redis.getAllConcurrencyStatus()
 
@@ -42,7 +43,7 @@ router.get('/concurrency', async (req, res) => {
  * GET /admin/concurrency/:apiKeyId
  * 获取特定 API Key 的并发状态详情
  */
-router.get('/concurrency/:apiKeyId', async (req, res) => {
+router.get('/concurrency/:apiKeyId', authenticateAdmin, async (req, res) => {
   try {
     const { apiKeyId } = req.params
     const status = await redis.getConcurrencyStatus(apiKeyId)
@@ -65,7 +66,7 @@ router.get('/concurrency/:apiKeyId', async (req, res) => {
  * DELETE /admin/concurrency/:apiKeyId
  * 强制清理特定 API Key 的并发计数
  */
-router.delete('/concurrency/:apiKeyId', async (req, res) => {
+router.delete('/concurrency/:apiKeyId', authenticateAdmin, async (req, res) => {
   try {
     const { apiKeyId } = req.params
     const result = await redis.forceClearConcurrency(apiKeyId)
@@ -93,7 +94,7 @@ router.delete('/concurrency/:apiKeyId', async (req, res) => {
  * DELETE /admin/concurrency
  * 强制清理所有并发计数
  */
-router.delete('/concurrency', async (req, res) => {
+router.delete('/concurrency', authenticateAdmin, async (req, res) => {
   try {
     const result = await redis.forceClearAllConcurrency()
 
@@ -118,7 +119,7 @@ router.delete('/concurrency', async (req, res) => {
  * POST /admin/concurrency/cleanup
  * 清理过期的并发条目（不影响活跃请求）
  */
-router.post('/concurrency/cleanup', async (req, res) => {
+router.post('/concurrency/cleanup', authenticateAdmin, async (req, res) => {
   try {
     const { apiKeyId } = req.body
     const result = await redis.cleanupExpiredConcurrency(apiKeyId || null)

src/routes/admin/index.js

@@ -23,6 +23,7 @@ const dashboardRoutes = require('./dashboard')
 const usageStatsRoutes = require('./usageStats')
 const systemRoutes = require('./system')
 const concurrencyRoutes = require('./concurrency')
+const claudeRelayConfigRoutes = require('./claudeRelayConfig')
 
 // 挂载所有子路由
 // 使用完整路径的模块（直接挂载到根路径）
@@ -37,6 +38,7 @@ router.use('/', dashboardRoutes)
 router.use('/', usageStatsRoutes)
 router.use('/', systemRoutes)
 router.use('/', concurrencyRoutes)
+router.use('/', claudeRelayConfigRoutes)
 
 // 使用相对路径的模块（需要指定基础路径前缀）
 router.use('/account-groups', accountGroupsRoutes)

src/routes/api.js

@@ -11,6 +11,7 @@ const logger = require('../utils/logger')
 const { getEffectiveModel, parseVendorPrefixedModel } = require('../utils/modelHelper')
 const sessionHelper = require('../utils/sessionHelper')
 const { updateRateLimitCounters } = require('../utils/rateLimitHelper')
+const claudeRelayConfigService = require('../services/claudeRelayConfigService')
 const { sanitizeUpstreamError } = require('../utils/errorSanitizer')
 const router = express.Router()
 
@@ -37,6 +38,73 @@ function queueRateLimitUpdate(rateLimitInfo, usageSummary, model, context = '')
     })
 }
 
+/**
+ * 判断是否为旧会话（污染的会话）
+ * Claude Code 发送的请求特点：
+ * - messages 数组通常只有 1 个元素
+ * - 历史对话记录嵌套在单个 message 的 content 数组中
+ * - content 数组中包含 <system-reminder> 开头的系统注入内容
+ *
+ * 污染会话的特征：
+ * 1. messages.length > 1
+ * 2. messages.length === 1 但 content 中有多个用户输入
+ * 3. "warmup" 请求：单条简单消息 + 无 tools（真正新会话会带 tools）
+ *
+ * @param {Object} body - 请求体
+ * @returns {boolean} 是否为旧会话
+ */
+function isOldSession(body) {
+  const messages = body?.messages
+  const tools = body?.tools
+
+  if (!messages || messages.length === 0) {
+    return false
+  }
+
+  // 1. 多条消息 = 旧会话
+  if (messages.length > 1) {
+    return true
+  }
+
+  // 2. 单条消息，分析 content
+  const firstMessage = messages[0]
+  const content = firstMessage?.content
+
+  if (!content) {
+    return false
+  }
+
+  // 如果 content 是字符串，只有一条输入，需要检查 tools
+  if (typeof content === 'string') {
+    // 有 tools = 正常新会话，无 tools = 可疑
+    return !tools || tools.length === 0
+  }
+
+  // 如果 content 是数组，统计非 system-reminder 的元素
+  if (Array.isArray(content)) {
+    const userInputs = content.filter((item) => {
+      if (item.type !== 'text') {
+        return false
+      }
+      const text = item.text || ''
+      // 剔除以 <system-reminder> 开头的
+      return !text.trimStart().startsWith('<system-reminder>')
+    })
+
+    // 多个用户输入 = 旧会话
+    if (userInputs.length > 1) {
+      return true
+    }
+
+    // Warmup 检测：单个消息 + 无 tools = 旧会话
+    if (userInputs.length === 1 && (!tools || tools.length === 0)) {
+      return true
+    }
+  }
+
+  return false
+}
+
 // 🔧 共享的消息处理函数
 async function handleMessagesRequest(req, res) {
   try {
@@ -141,6 +209,56 @@ async function handleMessagesRequest(req, res) {
       // 生成会话哈希用于sticky会话
       const sessionHash = sessionHelper.generateSessionHash(req.body)
 
+      // 🔒 全局会话绑定验证
+      let forcedAccount = null
+      let needSessionBinding = false
+      let originalSessionIdForBinding = null
+
+      try {
+        const globalBindingEnabled = await claudeRelayConfigService.isGlobalSessionBindingEnabled()
+
+        if (globalBindingEnabled) {
+          const originalSessionId = claudeRelayConfigService.extractOriginalSessionId(req.body)
+
+          if (originalSessionId) {
+            const validation = await claudeRelayConfigService.validateNewSession(
+              req.body,
+              originalSessionId
+            )
+
+            if (!validation.valid) {
+              logger.api(
+                `❌ Session binding validation failed: ${validation.code} for session ${originalSessionId}`
+              )
+              return res.status(403).json({
+                error: {
+                  type: 'session_binding_error',
+                  message: validation.error
+                }
+              })
+            }
+
+            // 如果已有绑定，使用绑定的账户
+            if (validation.binding) {
+              forcedAccount = validation.binding
+              logger.api(
+                `🔗 Using bound account for session ${originalSessionId}: ${forcedAccount.accountId}`
+              )
+            }
+
+            // 标记需要在调度成功后建立绑定
+            if (validation.isNewSession) {
+              needSessionBinding = true
+              originalSessionIdForBinding = originalSessionId
+              logger.api(`📝 New session detected, will create binding: ${originalSessionId}`)
+            }
+          }
+        }
+      } catch (error) {
+        logger.error('❌ Error in global session binding check:', error)
+        // 配置服务出错时不阻断请求
+      }
+
       // 使用统一调度选择账号（传递请求的模型）
       const requestedModel = req.body.model
       let accountId
@@ -149,10 +267,21 @@ async function handleMessagesRequest(req, res) {
         const selection = await unifiedClaudeScheduler.selectAccountForApiKey(
           req.apiKey,
           sessionHash,
-          requestedModel
+          requestedModel,
+          forcedAccount
         )
         ;({ accountId, accountType } = selection)
       } catch (error) {
+        // 处理会话绑定账户不可用的错误
+        if (error.code === 'SESSION_BINDING_ACCOUNT_UNAVAILABLE') {
+          const errorMessage = await claudeRelayConfigService.getSessionBindingErrorMessage()
+          return res.status(403).json({
+            error: {
+              type: 'session_binding_error',
+              message: errorMessage
+            }
+          })
+        }
         if (error.code === 'CLAUDE_DEDICATED_RATE_LIMITED') {
           const limitMessage = claudeRelayService._buildStandardRateLimitMessage(
             error.rateLimitEndAt
@@ -170,6 +299,40 @@ async function handleMessagesRequest(req, res) {
         throw error
       }
 
+      // 🔗 在成功调度后建立会话绑定（仅 claude-official 类型）
+      // claude-official 只接受：1) 新会话 2) 已绑定的会话
+      if (
+        needSessionBinding &&
+        originalSessionIdForBinding &&
+        accountId &&
+        accountType === 'claude-official'
+      ) {
+        // 🚫 检测旧会话（污染的会话）
+        if (isOldSession(req.body)) {
+          const cfg = await claudeRelayConfigService.getConfig()
+          logger.warn(
+            `🚫 Old session rejected: sessionId=${originalSessionIdForBinding}, messages.length=${req.body?.messages?.length}, tools.length=${req.body?.tools?.length || 0}, isOldSession=true`
+          )
+          return res.status(400).json({
+            error: {
+              type: 'session_binding_error',
+              message: cfg.sessionBindingErrorMessage || '你的本地session已污染，请清理后使用。'
+            }
+          })
+        }
+
+        // 创建绑定
+        try {
+          await claudeRelayConfigService.setOriginalSessionBinding(
+            originalSessionIdForBinding,
+            accountId,
+            accountType
+          )
+        } catch (bindingError) {
+          logger.warn(`⚠️ Failed to create session binding:`, bindingError)
+        }
+      }
+
       // 根据账号类型选择对应的转发服务并调用
       if (accountType === 'claude-official') {
         // 官方Claude账号使用原有的转发服务（会自己选择账号）
@@ -503,6 +666,55 @@ async function handleMessagesRequest(req, res) {
       // 生成会话哈希用于sticky会话
       const sessionHash = sessionHelper.generateSessionHash(req.body)
 
+      // 🔒 全局会话绑定验证（非流式）
+      let forcedAccountNonStream = null
+      let needSessionBindingNonStream = false
+      let originalSessionIdForBindingNonStream = null
+
+      try {
+        const globalBindingEnabled = await claudeRelayConfigService.isGlobalSessionBindingEnabled()
+
+        if (globalBindingEnabled) {
+          const originalSessionId = claudeRelayConfigService.extractOriginalSessionId(req.body)
+
+          if (originalSessionId) {
+            const validation = await claudeRelayConfigService.validateNewSession(
+              req.body,
+              originalSessionId
+            )
+
+            if (!validation.valid) {
+              logger.api(
+                `❌ Session binding validation failed (non-stream): ${validation.code} for session ${originalSessionId}`
+              )
+              return res.status(403).json({
+                error: {
+                  type: 'session_binding_error',
+                  message: validation.error
+                }
+              })
+            }
+
+            if (validation.binding) {
+              forcedAccountNonStream = validation.binding
+              logger.api(
+                `🔗 Using bound account for session (non-stream) ${originalSessionId}: ${forcedAccountNonStream.accountId}`
+              )
+            }
+
+            if (validation.isNewSession) {
+              needSessionBindingNonStream = true
+              originalSessionIdForBindingNonStream = originalSessionId
+              logger.api(
+                `📝 New session detected (non-stream), will create binding: ${originalSessionId}`
+              )
+            }
+          }
+        }
+      } catch (error) {
+        logger.error('❌ Error in global session binding check (non-stream):', error)
+      }
+
       // 使用统一调度选择账号（传递请求的模型）
       const requestedModel = req.body.model
       let accountId
@@ -511,10 +723,20 @@ async function handleMessagesRequest(req, res) {
         const selection = await unifiedClaudeScheduler.selectAccountForApiKey(
           req.apiKey,
           sessionHash,
-          requestedModel
+          requestedModel,
+          forcedAccountNonStream
         )
         ;({ accountId, accountType } = selection)
       } catch (error) {
+        if (error.code === 'SESSION_BINDING_ACCOUNT_UNAVAILABLE') {
+          const errorMessage = await claudeRelayConfigService.getSessionBindingErrorMessage()
+          return res.status(403).json({
+            error: {
+              type: 'session_binding_error',
+              message: errorMessage
+            }
+          })
+        }
         if (error.code === 'CLAUDE_DEDICATED_RATE_LIMITED') {
           const limitMessage = claudeRelayService._buildStandardRateLimitMessage(
             error.rateLimitEndAt
@@ -527,6 +749,40 @@ async function handleMessagesRequest(req, res) {
         throw error
       }
 
+      // 🔗 在成功调度后建立会话绑定（非流式，仅 claude-official 类型）
+      // claude-official 只接受：1) 新会话 2) 已绑定的会话
+      if (
+        needSessionBindingNonStream &&
+        originalSessionIdForBindingNonStream &&
+        accountId &&
+        accountType === 'claude-official'
+      ) {
+        // 🚫 检测旧会话（污染的会话）
+        if (isOldSession(req.body)) {
+          const cfg = await claudeRelayConfigService.getConfig()
+          logger.warn(
+            `🚫 Old session rejected (non-stream): sessionId=${originalSessionIdForBindingNonStream}, messages.length=${req.body?.messages?.length}, tools.length=${req.body?.tools?.length || 0}, isOldSession=true`
+          )
+          return res.status(400).json({
+            error: {
+              type: 'session_binding_error',
+              message: cfg.sessionBindingErrorMessage || '你的本地session已污染，请清理后使用。'
+            }
+          })
+        }
+
+        // 创建绑定
+        try {
+          await claudeRelayConfigService.setOriginalSessionBinding(
+            originalSessionIdForBindingNonStream,
+            accountId,
+            accountType
+          )
+        } catch (bindingError) {
+          logger.warn(`⚠️ Failed to create session binding (non-stream):`, bindingError)
+        }
+      }
+
       // 根据账号类型选择对应的转发服务
       let response
       logger.debug(`[DEBUG] Request query params: ${JSON.stringify(req.query)}`)
@@ -966,6 +1222,41 @@ router.post('/v1/messages/count_tokens', authenticateApiKey, async (req, res) =>
     })
   }
 
+  // 🔗 会话绑定验证（与 messages 端点保持一致）
+  const originalSessionId = claudeRelayConfigService.extractOriginalSessionId(req.body)
+  const sessionValidation = await claudeRelayConfigService.validateNewSession(
+    req.body,
+    originalSessionId
+  )
+
+  if (!sessionValidation.valid) {
+    logger.warn(
+      `🚫 Session binding validation failed (count_tokens): ${sessionValidation.code} for session ${originalSessionId}`
+    )
+    return res.status(400).json({
+      error: {
+        type: 'session_binding_error',
+        message: sessionValidation.error
+      }
+    })
+  }
+
+  // 🔗 检测旧会话（污染的会话）- 仅对需要绑定的新会话检查
+  if (sessionValidation.isNewSession && originalSessionId) {
+    if (isOldSession(req.body)) {
+      const cfg = await claudeRelayConfigService.getConfig()
+      logger.warn(
+        `🚫 Old session rejected (count_tokens): sessionId=${originalSessionId}, messages.length=${req.body?.messages?.length}, tools.length=${req.body?.tools?.length || 0}, isOldSession=true`
+      )
+      return res.status(400).json({
+        error: {
+          type: 'session_binding_error',
+          message: cfg.sessionBindingErrorMessage || '你的本地session已污染，请清理后使用。'
+        }
+      })
+    }
+  }
+
   logger.info(`🔢 Processing token count request for key: ${req.apiKey.name}`)
 
   const sessionHash = sessionHelper.generateSessionHash(req.body)

src/services/bedrockRelayService.js

@@ -6,6 +6,7 @@ const {
 const { fromEnv } = require('@aws-sdk/credential-providers')
 const logger = require('../utils/logger')
 const config = require('../../config/config')
+const userMessageQueueService = require('./userMessageQueueService')
 
 class BedrockRelayService {
   constructor() {
@@ -69,7 +70,68 @@ class BedrockRelayService {
 
   // 处理非流式请求
   async handleNonStreamRequest(requestBody, bedrockAccount = null) {
+    const accountId = bedrockAccount?.id
+    let queueLockAcquired = false
+    let queueRequestId = null
+
     try {
+      // 📬 用户消息队列处理
+      if (userMessageQueueService.isUserMessageRequest(requestBody)) {
+        // 校验 accountId 非空，避免空值污染队列锁键
+        if (!accountId || accountId === '') {
+          logger.error('❌ accountId missing for queue lock in Bedrock handleNonStreamRequest')
+          throw new Error('accountId missing for queue lock')
+        }
+        const queueResult = await userMessageQueueService.acquireQueueLock(accountId)
+        if (!queueResult.acquired && !queueResult.skipped) {
+          // 区分 Redis 后端错误和队列超时
+          const isBackendError = queueResult.error === 'queue_backend_error'
+          const errorCode = isBackendError ? 'QUEUE_BACKEND_ERROR' : 'QUEUE_TIMEOUT'
+          const errorType = isBackendError ? 'queue_backend_error' : 'queue_timeout'
+          const errorMessage = isBackendError
+            ? 'Queue service temporarily unavailable, please retry later'
+            : 'User message queue wait timeout, please retry later'
+          const statusCode = isBackendError ? 500 : 503
+
+          // 结构化性能日志，用于后续统计
+          logger.performance('user_message_queue_error', {
+            errorType,
+            errorCode,
+            accountId,
+            statusCode,
+            backendError: isBackendError ? queueResult.errorMessage : undefined
+          })
+
+          logger.warn(
+            `📬 User message queue ${errorType} for Bedrock account ${accountId}`,
+            isBackendError ? { backendError: queueResult.errorMessage } : {}
+          )
+          return {
+            statusCode,
+            headers: {
+              'Content-Type': 'application/json',
+              'x-user-message-queue-error': errorType
+            },
+            body: JSON.stringify({
+              type: 'error',
+              error: {
+                type: errorType,
+                code: errorCode,
+                message: errorMessage
+              }
+            }),
+            success: false
+          }
+        }
+        if (queueResult.acquired && !queueResult.skipped) {
+          queueLockAcquired = true
+          queueRequestId = queueResult.requestId
+          logger.debug(
+            `📬 User message queue lock acquired for Bedrock account ${accountId}, requestId: ${queueRequestId}`
+          )
+        }
+      }
+
       const modelId = this._selectModel(requestBody, bedrockAccount)
       const region = this._selectRegion(modelId, bedrockAccount)
       const client = this._getBedrockClient(region, bedrockAccount)
@@ -90,6 +152,23 @@ class BedrockRelayService {
       const response = await client.send(command)
       const duration = Date.now() - startTime
 
+      // 📬 请求已发送成功，立即释放队列锁（无需等待响应处理完成）
+      // 因为限流基于请求发送时刻计算（RPM），不是请求完成时刻
+      if (queueLockAcquired && queueRequestId && accountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+          queueLockAcquired = false // 标记已释放，防止 finally 重复释放
+          logger.debug(
+            `📬 User message queue lock released early for Bedrock account ${accountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock early for Bedrock account ${accountId}:`,
+            releaseError.message
+          )
+        }
+      }
+
       // 解析响应
       const responseBody = JSON.parse(new TextDecoder().decode(response.body))
       const claudeResponse = this._convertFromBedrockFormat(responseBody)
@@ -106,12 +185,93 @@ class BedrockRelayService {
     } catch (error) {
       logger.error('❌ Bedrock非流式请求失败:', error)
       throw this._handleBedrockError(error)
+    } finally {
+      // 📬 释放用户消息队列锁（兜底，正常情况下已在请求发送后提前释放）
+      if (queueLockAcquired && queueRequestId && accountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+          logger.debug(
+            `📬 User message queue lock released in finally for Bedrock account ${accountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock for Bedrock account ${accountId}:`,
+            releaseError.message
+          )
+        }
+      }
     }
   }
 
   // 处理流式请求
   async handleStreamRequest(requestBody, bedrockAccount = null, res) {
+    const accountId = bedrockAccount?.id
+    let queueLockAcquired = false
+    let queueRequestId = null
+
     try {
+      // 📬 用户消息队列处理
+      if (userMessageQueueService.isUserMessageRequest(requestBody)) {
+        // 校验 accountId 非空，避免空值污染队列锁键
+        if (!accountId || accountId === '') {
+          logger.error('❌ accountId missing for queue lock in Bedrock handleStreamRequest')
+          throw new Error('accountId missing for queue lock')
+        }
+        const queueResult = await userMessageQueueService.acquireQueueLock(accountId)
+        if (!queueResult.acquired && !queueResult.skipped) {
+          // 区分 Redis 后端错误和队列超时
+          const isBackendError = queueResult.error === 'queue_backend_error'
+          const errorCode = isBackendError ? 'QUEUE_BACKEND_ERROR' : 'QUEUE_TIMEOUT'
+          const errorType = isBackendError ? 'queue_backend_error' : 'queue_timeout'
+          const errorMessage = isBackendError
+            ? 'Queue service temporarily unavailable, please retry later'
+            : 'User message queue wait timeout, please retry later'
+          const statusCode = isBackendError ? 500 : 503
+
+          // 结构化性能日志，用于后续统计
+          logger.performance('user_message_queue_error', {
+            errorType,
+            errorCode,
+            accountId,
+            statusCode,
+            stream: true,
+            backendError: isBackendError ? queueResult.errorMessage : undefined
+          })
+
+          logger.warn(
+            `📬 User message queue ${errorType} for Bedrock account ${accountId} (stream)`,
+            isBackendError ? { backendError: queueResult.errorMessage } : {}
+          )
+          if (!res.headersSent) {
+            res.writeHead(statusCode, {
+              'Content-Type': 'text/event-stream',
+              'Cache-Control': 'no-cache',
+              Connection: 'keep-alive',
+              'x-user-message-queue-error': errorType
+            })
+          }
+          const errorEvent = `event: error\ndata: ${JSON.stringify({
+            type: 'error',
+            error: {
+              type: errorType,
+              code: errorCode,
+              message: errorMessage
+            }
+          })}\n\n`
+          res.write(errorEvent)
+          res.write('data: [DONE]\n\n')
+          res.end()
+          return { success: false, error: errorType }
+        }
+        if (queueResult.acquired && !queueResult.skipped) {
+          queueLockAcquired = true
+          queueRequestId = queueResult.requestId
+          logger.debug(
+            `📬 User message queue lock acquired for Bedrock account ${accountId} (stream), requestId: ${queueRequestId}`
+          )
+        }
+      }
+
       const modelId = this._selectModel(requestBody, bedrockAccount)
       const region = this._selectRegion(modelId, bedrockAccount)
       const client = this._getBedrockClient(region, bedrockAccount)
@@ -131,6 +291,23 @@ class BedrockRelayService {
       const startTime = Date.now()
       const response = await client.send(command)
 
+      // 📬 请求已发送成功，立即释放队列锁（无需等待响应处理完成）
+      // 因为限流基于请求发送时刻计算（RPM），不是请求完成时刻
+      if (queueLockAcquired && queueRequestId && accountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+          queueLockAcquired = false // 标记已释放，防止 finally 重复释放
+          logger.debug(
+            `📬 User message queue lock released early for Bedrock stream account ${accountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock early for Bedrock stream account ${accountId}:`,
+            releaseError.message
+          )
+        }
+      }
+
       // 设置SSE响应头
       res.writeHead(200, {
         'Content-Type': 'text/event-stream',
@@ -191,6 +368,21 @@ class BedrockRelayService {
       res.end()
 
       throw this._handleBedrockError(error)
+    } finally {
+      // 📬 释放用户消息队列锁（兜底，正常情况下已在请求发送后提前释放）
+      if (queueLockAcquired && queueRequestId && accountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+          logger.debug(
+            `📬 User message queue lock released in finally for Bedrock stream account ${accountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock for Bedrock stream account ${accountId}:`,
+            releaseError.message
+          )
+        }
+      }
     }
   }
 

src/services/ccrRelayService.js

@@ -3,6 +3,7 @@ const ccrAccountService = require('./ccrAccountService')
 const logger = require('../utils/logger')
 const config = require('../../config/config')
 const { parseVendorPrefixedModel } = require('../utils/modelHelper')
+const userMessageQueueService = require('./userMessageQueueService')
 
 class CcrRelayService {
   constructor() {
@@ -21,8 +22,67 @@ class CcrRelayService {
   ) {
     let abortController = null
     let account = null
+    let queueLockAcquired = false
+    let queueRequestId = null
 
     try {
+      // 📬 用户消息队列处理
+      if (userMessageQueueService.isUserMessageRequest(requestBody)) {
+        // 校验 accountId 非空，避免空值污染队列锁键
+        if (!accountId || accountId === '') {
+          logger.error('❌ accountId missing for queue lock in CCR relayRequest')
+          throw new Error('accountId missing for queue lock')
+        }
+        const queueResult = await userMessageQueueService.acquireQueueLock(accountId)
+        if (!queueResult.acquired && !queueResult.skipped) {
+          // 区分 Redis 后端错误和队列超时
+          const isBackendError = queueResult.error === 'queue_backend_error'
+          const errorCode = isBackendError ? 'QUEUE_BACKEND_ERROR' : 'QUEUE_TIMEOUT'
+          const errorType = isBackendError ? 'queue_backend_error' : 'queue_timeout'
+          const errorMessage = isBackendError
+            ? 'Queue service temporarily unavailable, please retry later'
+            : 'User message queue wait timeout, please retry later'
+          const statusCode = isBackendError ? 500 : 503
+
+          // 结构化性能日志，用于后续统计
+          logger.performance('user_message_queue_error', {
+            errorType,
+            errorCode,
+            accountId,
+            statusCode,
+            backendError: isBackendError ? queueResult.errorMessage : undefined
+          })
+
+          logger.warn(
+            `📬 User message queue ${errorType} for CCR account ${accountId}`,
+            isBackendError ? { backendError: queueResult.errorMessage } : {}
+          )
+          return {
+            statusCode,
+            headers: {
+              'Content-Type': 'application/json',
+              'x-user-message-queue-error': errorType
+            },
+            body: JSON.stringify({
+              type: 'error',
+              error: {
+                type: errorType,
+                code: errorCode,
+                message: errorMessage
+              }
+            }),
+            accountId
+          }
+        }
+        if (queueResult.acquired && !queueResult.skipped) {
+          queueLockAcquired = true
+          queueRequestId = queueResult.requestId
+          logger.debug(
+            `📬 User message queue lock acquired for CCR account ${accountId}, requestId: ${queueRequestId}`
+          )
+        }
+      }
+
       // 获取账户信息
       account = await ccrAccountService.getAccount(accountId)
       if (!account) {
@@ -162,6 +222,23 @@ class CcrRelayService {
       )
       const response = await axios(requestConfig)
 
+      // 📬 请求已发送成功，立即释放队列锁（无需等待响应处理完成）
+      // 因为 Claude API 限流基于请求发送时刻计算（RPM），不是请求完成时刻
+      if (queueLockAcquired && queueRequestId && accountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+          queueLockAcquired = false // 标记已释放，防止 finally 重复释放
+          logger.debug(
+            `📬 User message queue lock released early for CCR account ${accountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock early for CCR account ${accountId}:`,
+            releaseError.message
+          )
+        }
+      }
+
       // 移除监听器（请求成功完成）
       if (clientRequest) {
         clientRequest.removeListener('close', handleClientDisconnect)
@@ -233,6 +310,21 @@ class CcrRelayService {
       )
 
       throw error
+    } finally {
+      // 📬 释放用户消息队列锁（兜底，正常情况下已在请求发送后提前释放）
+      if (queueLockAcquired && queueRequestId && accountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+          logger.debug(
+            `📬 User message queue lock released in finally for CCR account ${accountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock for CCR account ${accountId}:`,
+            releaseError.message
+          )
+        }
+      }
     }
   }
 
@@ -248,7 +340,74 @@ class CcrRelayService {
     options = {}
   ) {
     let account = null
+    let queueLockAcquired = false
+    let queueRequestId = null
+
     try {
+      // 📬 用户消息队列处理
+      if (userMessageQueueService.isUserMessageRequest(requestBody)) {
+        // 校验 accountId 非空，避免空值污染队列锁键
+        if (!accountId || accountId === '') {
+          logger.error(
+            '❌ accountId missing for queue lock in CCR relayStreamRequestWithUsageCapture'
+          )
+          throw new Error('accountId missing for queue lock')
+        }
+        const queueResult = await userMessageQueueService.acquireQueueLock(accountId)
+        if (!queueResult.acquired && !queueResult.skipped) {
+          // 区分 Redis 后端错误和队列超时
+          const isBackendError = queueResult.error === 'queue_backend_error'
+          const errorCode = isBackendError ? 'QUEUE_BACKEND_ERROR' : 'QUEUE_TIMEOUT'
+          const errorType = isBackendError ? 'queue_backend_error' : 'queue_timeout'
+          const errorMessage = isBackendError
+            ? 'Queue service temporarily unavailable, please retry later'
+            : 'User message queue wait timeout, please retry later'
+          const statusCode = isBackendError ? 500 : 503
+
+          // 结构化性能日志，用于后续<E5908E><E7BBAD>计
+          logger.performance('user_message_queue_error', {
+            errorType,
+            errorCode,
+            accountId,
+            statusCode,
+            stream: true,
+            backendError: isBackendError ? queueResult.errorMessage : undefined
+          })
+
+          logger.warn(
+            `📬 User message queue ${errorType} for CCR account ${accountId} (stream)`,
+            isBackendError ? { backendError: queueResult.errorMessage } : {}
+          )
+          if (!responseStream.headersSent) {
+            responseStream.writeHead(statusCode, {
+              'Content-Type': 'text/event-stream',
+              'Cache-Control': 'no-cache',
+              Connection: 'keep-alive',
+              'x-user-message-queue-error': errorType
+            })
+          }
+          const errorEvent = `event: error\ndata: ${JSON.stringify({
+            type: 'error',
+            error: {
+              type: errorType,
+              code: errorCode,
+              message: errorMessage
+            }
+          })}\n\n`
+          responseStream.write(errorEvent)
+          responseStream.write('data: [DONE]\n\n')
+          responseStream.end()
+          return
+        }
+        if (queueResult.acquired && !queueResult.skipped) {
+          queueLockAcquired = true
+          queueRequestId = queueResult.requestId
+          logger.debug(
+            `📬 User message queue lock acquired for CCR account ${accountId} (stream), requestId: ${queueRequestId}`
+          )
+        }
+      }
+
       // 获取账户信息
       account = await ccrAccountService.getAccount(accountId)
       if (!account) {
@@ -296,14 +455,53 @@ class CcrRelayService {
         accountId,
         usageCallback,
         streamTransformer,
-        options
+        options,
+        // 📬 回调：在收到响应头时释放队列锁
+        async () => {
+          if (queueLockAcquired && queueRequestId && accountId) {
+            try {
+              await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+              queueLockAcquired = false // 标记已释放，防止 finally 重复释放
+              logger.debug(
+                `📬 User message queue lock released early for CCR stream account ${accountId}, requestId: ${queueRequestId}`
+              )
+            } catch (releaseError) {
+              logger.error(
+                `❌ Failed to release user message queue lock early for CCR stream account ${accountId}:`,
+                releaseError.message
+              )
+            }
+          }
+        }
       )
 
       // 更新最后使用时间
       await this._updateLastUsedTime(accountId)
     } catch (error) {
-      logger.error(`❌ CCR stream relay failed (Account: ${account?.name || accountId}):`, error)
+      // 客户端主动断开连接是正常情况，使用 INFO 级别
+      if (error.message === 'Client disconnected') {
+        logger.info(
+          `🔌 CCR stream relay ended: Client disconnected (Account: ${account?.name || accountId})`
+        )
+      } else {
+        logger.error(`❌ CCR stream relay failed (Account: ${account?.name || accountId}):`, error)
+      }
       throw error
+    } finally {
+      // 📬 释放用户消息队列锁（兜底，正常情况下已在收到响应头后提前释放）
+      if (queueLockAcquired && queueRequestId && accountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+          logger.debug(
+            `📬 User message queue lock released in finally for CCR stream account ${accountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock for CCR stream account ${accountId}:`,
+            releaseError.message
+          )
+        }
+      }
     }
   }
 
@@ -317,7 +515,8 @@ class CcrRelayService {
     accountId,
     usageCallback,
     streamTransformer = null,
-    requestOptions = {}
+    requestOptions = {},
+    onResponseHeaderReceived = null
   ) {
     return new Promise((resolve, reject) => {
       let aborted = false
@@ -380,8 +579,11 @@ class CcrRelayService {
       // 发送请求
       const request = axios(requestConfig)
 
+      // 注意：使用 .then(async ...) 模式处理响应
+      // - 内部的 releaseQueueLock 有独立的 try-catch，不会导致未捕获异常
+      // - queueLockAcquired = false 的赋值会在 finally 执行前完成（JS 单线程保证）
       request
-        .then((response) => {
+        .then(async (response) => {
           logger.debug(`🌊 CCR stream response status: ${response.status}`)
 
           // 错误响应处理
@@ -431,6 +633,19 @@ class CcrRelayService {
             return
           }
 
+          // 📬 收到成功响应头（HTTP 200），调用回调释放队列锁
+          // 此时请求已被 Claude API 接受并计入 RPM 配额，无需等待响应完成
+          if (onResponseHeaderReceived && typeof onResponseHeaderReceived === 'function') {
+            try {
+              await onResponseHeaderReceived()
+            } catch (callbackError) {
+              logger.error(
+                `❌ Failed to execute onResponseHeaderReceived callback for CCR stream account ${accountId}:`,
+                callbackError.message
+              )
+            }
+          }
+
           // 成功响应，检查并移除错误状态
           ccrAccountService.isAccountRateLimited(accountId).then((isRateLimited) => {
             if (isRateLimited) {

src/services/claudeConsoleRelayService.js

@@ -9,6 +9,7 @@ const {
   sanitizeErrorMessage,
   isAccountDisabledError
 } = require('../utils/errorSanitizer')
+const userMessageQueueService = require('./userMessageQueueService')
 
 class ClaudeConsoleRelayService {
   constructor() {
@@ -29,8 +30,68 @@ class ClaudeConsoleRelayService {
     let account = null
     const requestId = uuidv4() // 用于并发追踪
     let concurrencyAcquired = false
+    let queueLockAcquired = false
+    let queueRequestId = null
 
     try {
+      // 📬 用户消息队列处理：如果是用户消息请求，需要获取队列锁
+      if (userMessageQueueService.isUserMessageRequest(requestBody)) {
+        // 校验 accountId 非空，避免空值污染队列锁键
+        if (!accountId || accountId === '') {
+          logger.error('❌ accountId missing for queue lock in console relayRequest')
+          throw new Error('accountId missing for queue lock')
+        }
+        const queueResult = await userMessageQueueService.acquireQueueLock(accountId)
+        if (!queueResult.acquired && !queueResult.skipped) {
+          // 区分 Redis 后端错误和队列超时
+          const isBackendError = queueResult.error === 'queue_backend_error'
+          const errorCode = isBackendError ? 'QUEUE_BACKEND_ERROR' : 'QUEUE_TIMEOUT'
+          const errorType = isBackendError ? 'queue_backend_error' : 'queue_timeout'
+          const errorMessage = isBackendError
+            ? 'Queue service temporarily unavailable, please retry later'
+            : 'User message queue wait timeout, please retry later'
+          const statusCode = isBackendError ? 500 : 503
+
+          // 结构化性能日志，用于后续统计
+          logger.performance('user_message_queue_error', {
+            errorType,
+            errorCode,
+            accountId,
+            statusCode,
+            apiKeyName: apiKeyData.name,
+            backendError: isBackendError ? queueResult.errorMessage : undefined
+          })
+
+          logger.warn(
+            `📬 User message queue ${errorType} for console account ${accountId}, key: ${apiKeyData.name}`,
+            isBackendError ? { backendError: queueResult.errorMessage } : {}
+          )
+          return {
+            statusCode,
+            headers: {
+              'Content-Type': 'application/json',
+              'x-user-message-queue-error': errorType
+            },
+            body: JSON.stringify({
+              type: 'error',
+              error: {
+                type: errorType,
+                code: errorCode,
+                message: errorMessage
+              }
+            }),
+            accountId
+          }
+        }
+        if (queueResult.acquired && !queueResult.skipped) {
+          queueLockAcquired = true
+          queueRequestId = queueResult.requestId
+          logger.debug(
+            `📬 User message queue lock acquired for console account ${accountId}, requestId: ${queueRequestId}`
+          )
+        }
+      }
+
       // 获取账户信息
       account = await claudeConsoleAccountService.getAccount(accountId)
       if (!account) {
@@ -203,6 +264,23 @@ class ClaudeConsoleRelayService {
       )
       const response = await axios(requestConfig)
 
+      // 📬 请求已发送成功，立即释放队列锁（无需等待响应处理完成）
+      // 因为 Claude API 限流基于请求发送时刻计算（RPM），不是请求完成时刻
+      if (queueLockAcquired && queueRequestId && accountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+          queueLockAcquired = false // 标记已释放，防止 finally 重复释放
+          logger.debug(
+            `📬 User message queue lock released early for console account ${accountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock early for console account ${accountId}:`,
+            releaseError.message
+          )
+        }
+      }
+
       // 移除监听器（请求成功完成）
       if (clientRequest) {
         clientRequest.removeListener('close', handleClientDisconnect)
@@ -366,6 +444,21 @@ class ClaudeConsoleRelayService {
           )
         }
       }
+
+      // 📬 释放用户消息队列锁（兜底，正常情况下已在请求发送后提前释放）
+      if (queueLockAcquired && queueRequestId && accountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+          logger.debug(
+            `📬 User message queue lock released in finally for console account ${accountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock for account ${accountId}:`,
+            releaseError.message
+          )
+        }
+      }
     }
   }
 
@@ -384,8 +477,68 @@ class ClaudeConsoleRelayService {
     const requestId = uuidv4() // 用于并发追踪
     let concurrencyAcquired = false
     let leaseRefreshInterval = null // 租约刷新定时器
+    let queueLockAcquired = false
+    let queueRequestId = null
 
     try {
+      // 📬 用户消息队列处理：如果是用户消息请求，需要获取队列锁
+      if (userMessageQueueService.isUserMessageRequest(requestBody)) {
+        // 校验 accountId 非空，避免空值污染队列锁键
+        if (!accountId || accountId === '') {
+          logger.error(
+            '❌ accountId missing for queue lock in console relayStreamRequestWithUsageCapture'
+          )
+          throw new Error('accountId missing for queue lock')
+        }
+        const queueResult = await userMessageQueueService.acquireQueueLock(accountId)
+        if (!queueResult.acquired && !queueResult.skipped) {
+          // 区分 Redis 后端错误和队列超时
+          const isBackendError = queueResult.error === 'queue_backend_error'
+          const errorCode = isBackendError ? 'QUEUE_BACKEND_ERROR' : 'QUEUE_TIMEOUT'
+          const errorType = isBackendError ? 'queue_backend_error' : 'queue_timeout'
+          const errorMessage = isBackendError
+            ? 'Queue service temporarily unavailable, please retry later'
+            : 'User message queue wait timeout, please retry later'
+          const statusCode = isBackendError ? 500 : 503
+
+          // 结构化性能日志，用于后续统计
+          logger.performance('user_message_queue_error', {
+            errorType,
+            errorCode,
+            accountId,
+            statusCode,
+            stream: true,
+            apiKeyName: apiKeyData.name,
+            backendError: isBackendError ? queueResult.errorMessage : undefined
+          })
+
+          logger.warn(
+            `📬 User message queue ${errorType} for console account ${accountId} (stream), key: ${apiKeyData.name}`,
+            isBackendError ? { backendError: queueResult.errorMessage } : {}
+          )
+          if (!responseStream.headersSent) {
+            responseStream.writeHead(statusCode, {
+              'Content-Type': 'text/event-stream',
+              'Cache-Control': 'no-cache',
+              Connection: 'keep-alive',
+              'x-user-message-queue-error': errorType
+            })
+          }
+          const errorEvent = `event: error\ndata: ${JSON.stringify({ type: 'error', error: { type: errorType, code: errorCode, message: errorMessage } })}\n\n`
+          responseStream.write(errorEvent)
+          responseStream.write('data: [DONE]\n\n')
+          responseStream.end()
+          return
+        }
+        if (queueResult.acquired && !queueResult.skipped) {
+          queueLockAcquired = true
+          queueRequestId = queueResult.requestId
+          logger.debug(
+            `📬 User message queue lock acquired for console account ${accountId} (stream), requestId: ${queueRequestId}`
+          )
+        }
+      }
+
       // 获取账户信息
       account = await claudeConsoleAccountService.getAccount(accountId)
       if (!account) {
@@ -483,16 +636,40 @@ class ClaudeConsoleRelayService {
         accountId,
         usageCallback,
         streamTransformer,
-        options
+        options,
+        // 📬 回调：在收到响应头时释放队列锁
+        async () => {
+          if (queueLockAcquired && queueRequestId && accountId) {
+            try {
+              await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+              queueLockAcquired = false // 标记已释放，防止 finally 重复释放
+              logger.debug(
+                `📬 User message queue lock released early for console stream account ${accountId}, requestId: ${queueRequestId}`
+              )
+            } catch (releaseError) {
+              logger.error(
+                `❌ Failed to release user message queue lock early for console stream account ${accountId}:`,
+                releaseError.message
+              )
+            }
+          }
+        }
       )
 
       // 更新最后使用时间
       await this._updateLastUsedTime(accountId)
     } catch (error) {
-      logger.error(
+      // 客户端主动断开连接是正常情况，使用 INFO 级别
-        `❌ Claude Console stream relay failed (Account: ${account?.name || accountId}):`,
+      if (error.message === 'Client disconnected') {
-        error
+        logger.info(
-      )
+          `🔌 Claude Console stream relay ended: Client disconnected (Account: ${account?.name || accountId})`
+        )
+      } else {
+        logger.error(
+          `❌ Claude Console stream relay failed (Account: ${account?.name || accountId}):`,
+          error
+        )
+      }
       throw error
     } finally {
       // 🛑 清理租约刷新定时器
@@ -517,6 +694,21 @@ class ClaudeConsoleRelayService {
           )
         }
       }
+
+      // 📬 释放用户消息队列锁（兜底，正常情况下已在收到响应头后提前释放）
+      if (queueLockAcquired && queueRequestId && accountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(accountId, queueRequestId)
+          logger.debug(
+            `📬 User message queue lock released in finally for console stream account ${accountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock for stream account ${accountId}:`,
+            releaseError.message
+          )
+        }
+      }
     }
   }
 
@@ -530,7 +722,8 @@ class ClaudeConsoleRelayService {
     accountId,
     usageCallback,
     streamTransformer = null,
-    requestOptions = {}
+    requestOptions = {},
+    onResponseHeaderReceived = null
   ) {
     return new Promise((resolve, reject) => {
       let aborted = false
@@ -593,8 +786,11 @@ class ClaudeConsoleRelayService {
       // 发送请求
       const request = axios(requestConfig)
 
+      // 注意：使用 .then(async ...) 模式处理响应
+      // - 内部的 releaseQueueLock 有独立的 try-catch，不会导致未捕获异常
+      // - queueLockAcquired = false 的赋值会在 finally 执行前完成（JS 单线程保证）
       request
-        .then((response) => {
+        .then(async (response) => {
           logger.debug(`🌊 Claude Console Claude stream response status: ${response.status}`)
 
           // 错误响应处理
@@ -701,6 +897,19 @@ class ClaudeConsoleRelayService {
             return
           }
 
+          // 📬 收到成功响应头（HTTP 200），调用回调释放队列锁
+          // 此时请求已被 Claude API 接受并计入 RPM 配额，无需等待响应完成
+          if (onResponseHeaderReceived && typeof onResponseHeaderReceived === 'function') {
+            try {
+              await onResponseHeaderReceived()
+            } catch (callbackError) {
+              logger.error(
+                `❌ Failed to execute onResponseHeaderReceived callback for console stream account ${accountId}:`,
+                callbackError.message
+              )
+            }
+          }
+
           // 成功响应，检查并移除错误状态
           claudeConsoleAccountService.isAccountRateLimited(accountId).then((isRateLimited) => {
             if (isRateLimited) {

src/services/claudeRelayConfigService.js

@@ -0,0 +1,443 @@
+/**
+ * Claude 转发配置服务
+ * 管理全局 Claude Code 限制和会话绑定配置
+ */
+
+const redis = require('../models/redis')
+const logger = require('../utils/logger')
+
+const CONFIG_KEY = 'claude_relay_config'
+const SESSION_BINDING_PREFIX = 'original_session_binding:'
+
+// 默认配置
+const DEFAULT_CONFIG = {
+  claudeCodeOnlyEnabled: false,
+  globalSessionBindingEnabled: false,
+  sessionBindingErrorMessage: '你的本地session已污染，请清理后使用。',
+  sessionBindingTtlDays: 30, // 会话绑定 TTL（天），默认30天
+  // 用户消息队列配置
+  userMessageQueueEnabled: false, // 是否启用用户消息队列（默认关闭）
+  userMessageQueueDelayMs: 200, // 请求间隔（毫秒）
+  userMessageQueueTimeoutMs: 5000, // 队列等待超时（毫秒），优化后锁持有时间短无需长等待
+  userMessageQueueLockTtlMs: 5000, // 锁TTL（毫秒），请求发送后立即释放无需长TTL
+  updatedAt: null,
+  updatedBy: null
+}
+
+// 内存缓存（避免频繁 Redis 查询）
+let configCache = null
+let configCacheTime = 0
+const CONFIG_CACHE_TTL = 60000 // 1分钟缓存
+
+class ClaudeRelayConfigService {
+  /**
+   * 从 metadata.user_id 中提取原始 sessionId
+   * 格式: user_{64位十六进制}_account__session_{uuid}
+   * @param {Object} requestBody - 请求体
+   * @returns {string|null} 原始 sessionId 或 null
+   */
+  extractOriginalSessionId(requestBody) {
+    if (!requestBody?.metadata?.user_id) {
+      return null
+    }
+
+    const userId = requestBody.metadata.user_id
+    const match = userId.match(/session_([a-f0-9-]{36})$/i)
+    return match ? match[1] : null
+  }
+
+  /**
+   * 获取配置（带缓存）
+   * @returns {Promise<Object>} 配置对象
+   */
+  async getConfig() {
+    try {
+      // 检查缓存
+      if (configCache && Date.now() - configCacheTime < CONFIG_CACHE_TTL) {
+        return configCache
+      }
+
+      const client = redis.getClient()
+      if (!client) {
+        logger.warn('⚠️ Redis not connected, using default config')
+        return { ...DEFAULT_CONFIG }
+      }
+
+      const data = await client.get(CONFIG_KEY)
+
+      if (data) {
+        configCache = { ...DEFAULT_CONFIG, ...JSON.parse(data) }
+      } else {
+        configCache = { ...DEFAULT_CONFIG }
+      }
+
+      configCacheTime = Date.now()
+      return configCache
+    } catch (error) {
+      logger.error('❌ Failed to get Claude relay config:', error)
+      return { ...DEFAULT_CONFIG }
+    }
+  }
+
+  /**
+   * 更新配置
+   * @param {Object} newConfig - 新配置
+   * @param {string} updatedBy - 更新者
+   * @returns {Promise<Object>} 更新后的配置
+   */
+  async updateConfig(newConfig, updatedBy) {
+    try {
+      const client = redis.getClientSafe()
+      const currentConfig = await this.getConfig()
+
+      const updatedConfig = {
+        ...currentConfig,
+        ...newConfig,
+        updatedAt: new Date().toISOString(),
+        updatedBy
+      }
+
+      await client.set(CONFIG_KEY, JSON.stringify(updatedConfig))
+
+      // 更新缓存
+      configCache = updatedConfig
+      configCacheTime = Date.now()
+
+      logger.info(`✅ Claude relay config updated by ${updatedBy}:`, {
+        claudeCodeOnlyEnabled: updatedConfig.claudeCodeOnlyEnabled,
+        globalSessionBindingEnabled: updatedConfig.globalSessionBindingEnabled
+      })
+
+      return updatedConfig
+    } catch (error) {
+      logger.error('❌ Failed to update Claude relay config:', error)
+      throw error
+    }
+  }
+
+  /**
+   * 检查是否启用全局 Claude Code 限制
+   * @returns {Promise<boolean>}
+   */
+  async isClaudeCodeOnlyEnabled() {
+    const cfg = await this.getConfig()
+    return cfg.claudeCodeOnlyEnabled === true
+  }
+
+  /**
+   * 检查是否启用全局会话绑定
+   * @returns {Promise<boolean>}
+   */
+  async isGlobalSessionBindingEnabled() {
+    const cfg = await this.getConfig()
+    return cfg.globalSessionBindingEnabled === true
+  }
+
+  /**
+   * 获取会话绑定错误信息
+   * @returns {Promise<string>}
+   */
+  async getSessionBindingErrorMessage() {
+    const cfg = await this.getConfig()
+    return cfg.sessionBindingErrorMessage || DEFAULT_CONFIG.sessionBindingErrorMessage
+  }
+
+  /**
+   * 获取原始会话绑定
+   * @param {string} originalSessionId - 原始会话ID
+   * @returns {Promise<Object|null>} 绑定信息或 null
+   */
+  async getOriginalSessionBinding(originalSessionId) {
+    if (!originalSessionId) {
+      return null
+    }
+
+    try {
+      const client = redis.getClient()
+      if (!client) {
+        return null
+      }
+
+      const key = `${SESSION_BINDING_PREFIX}${originalSessionId}`
+      const data = await client.get(key)
+
+      if (data) {
+        return JSON.parse(data)
+      }
+      return null
+    } catch (error) {
+      logger.error(`❌ Failed to get session binding for ${originalSessionId}:`, error)
+      return null
+    }
+  }
+
+  /**
+   * 设置原始会话绑定
+   * @param {string} originalSessionId - 原始会话ID
+   * @param {string} accountId - 账户ID
+   * @param {string} accountType - 账户类型
+   * @returns {Promise<Object>} 绑定信息
+   */
+  async setOriginalSessionBinding(originalSessionId, accountId, accountType) {
+    if (!originalSessionId || !accountId || !accountType) {
+      throw new Error('Invalid parameters for session binding')
+    }
+
+    try {
+      const client = redis.getClientSafe()
+      const key = `${SESSION_BINDING_PREFIX}${originalSessionId}`
+
+      const binding = {
+        accountId,
+        accountType,
+        createdAt: new Date().toISOString(),
+        lastUsedAt: new Date().toISOString()
+      }
+
+      // 使用配置的 TTL（默认30天）
+      const cfg = await this.getConfig()
+      const ttlDays = cfg.sessionBindingTtlDays || DEFAULT_CONFIG.sessionBindingTtlDays
+      const ttlSeconds = Math.floor(ttlDays * 24 * 3600)
+
+      await client.set(key, JSON.stringify(binding), 'EX', ttlSeconds)
+
+      logger.info(
+        `🔗 Session binding created: ${originalSessionId} -> ${accountId} (${accountType})`
+      )
+
+      return binding
+    } catch (error) {
+      logger.error(`❌ Failed to set session binding for ${originalSessionId}:`, error)
+      throw error
+    }
+  }
+
+  /**
+   * 更新会话绑定的最后使用时间（续期）
+   * @param {string} originalSessionId - 原始会话ID
+   */
+  async touchOriginalSessionBinding(originalSessionId) {
+    if (!originalSessionId) {
+      return
+    }
+
+    try {
+      const binding = await this.getOriginalSessionBinding(originalSessionId)
+      if (!binding) {
+        return
+      }
+
+      binding.lastUsedAt = new Date().toISOString()
+
+      const client = redis.getClientSafe()
+      const key = `${SESSION_BINDING_PREFIX}${originalSessionId}`
+
+      // 使用配置的 TTL（默认30天）
+      const cfg = await this.getConfig()
+      const ttlDays = cfg.sessionBindingTtlDays || DEFAULT_CONFIG.sessionBindingTtlDays
+      const ttlSeconds = Math.floor(ttlDays * 24 * 3600)
+
+      await client.set(key, JSON.stringify(binding), 'EX', ttlSeconds)
+    } catch (error) {
+      logger.warn(`⚠️ Failed to touch session binding for ${originalSessionId}:`, error)
+    }
+  }
+
+  /**
+   * 检查原始会话是否已绑定
+   * @param {string} originalSessionId - 原始会话ID
+   * @returns {Promise<boolean>}
+   */
+  async isOriginalSessionBound(originalSessionId) {
+    const binding = await this.getOriginalSessionBinding(originalSessionId)
+    return binding !== null
+  }
+
+  /**
+   * 验证绑定的账户是否可用
+   * @param {Object} binding - 绑定信息
+   * @returns {Promise<boolean>}
+   */
+  async validateBoundAccount(binding) {
+    if (!binding || !binding.accountId || !binding.accountType) {
+      return false
+    }
+
+    try {
+      const { accountType } = binding
+      const { accountId } = binding
+
+      let accountService
+      switch (accountType) {
+        case 'claude-official':
+          accountService = require('./claudeAccountService')
+          break
+        case 'claude-console':
+          accountService = require('./claudeConsoleAccountService')
+          break
+        case 'bedrock':
+          accountService = require('./bedrockAccountService')
+          break
+        case 'ccr':
+          accountService = require('./ccrAccountService')
+          break
+        default:
+          logger.warn(`Unknown account type for validation: ${accountType}`)
+          return false
+      }
+
+      const account = await accountService.getAccount(accountId)
+
+      // getAccount() 直接返回账户数据对象或 null，不是 { success, data } 格式
+      if (!account) {
+        logger.warn(`Session binding account not found: ${accountId} (${accountType})`)
+        return false
+      }
+
+      const accountData = account
+
+      // 检查账户是否激活
+      if (accountData.isActive === false || accountData.isActive === 'false') {
+        logger.warn(
+          `Session binding account not active: ${accountId} (${accountType}), isActive: ${accountData.isActive}`
+        )
+        return false
+      }
+
+      // 检查账户状态（如果存在）
+      if (accountData.status && accountData.status === 'error') {
+        logger.warn(
+          `Session binding account has error status: ${accountId} (${accountType}), status: ${accountData.status}`
+        )
+        return false
+      }
+
+      return true
+    } catch (error) {
+      logger.error(`❌ Failed to validate bound account ${binding.accountId}:`, error)
+      return false
+    }
+  }
+
+  /**
+   * 验证新会话请求
+   * @param {Object} _requestBody - 请求体（预留参数，当前未使用）
+   * @param {string} originalSessionId - 原始会话ID
+   * @returns {Promise<Object>} { valid: boolean, error?: string, binding?: object, isNewSession?: boolean }
+   */
+  async validateNewSession(_requestBody, originalSessionId) {
+    const cfg = await this.getConfig()
+
+    if (!cfg.globalSessionBindingEnabled) {
+      return { valid: true }
+    }
+
+    // 如果没有 sessionId，跳过验证（可能是非 Claude Code 客户端）
+    if (!originalSessionId) {
+      return { valid: true }
+    }
+
+    const existingBinding = await this.getOriginalSessionBinding(originalSessionId)
+
+    // 如果会话已存在绑定
+    if (existingBinding) {
+      // ⚠️ 只有 claude-official 类型账户受全局会话绑定限制
+      // 其他类型（bedrock, ccr, claude-console等）忽略绑定，走正常调度
+      if (existingBinding.accountType !== 'claude-official') {
+        logger.info(
+          `🔗 Session binding ignored for non-official account type: ${existingBinding.accountType}`
+        )
+        return { valid: true }
+      }
+
+      const accountValid = await this.validateBoundAccount(existingBinding)
+
+      if (!accountValid) {
+        return {
+          valid: false,
+          error: cfg.sessionBindingErrorMessage,
+          code: 'SESSION_BINDING_INVALID'
+        }
+      }
+
+      // 续期
+      await this.touchOriginalSessionBinding(originalSessionId)
+
+      // 已有绑定，允许继续（这是正常的会话延续）
+      return { valid: true, binding: existingBinding }
+    }
+
+    // 没有绑定，是新会话
+    // 注意：messages.length 检查在此处无法执行，因为我们不知道最终会调度到哪种账户类型
+    // 绑定会在调度后创建，仅针对 claude-official 账户
+    return { valid: true, isNewSession: true }
+  }
+
+  /**
+   * 删除原始会话绑定
+   * @param {string} originalSessionId - 原始会话ID
+   */
+  async deleteOriginalSessionBinding(originalSessionId) {
+    if (!originalSessionId) {
+      return
+    }
+
+    try {
+      const client = redis.getClient()
+      if (!client) {
+        return
+      }
+
+      const key = `${SESSION_BINDING_PREFIX}${originalSessionId}`
+      await client.del(key)
+      logger.info(`🗑️ Session binding deleted: ${originalSessionId}`)
+    } catch (error) {
+      logger.error(`❌ Failed to delete session binding for ${originalSessionId}:`, error)
+    }
+  }
+
+  /**
+   * 获取会话绑定统计
+   * @returns {Promise<Object>}
+   */
+  async getSessionBindingStats() {
+    try {
+      const client = redis.getClient()
+      if (!client) {
+        return { totalBindings: 0 }
+      }
+
+      let cursor = '0'
+      let count = 0
+
+      do {
+        const [newCursor, keys] = await client.scan(
+          cursor,
+          'MATCH',
+          `${SESSION_BINDING_PREFIX}*`,
+          'COUNT',
+          100
+        )
+        cursor = newCursor
+        count += keys.length
+      } while (cursor !== '0')
+
+      return {
+        totalBindings: count
+      }
+    } catch (error) {
+      logger.error('❌ Failed to get session binding stats:', error)
+      return { totalBindings: 0 }
+    }
+  }
+
+  /**
+   * 清除配置缓存（用于测试或强制刷新）
+   */
+  clearCache() {
+    configCache = null
+    configCacheTime = 0
+  }
+}
+
+module.exports = new ClaudeRelayConfigService()

src/services/claudeRelayService.js

@@ -15,10 +15,11 @@ const ClaudeCodeValidator = require('../validators/clients/claudeCodeValidator')
 const { formatDateWithTimezone } = require('../utils/dateHelper')
 const requestIdentityService = require('./requestIdentityService')
 const { createClaudeTestPayload } = require('../utils/testPayloadHelper')
+const userMessageQueueService = require('./userMessageQueueService')
 
 class ClaudeRelayService {
   constructor() {
-    this.claudeApiUrl = config.claude.apiUrl
+    this.claudeApiUrl = 'https://api.anthropic.com/v1/messages?beta=true'
     this.apiVersion = config.claude.apiVersion
     this.betaHeader = config.claude.betaHeader
     this.systemPrompt = config.claude.systemPrompt
@@ -148,6 +149,9 @@ class ClaudeRelayService {
     options = {}
   ) {
     let upstreamRequest = null
+    let queueLockAcquired = false
+    let queueRequestId = null
+    let selectedAccountId = null
 
     try {
       // 调试日志：查看API Key数据
@@ -192,11 +196,70 @@ class ClaudeRelayService {
       }
       const { accountId } = accountSelection
       const { accountType } = accountSelection
+      selectedAccountId = accountId
 
       logger.info(
         `📤 Processing API request for key: ${apiKeyData.name || apiKeyData.id}, account: ${accountId} (${accountType})${sessionHash ? `, session: ${sessionHash}` : ''}`
       )
 
+      // 📬 用户消息队列处理：如果是用户消息请求，需要获取队列锁
+      if (userMessageQueueService.isUserMessageRequest(requestBody)) {
+        // 校验 accountId 非空，避免空值污染队列锁键
+        if (!accountId || accountId === '') {
+          logger.error('❌ accountId missing for queue lock in relayRequest')
+          throw new Error('accountId missing for queue lock')
+        }
+        const queueResult = await userMessageQueueService.acquireQueueLock(accountId)
+        if (!queueResult.acquired && !queueResult.skipped) {
+          // 区分 Redis 后端错误和队列超时
+          const isBackendError = queueResult.error === 'queue_backend_error'
+          const errorCode = isBackendError ? 'QUEUE_BACKEND_ERROR' : 'QUEUE_TIMEOUT'
+          const errorType = isBackendError ? 'queue_backend_error' : 'queue_timeout'
+          const errorMessage = isBackendError
+            ? 'Queue service temporarily unavailable, please retry later'
+            : 'User message queue wait timeout, please retry later'
+          const statusCode = isBackendError ? 500 : 503
+
+          // 结构化性能日志，用于后续统计
+          logger.performance('user_message_queue_error', {
+            errorType,
+            errorCode,
+            accountId,
+            statusCode,
+            apiKeyName: apiKeyData.name,
+            backendError: isBackendError ? queueResult.errorMessage : undefined
+          })
+
+          logger.warn(
+            `📬 User message queue ${errorType} for account ${accountId}, key: ${apiKeyData.name}`,
+            isBackendError ? { backendError: queueResult.errorMessage } : {}
+          )
+          return {
+            statusCode,
+            headers: {
+              'Content-Type': 'application/json',
+              'x-user-message-queue-error': errorType
+            },
+            body: JSON.stringify({
+              type: 'error',
+              error: {
+                type: errorType,
+                code: errorCode,
+                message: errorMessage
+              }
+            }),
+            accountId
+          }
+        }
+        if (queueResult.acquired && !queueResult.skipped) {
+          queueLockAcquired = true
+          queueRequestId = queueResult.requestId
+          logger.debug(
+            `📬 User message queue lock acquired for account ${accountId}, requestId: ${queueRequestId}`
+          )
+        }
+      }
+
       // 获取账户信息
       let account = await claudeAccountService.getAccount(accountId)
 
@@ -271,6 +334,23 @@ class ClaudeRelayService {
         options
       )
 
+      // 📬 请求已发送成功，立即释放队列锁（无需等待响应处理完成）
+      // 因为 Claude API 限流基于请求发送时刻计算（RPM），不是请求完成时刻
+      if (queueLockAcquired && queueRequestId && selectedAccountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(selectedAccountId, queueRequestId)
+          queueLockAcquired = false // 标记已释放，防止 finally 重复释放
+          logger.debug(
+            `📬 User message queue lock released early for account ${selectedAccountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock early for account ${selectedAccountId}:`,
+            releaseError.message
+          )
+        }
+      }
+
       response.accountId = accountId
       response.accountType = accountType
 
@@ -539,6 +619,21 @@ class ClaudeRelayService {
         error.message
       )
       throw error
+    } finally {
+      // 📬 释放用户消息队列锁（兜底，正常情况下已在请求发送后提前释放）
+      if (queueLockAcquired && queueRequestId && selectedAccountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(selectedAccountId, queueRequestId)
+          logger.debug(
+            `📬 User message queue lock released in finally for account ${selectedAccountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock for account ${selectedAccountId}:`,
+            releaseError.message
+          )
+        }
+      }
     }
   }
 
@@ -878,11 +973,102 @@ class ClaudeRelayService {
 
   // 🔧 过滤客户端请求头
   _filterClientHeaders(clientHeaders) {
-    // 使用统一的 headerFilter 工具类 - 移除 CDN、浏览器和代理相关 headers
+    // 使用统一的 headerFilter 工具类
     // 同时伪装成正常的直接客户端请求，避免触发上游 API 的安全检查
     return filterForClaude(clientHeaders)
   }
 
+  // 🔧 准备请求头和 payload（抽离公共逻辑）
+  async _prepareRequestHeadersAndPayload(
+    body,
+    clientHeaders,
+    accountId,
+    accessToken,
+    options = {}
+  ) {
+    const { account, accountType, sessionHash, requestOptions = {}, isStream = false } = options
+
+    // 获取统一的 User-Agent
+    const unifiedUA = await this.captureAndGetUnifiedUserAgent(clientHeaders, account)
+
+    // 获取过滤后的客户端 headers
+    const filteredHeaders = this._filterClientHeaders(clientHeaders)
+
+    // 判断是否是真实的 Claude Code 请求
+    const isRealClaudeCode = this.isRealClaudeCodeRequest(body)
+
+    // 如果不是真实的 Claude Code 请求，需要使用从账户获取的 Claude Code headers
+    let finalHeaders = { ...filteredHeaders }
+    let requestPayload = body
+
+    if (!isRealClaudeCode) {
+      // 获取该账号存储的 Claude Code headers
+      const claudeCodeHeaders = await claudeCodeHeadersService.getAccountHeaders(accountId)
+
+      // 只添加客户端没有提供的 headers
+      Object.keys(claudeCodeHeaders).forEach((key) => {
+        const lowerKey = key.toLowerCase()
+        if (!finalHeaders[key] && !finalHeaders[lowerKey]) {
+          finalHeaders[key] = claudeCodeHeaders[key]
+        }
+      })
+    }
+
+    // 应用请求身份转换
+    const extensionResult = this._applyRequestIdentityTransform(requestPayload, finalHeaders, {
+      account,
+      accountId,
+      accountType,
+      sessionHash,
+      clientHeaders,
+      requestOptions,
+      isStream
+    })
+
+    if (extensionResult.abortResponse) {
+      return { abortResponse: extensionResult.abortResponse }
+    }
+
+    requestPayload = extensionResult.body
+    finalHeaders = extensionResult.headers
+
+    // 序列化请求体，计算 content-length
+    const bodyString = JSON.stringify(requestPayload)
+    const contentLength = Buffer.byteLength(bodyString, 'utf8')
+
+    // 构建最终请求头（包含认证、版本、User-Agent、Beta 等）
+    const headers = {
+      host: 'api.anthropic.com',
+      connection: 'keep-alive',
+      'content-type': 'application/json',
+      'content-length': String(contentLength),
+      authorization: `Bearer ${accessToken}`,
+      'anthropic-version': this.apiVersion,
+      ...finalHeaders
+    }
+
+    // 使用统一 User-Agent 或客户端提供的，最后使用默认值
+    const userAgent = unifiedUA || headers['user-agent'] || 'claude-cli/1.0.119 (external, cli)'
+    const acceptHeader = headers['accept'] || 'application/json'
+    delete headers['user-agent']
+    delete headers['accept']
+    headers['User-Agent'] = userAgent
+    headers['Accept'] = acceptHeader
+
+    logger.info(`🔗 指纹是这个: ${headers['User-Agent']}`)
+
+    // 根据模型和客户端传递的 anthropic-beta 动态设置 header
+    const modelId = requestPayload?.model || body?.model
+    const clientBetaHeader = clientHeaders?.['anthropic-beta']
+    headers['anthropic-beta'] = this._getBetaHeader(modelId, clientBetaHeader)
+    return {
+      requestPayload,
+      bodyString,
+      headers,
+      isRealClaudeCode
+    }
+  }
+
   _applyRequestIdentityTransform(body, headers, context = {}) {
     const normalizedHeaders = headers && typeof headers === 'object' ? { ...headers } : {}
 
@@ -928,46 +1114,24 @@ class ClaudeRelayService {
     // 获取账户信息用于统一 User-Agent
     const account = await claudeAccountService.getAccount(accountId)
 
-    // 获取统一的 User-Agent
+    // 使用公共方法准备请求头和 payload
-    const unifiedUA = await this.captureAndGetUnifiedUserAgent(clientHeaders, account)
+    const prepared = await this._prepareRequestHeadersAndPayload(
-
+      body,
-    // 获取过滤后的客户端 headers
-    const filteredHeaders = this._filterClientHeaders(clientHeaders)
-
-    // 判断是否是真实的 Claude Code 请求
-    const isRealClaudeCode = this.isRealClaudeCodeRequest(body)
-
-    // 如果不是真实的 Claude Code 请求，需要使用从账户获取的 Claude Code headers
-    let finalHeaders = { ...filteredHeaders }
-    let requestPayload = body
-
-    if (!isRealClaudeCode) {
-      // 获取该账号存储的 Claude Code headers
-      const claudeCodeHeaders = await claudeCodeHeadersService.getAccountHeaders(accountId)
-
-      // 只添加客户端没有提供的 headers
-      Object.keys(claudeCodeHeaders).forEach((key) => {
-        const lowerKey = key.toLowerCase()
-        if (!finalHeaders[key] && !finalHeaders[lowerKey]) {
-          finalHeaders[key] = claudeCodeHeaders[key]
-        }
-      })
-    }
-
-    const extensionResult = this._applyRequestIdentityTransform(requestPayload, finalHeaders, {
-      account,
-      accountId,
       clientHeaders,
-      requestOptions,
+      accountId,
-      isStream: false
+      accessToken,
-    })
+      {
+        account,
+        requestOptions,
+        isStream: false
+      }
+    )
 
-    if (extensionResult.abortResponse) {
+    if (prepared.abortResponse) {
-      return extensionResult.abortResponse
+      return prepared.abortResponse
     }
 
-    requestPayload = extensionResult.body
+    const { bodyString, headers } = prepared
-    finalHeaders = extensionResult.headers
 
     return new Promise((resolve, reject) => {
       // 支持自定义路径（如 count_tokens）
@@ -981,31 +1145,13 @@ class ClaudeRelayService {
       const options = {
         hostname: url.hostname,
         port: url.port || 443,
-        path: requestPath,
+        path: requestPath + (url.search || ''),
         method: 'POST',
-        headers: {
+        headers,
-          'Content-Type': 'application/json',
-          Authorization: `Bearer ${accessToken}`,
-          'anthropic-version': this.apiVersion,
-          ...finalHeaders
-        },
         agent: proxyAgent,
         timeout: config.requestTimeout || 600000
       }
 
-      // 使用统一 User-Agent 或客户端提供的，最后使用默认值
-      if (!options.headers['user-agent'] || unifiedUA !== null) {
-        const userAgent = unifiedUA || 'claude-cli/1.0.119 (external, cli)'
-        options.headers['user-agent'] = userAgent
-      }
-
-      logger.info(`🔗 指纹是这个: ${options.headers['user-agent']}`)
-
-      // 根据模型和客户端传递的 anthropic-beta 动态设置 header
-      const modelId = requestPayload?.model || body?.model
-      const clientBetaHeader = clientHeaders?.['anthropic-beta']
-      options.headers['anthropic-beta'] = this._getBetaHeader(modelId, clientBetaHeader)
-
       const req = https.request(options, (res) => {
         let responseData = Buffer.alloc(0)
 
@@ -1015,32 +1161,32 @@ class ClaudeRelayService {
 
         res.on('end', () => {
           try {
-            let bodyString = ''
+            let responseBody = ''
 
             // 根据Content-Encoding处理响应数据
             const contentEncoding = res.headers['content-encoding']
             if (contentEncoding === 'gzip') {
               try {
-                bodyString = zlib.gunzipSync(responseData).toString('utf8')
+                responseBody = zlib.gunzipSync(responseData).toString('utf8')
               } catch (unzipError) {
                 logger.error('❌ Failed to decompress gzip response:', unzipError)
-                bodyString = responseData.toString('utf8')
+                responseBody = responseData.toString('utf8')
               }
             } else if (contentEncoding === 'deflate') {
               try {
-                bodyString = zlib.inflateSync(responseData).toString('utf8')
+                responseBody = zlib.inflateSync(responseData).toString('utf8')
               } catch (unzipError) {
                 logger.error('❌ Failed to decompress deflate response:', unzipError)
-                bodyString = responseData.toString('utf8')
+                responseBody = responseData.toString('utf8')
               }
             } else {
-              bodyString = responseData.toString('utf8')
+              responseBody = responseData.toString('utf8')
             }
 
             const response = {
               statusCode: res.statusCode,
               headers: res.headers,
-              body: bodyString
+              body: responseBody
             }
 
             logger.debug(`🔗 Claude API response: ${res.statusCode}`)
@@ -1059,7 +1205,6 @@ class ClaudeRelayService {
       }
 
       req.on('error', async (error) => {
-        console.error(': ❌ ', error)
         logger.error(`❌ Claude API request error (Account: ${accountId}):`, error.message, {
           code: error.code,
           errno: error.errno,
@@ -1095,7 +1240,7 @@ class ClaudeRelayService {
       })
 
       // 写入请求体
-      req.write(JSON.stringify(requestPayload))
+      req.write(bodyString)
       req.end()
     })
   }
@@ -1110,6 +1255,10 @@ class ClaudeRelayService {
     streamTransformer = null,
     options = {}
   ) {
+    let queueLockAcquired = false
+    let queueRequestId = null
+    let selectedAccountId = null
+
     try {
       // 调试日志：查看API Key数据（流式请求）
       logger.info('🔍 [Stream] API Key data received:', {
@@ -1153,6 +1302,70 @@ class ClaudeRelayService {
       }
       const { accountId } = accountSelection
       const { accountType } = accountSelection
+      selectedAccountId = accountId
+
+      // 📬 用户消息队列处理：如果是用户消息请求，需要获取队列锁
+      if (userMessageQueueService.isUserMessageRequest(requestBody)) {
+        // 校验 accountId 非空，避免空值污染队列锁键
+        if (!accountId || accountId === '') {
+          logger.error('❌ accountId missing for queue lock in relayStreamRequestWithUsageCapture')
+          throw new Error('accountId missing for queue lock')
+        }
+        const queueResult = await userMessageQueueService.acquireQueueLock(accountId)
+        if (!queueResult.acquired && !queueResult.skipped) {
+          // 区分 Redis 后端错误和队列超时
+          const isBackendError = queueResult.error === 'queue_backend_error'
+          const errorCode = isBackendError ? 'QUEUE_BACKEND_ERROR' : 'QUEUE_TIMEOUT'
+          const errorType = isBackendError ? 'queue_backend_error' : 'queue_timeout'
+          const errorMessage = isBackendError
+            ? 'Queue service temporarily unavailable, please retry later'
+            : 'User message queue wait timeout, please retry later'
+          const statusCode = isBackendError ? 500 : 503
+
+          // 结构化性能日志，用于后续统计
+          logger.performance('user_message_queue_error', {
+            errorType,
+            errorCode,
+            accountId,
+            statusCode,
+            stream: true,
+            apiKeyName: apiKeyData.name,
+            backendError: isBackendError ? queueResult.errorMessage : undefined
+          })
+
+          logger.warn(
+            `📬 User message queue ${errorType} for account ${accountId} (stream), key: ${apiKeyData.name}`,
+            isBackendError ? { backendError: queueResult.errorMessage } : {}
+          )
+          if (!responseStream.headersSent) {
+            responseStream.writeHead(statusCode, {
+              'Content-Type': 'text/event-stream',
+              'Cache-Control': 'no-cache',
+              Connection: 'keep-alive',
+              'x-user-message-queue-error': errorType
+            })
+          }
+          const errorEvent = `event: error\ndata: ${JSON.stringify({
+            type: 'error',
+            error: {
+              type: errorType,
+              code: errorCode,
+              message: errorMessage
+            }
+          })}\n\n`
+          responseStream.write(errorEvent)
+          responseStream.write('data: [DONE]\n\n')
+          responseStream.end()
+          return
+        }
+        if (queueResult.acquired && !queueResult.skipped) {
+          queueLockAcquired = true
+          queueRequestId = queueResult.requestId
+          logger.debug(
+            `📬 User message queue lock acquired for account ${accountId} (stream), requestId: ${queueRequestId}`
+          )
+        }
+      }
 
       logger.info(
         `📡 Processing streaming API request with usage capture for key: ${apiKeyData.name || apiKeyData.id}, account: ${accountId} (${accountType})${sessionHash ? `, session: ${sessionHash}` : ''}`
@@ -1219,11 +1432,48 @@ class ClaudeRelayService {
         sessionHash,
         streamTransformer,
         options,
-        isDedicatedOfficialAccount
+        isDedicatedOfficialAccount,
+        // 📬 新增回调：在收到响应头时释放队列锁
+        async () => {
+          if (queueLockAcquired && queueRequestId && selectedAccountId) {
+            try {
+              await userMessageQueueService.releaseQueueLock(selectedAccountId, queueRequestId)
+              queueLockAcquired = false // 标记已释放，防止 finally 重复释放
+              logger.debug(
+                `📬 User message queue lock released early for stream account ${selectedAccountId}, requestId: ${queueRequestId}`
+              )
+            } catch (releaseError) {
+              logger.error(
+                `❌ Failed to release user message queue lock early for stream account ${selectedAccountId}:`,
+                releaseError.message
+              )
+            }
+          }
+        }
       )
     } catch (error) {
-      logger.error(`❌ Claude stream relay with usage capture failed:`, error)
+      // 客户端主动断开连接是正常情况，使用 INFO 级别
+      if (error.message === 'Client disconnected') {
+        logger.info(`🔌 Claude stream relay ended: Client disconnected`)
+      } else {
+        logger.error(`❌ Claude stream relay with usage capture failed:`, error)
+      }
       throw error
+    } finally {
+      // 📬 释放用户消息队列锁（兜底，正常情况下已在收到响应头后提前释放）
+      if (queueLockAcquired && queueRequestId && selectedAccountId) {
+        try {
+          await userMessageQueueService.releaseQueueLock(selectedAccountId, queueRequestId)
+          logger.debug(
+            `📬 User message queue lock released in finally for stream account ${selectedAccountId}, requestId: ${queueRequestId}`
+          )
+        } catch (releaseError) {
+          logger.error(
+            `❌ Failed to release user message queue lock for stream account ${selectedAccountId}:`,
+            releaseError.message
+          )
+        }
+      }
     }
   }
 
@@ -1240,7 +1490,8 @@ class ClaudeRelayService {
     sessionHash,
     streamTransformer = null,
     requestOptions = {},
-    isDedicatedOfficialAccount = false
+    isDedicatedOfficialAccount = false,
+    onResponseStart = null // 📬 新增：收到响应头时的回调，用于提前释放队列锁
   ) {
     // 获取账户信息用于统一 User-Agent
     const account = await claudeAccountService.getAccount(accountId)
@@ -1248,79 +1499,39 @@ class ClaudeRelayService {
     const isOpusModelRequest =
       typeof body?.model === 'string' && body.model.toLowerCase().includes('opus')
 
-    // 获取统一的 User-Agent
+    // 使用公共方法准备请求头和 payload
-    const unifiedUA = await this.captureAndGetUnifiedUserAgent(clientHeaders, account)
+    const prepared = await this._prepareRequestHeadersAndPayload(
-
+      body,
-    // 获取过滤后的客户端 headers
-    const filteredHeaders = this._filterClientHeaders(clientHeaders)
-
-    // 判断是否是真实的 Claude Code 请求
-    const isRealClaudeCode = this.isRealClaudeCodeRequest(body)
-
-    // 如果不是真实的 Claude Code 请求，需要使用从账户获取的 Claude Code headers
-    let finalHeaders = { ...filteredHeaders }
-    let requestPayload = body
-
-    if (!isRealClaudeCode) {
-      // 获取该账号存储的 Claude Code headers
-      const claudeCodeHeaders = await claudeCodeHeadersService.getAccountHeaders(accountId)
-
-      // 只添加客户端没有提供的 headers
-      Object.keys(claudeCodeHeaders).forEach((key) => {
-        const lowerKey = key.toLowerCase()
-        if (!finalHeaders[key] && !finalHeaders[lowerKey]) {
-          finalHeaders[key] = claudeCodeHeaders[key]
-        }
-      })
-    }
-
-    const extensionResult = this._applyRequestIdentityTransform(requestPayload, finalHeaders, {
-      account,
-      accountId,
-      accountType,
-      sessionHash,
       clientHeaders,
-      requestOptions,
+      accountId,
-      isStream: true
+      accessToken,
-    })
+      {
+        account,
+        accountType,
+        sessionHash,
+        requestOptions,
+        isStream: true
+      }
+    )
 
-    if (extensionResult.abortResponse) {
+    if (prepared.abortResponse) {
-      return extensionResult.abortResponse
+      return prepared.abortResponse
     }
 
-    requestPayload = extensionResult.body
+    const { bodyString, headers } = prepared
-    finalHeaders = extensionResult.headers
 
     return new Promise((resolve, reject) => {
       const url = new URL(this.claudeApiUrl)
-
       const options = {
         hostname: url.hostname,
         port: url.port || 443,
-        path: url.pathname,
+        path: url.pathname + (url.search || ''),
         method: 'POST',
-        headers: {
+        headers,
-          'Content-Type': 'application/json',
-          Authorization: `Bearer ${accessToken}`,
-          'anthropic-version': this.apiVersion,
-          ...finalHeaders
-        },
         agent: proxyAgent,
         timeout: config.requestTimeout || 600000
       }
 
-      // 使用统一 User-Agent 或客户端提供的，最后使用默认值
-      if (!options.headers['user-agent'] || unifiedUA !== null) {
-        const userAgent = unifiedUA || 'claude-cli/1.0.119 (external, cli)'
-        options.headers['user-agent'] = userAgent
-      }
-
-      logger.info(`🔗 指纹是这个: ${options.headers['user-agent']}`)
-      // 根据模型和客户端传递的 anthropic-beta 动态设置 header
-      const modelId = body?.model
-      const clientBetaHeader = clientHeaders?.['anthropic-beta']
-      options.headers['anthropic-beta'] = this._getBetaHeader(modelId, clientBetaHeader)
-
       const req = https.request(options, async (res) => {
         logger.debug(`🌊 Claude stream response status: ${res.statusCode}`)
 
@@ -1465,7 +1676,6 @@ class ClaudeRelayService {
           })
 
           res.on('end', () => {
-            console.error(': ❌ ', errorData)
             logger.error(
               `❌ Claude API error response (Account: ${account?.name || accountId}):`,
               errorData
@@ -1527,6 +1737,16 @@ class ClaudeRelayService {
           return
         }
 
+        // 📬 收到成功响应头（HTTP 200），立即调用回调释放队列锁
+        // 此时请求已被 Claude API 接受并计入 RPM 配额，无需等待响应完成
+        if (onResponseStart && typeof onResponseStart === 'function') {
+          try {
+            await onResponseStart()
+          } catch (callbackError) {
+            logger.error('❌ Error in onResponseStart callback:', callbackError.message)
+          }
+        }
+
         let buffer = ''
         const allUsageData = [] // 收集所有的usage事件
         let currentUsageData = {} // 当前正在收集的usage数据
@@ -1766,15 +1986,15 @@ class ClaudeRelayService {
 
           // 提取5小时会话窗口状态
           // 使用大小写不敏感的方式获取响应头
-          const get5hStatus = (headers) => {
+          const get5hStatus = (resHeaders) => {
-            if (!headers) {
+            if (!resHeaders) {
               return null
             }
             // HTTP头部名称不区分大小写，需要处理不同情况
             return (
-              headers['anthropic-ratelimit-unified-5h-status'] ||
+              resHeaders['anthropic-ratelimit-unified-5h-status'] ||
-              headers['Anthropic-Ratelimit-Unified-5h-Status'] ||
+              resHeaders['Anthropic-Ratelimit-Unified-5h-Status'] ||
-              headers['ANTHROPIC-RATELIMIT-UNIFIED-5H-STATUS']
+              resHeaders['ANTHROPIC-RATELIMIT-UNIFIED-5H-STATUS']
             )
           }
 
@@ -1937,12 +2157,12 @@ class ClaudeRelayService {
       responseStream.on('close', () => {
         logger.debug('🔌 Client disconnected, cleaning up stream')
         if (!req.destroyed) {
-          req.destroy()
+          req.destroy(new Error('Client disconnected'))
         }
       })
 
       // 写入请求体
-      req.write(JSON.stringify(requestPayload))
+      req.write(bodyString)
       req.end()
     })
   }

src/services/droidRelayService.js

@@ -336,7 +336,12 @@ class DroidRelayService {
         )
       }
     } catch (error) {
-      logger.error(`❌ Droid relay error: ${error.message}`, error)
+      // 客户端主动断开连接是正常情况，使用 INFO 级别
+      if (error.message === 'Client disconnected') {
+        logger.info(`🔌 Droid relay ended: Client disconnected`)
+      } else {
+        logger.error(`❌ Droid relay error: ${error.message}`, error)
+      }
 
       const status = error?.response?.status
       if (status >= 400 && status < 500) {
@@ -634,7 +639,7 @@ class DroidRelayService {
       // 客户端断开连接时清理
       clientResponse.on('close', () => {
         if (req && !req.destroyed) {
-          req.destroy()
+          req.destroy(new Error('Client disconnected'))
         }
       })
 

src/services/openaiResponsesRelayService.js

@@ -426,9 +426,9 @@ class OpenAIResponsesRelayService {
       const lines = data.split('\n')
 
       for (const line of lines) {
-        if (line.startsWith('data: ')) {
+        if (line.startsWith('data:')) {
           try {
-            const jsonStr = line.slice(6)
+            const jsonStr = line.slice(5).trim()
             if (jsonStr === '[DONE]') {
               continue
             }

src/services/requestIdentityService.js

@@ -22,6 +22,18 @@ const STAINLESS_HEADER_KEYS = [
   'x-stainless-runtime',
   'x-stainless-runtime-version'
 ]
+
+// 小写 key 到正确大小写格式的映射（用于返回给上游时）
+const STAINLESS_HEADER_CASE_MAP = {
+  'x-stainless-retry-count': 'X-Stainless-Retry-Count',
+  'x-stainless-timeout': 'X-Stainless-Timeout',
+  'x-stainless-lang': 'X-Stainless-Lang',
+  'x-stainless-package-version': 'X-Stainless-Package-Version',
+  'x-stainless-os': 'X-Stainless-OS',
+  'x-stainless-arch': 'X-Stainless-Arch',
+  'x-stainless-runtime': 'X-Stainless-Runtime',
+  'x-stainless-runtime-version': 'X-Stainless-Runtime-Version'
+}
 const MIN_FINGERPRINT_FIELDS = 4
 const REDIS_KEY_PREFIX = 'fmt_claude_req:stainless_headers:'
 
@@ -135,7 +147,9 @@ function applyFingerprintToHeaders(headers, fingerprint) {
       return
     }
     removeHeaderCaseInsensitive(nextHeaders, key)
-    nextHeaders[key] = fingerprint[key]
+    // 使用正确的大小写格式返回给上游
+    const properCaseKey = STAINLESS_HEADER_CASE_MAP[key] || key
+    nextHeaders[properCaseKey] = fingerprint[key]
   })
 
   return nextHeaders

src/services/unifiedClaudeScheduler.js

@@ -180,8 +180,56 @@ class UnifiedClaudeScheduler {
   }
 
   // 🎯 统一调度Claude账号（官方和Console）
-  async selectAccountForApiKey(apiKeyData, sessionHash = null, requestedModel = null) {
+  async selectAccountForApiKey(
+    apiKeyData,
+    sessionHash = null,
+    requestedModel = null,
+    forcedAccount = null
+  ) {
     try {
+      // 🔒 如果有强制绑定的账户（全局会话绑定），仅 claude-official 类型受影响
+      if (forcedAccount && forcedAccount.accountId && forcedAccount.accountType) {
+        // ⚠️ 只有 claude-official 类型账户受全局会话绑定限制
+        // 其他类型（bedrock, ccr, claude-console等）忽略绑定，走正常调度
+        if (forcedAccount.accountType !== 'claude-official') {
+          logger.info(
+            `🔗 Session binding ignored for non-official account type: ${forcedAccount.accountType}, proceeding with normal scheduling`
+          )
+          // 不使用 forcedAccount，继续走下面的正常调度逻辑
+        } else {
+          // claude-official 类型需要检查可用性并强制使用
+          logger.info(
+            `🔗 Forced session binding detected: ${forcedAccount.accountId} (${forcedAccount.accountType})`
+          )
+
+          const isAvailable = await this._isAccountAvailableForSessionBinding(
+            forcedAccount.accountId,
+            forcedAccount.accountType,
+            requestedModel
+          )
+
+          if (isAvailable) {
+            logger.info(
+              `✅ Using forced session binding account: ${forcedAccount.accountId} (${forcedAccount.accountType})`
+            )
+            return {
+              accountId: forcedAccount.accountId,
+              accountType: forcedAccount.accountType
+            }
+          } else {
+            // 绑定账户不可用，抛出特定错误（不 fallback）
+            logger.warn(
+              `❌ Forced session binding account unavailable: ${forcedAccount.accountId} (${forcedAccount.accountType})`
+            )
+            const error = new Error('Session binding account unavailable')
+            error.code = 'SESSION_BINDING_ACCOUNT_UNAVAILABLE'
+            error.accountId = forcedAccount.accountId
+            error.accountType = forcedAccount.accountType
+            throw error
+          }
+        }
+      }
+
       // 解析供应商前缀
       const { vendor, baseModel } = parseVendorPrefixedModel(requestedModel)
       const effectiveModel = vendor === 'ccr' ? baseModel : requestedModel
@@ -1711,6 +1759,67 @@ class UnifiedClaudeScheduler {
       return []
     }
   }
+
+  /**
+   * 🔒 检查 claude-official 账户是否可用于会话绑定
+   * 注意：此方法仅用于 claude-official 类型账户，其他类型不受会话绑定限制
+   * @param {string} accountId - 账户ID
+   * @param {string} accountType - 账户类型（应为 'claude-official'）
+   * @param {string} _requestedModel - 请求的模型（保留参数，当前未使用）
+   * @returns {Promise<boolean>}
+   */
+  async _isAccountAvailableForSessionBinding(accountId, accountType, _requestedModel = null) {
+    try {
+      // 此方法仅处理 claude-official 类型
+      if (accountType !== 'claude-official') {
+        logger.warn(
+          `Session binding: _isAccountAvailableForSessionBinding called for non-official type: ${accountType}`
+        )
+        return true // 非 claude-official 类型不受限制
+      }
+
+      const account = await redis.getClaudeAccount(accountId)
+      if (!account) {
+        logger.warn(`Session binding: Claude OAuth account ${accountId} not found`)
+        return false
+      }
+
+      const isActive = account.isActive === 'true' || account.isActive === true
+      const { status } = account
+
+      if (!isActive) {
+        logger.warn(`Session binding: Claude OAuth account ${accountId} is not active`)
+        return false
+      }
+
+      if (status === 'error' || status === 'temp_error') {
+        logger.warn(
+          `Session binding: Claude OAuth account ${accountId} has error status: ${status}`
+        )
+        return false
+      }
+
+      // 检查是否被限流
+      if (await claudeAccountService.isAccountRateLimited(accountId)) {
+        logger.warn(`Session binding: Claude OAuth account ${accountId} is rate limited`)
+        return false
+      }
+
+      // 检查临时不可用
+      if (await this.isAccountTemporarilyUnavailable(accountId, accountType)) {
+        logger.warn(`Session binding: Claude OAuth account ${accountId} is temporarily unavailable`)
+        return false
+      }
+
+      return true
+    } catch (error) {
+      logger.error(
+        `❌ Error checking account availability for session binding: ${accountId} (${accountType})`,
+        error
+      )
+      return false
+    }
+  }
 }
 
 module.exports = new UnifiedClaudeScheduler()

src/services/userMessageQueueService.js

@@ -0,0 +1,348 @@
+/**
+ * 用户消息队列服务
+ * 为 Claude 账户实现基于消息类型的串行排队机制
+ *
+ * 当请求的最后一条消息是用户输入（role: user）时，
+ * 同一账户的此类请求需要串行等待，并在请求之间添加延迟
+ */
+
+const { v4: uuidv4 } = require('uuid')
+const redis = require('../models/redis')
+const config = require('../../config/config')
+const logger = require('../utils/logger')
+
+// 清理任务间隔
+const CLEANUP_INTERVAL_MS = 60000 // 1分钟
+
+// 轮询等待配置
+const POLL_INTERVAL_BASE_MS = 50 // 基础轮询间隔
+const POLL_INTERVAL_MAX_MS = 500 // 最大轮询间隔
+const POLL_BACKOFF_FACTOR = 1.5 // 退避因子
+
+class UserMessageQueueService {
+  constructor() {
+    this.cleanupTimer = null
+  }
+
+  /**
+   * 检测请求是否为真正的用户消息请求
+   * 区分真正的用户输入和 tool_result 消息
+   *
+   * Claude API 消息格式：
+   * - 用户文本消息: { role: 'user', content: 'text' } 或 { role: 'user', content: [{ type: 'text', text: '...' }] }
+   * - 工具结果消息: { role: 'user', content: [{ type: 'tool_result', tool_use_id: '...', content: '...' }] }
+   *
+   * @param {Object} requestBody - 请求体
+   * @returns {boolean} - 是否为真正的用户消息（排除 tool_result）
+   */
+  isUserMessageRequest(requestBody) {
+    const messages = requestBody?.messages
+    if (!Array.isArray(messages) || messages.length === 0) {
+      return false
+    }
+    const lastMessage = messages[messages.length - 1]
+
+    // 检查 role 是否为 user
+    if (lastMessage?.role !== 'user') {
+      return false
+    }
+
+    // 检查 content 是否包含 tool_result 类型
+    const { content } = lastMessage
+    if (Array.isArray(content)) {
+      // 如果 content 数组中任何元素是 tool_result，则不是真正的用户消息
+      const hasToolResult = content.some(
+        (block) => block?.type === 'tool_result' || block?.type === 'tool_use_result'
+      )
+      if (hasToolResult) {
+        return false
+      }
+    }
+
+    // role 是 user 且不包含 tool_result，是真正的用户消息
+    return true
+  }
+
+  /**
+   * 获取当前配置（支持 Web 界面配置优先）
+   * @returns {Promise<Object>} 配置对象
+   */
+  async getConfig() {
+    // 默认配置（防止 config.userMessageQueue 未定义）
+    // 注意：优化后的默认值 - 锁持有时间从分钟级降到毫秒级，无需长等待
+    const queueConfig = config.userMessageQueue || {}
+    const defaults = {
+      enabled: queueConfig.enabled ?? false,
+      delayMs: queueConfig.delayMs ?? 200,
+      timeoutMs: queueConfig.timeoutMs ?? 5000, // 从 60000 降到 5000，因为锁持有时间短
+      lockTtlMs: queueConfig.lockTtlMs ?? 5000 // 从 120000 降到 5000，5秒足以覆盖请求发送
+    }
+
+    // 尝试从 claudeRelayConfigService 获取 Web 界面配置
+    try {
+      const claudeRelayConfigService = require('./claudeRelayConfigService')
+      const webConfig = await claudeRelayConfigService.getConfig()
+
+      return {
+        enabled:
+          webConfig.userMessageQueueEnabled !== undefined
+            ? webConfig.userMessageQueueEnabled
+            : defaults.enabled,
+        delayMs:
+          webConfig.userMessageQueueDelayMs !== undefined
+            ? webConfig.userMessageQueueDelayMs
+            : defaults.delayMs,
+        timeoutMs:
+          webConfig.userMessageQueueTimeoutMs !== undefined
+            ? webConfig.userMessageQueueTimeoutMs
+            : defaults.timeoutMs,
+        lockTtlMs:
+          webConfig.userMessageQueueLockTtlMs !== undefined
+            ? webConfig.userMessageQueueLockTtlMs
+            : defaults.lockTtlMs
+      }
+    } catch {
+      // 回退到环境变量配置
+      return defaults
+    }
+  }
+
+  /**
+   * 检查功能是否启用
+   * @returns {Promise<boolean>}
+   */
+  async isEnabled() {
+    const cfg = await this.getConfig()
+    return cfg.enabled === true
+  }
+
+  /**
+   * 获取账户队列锁（阻塞等待）
+   * @param {string} accountId - 账户ID
+   * @param {string} requestId - 请求ID（可选，会自动生成）
+   * @param {number} timeoutMs - 超时时间（可选，使用配置默认值）
+   * @returns {Promise<{acquired: boolean, requestId: string, error?: string}>}
+   */
+  async acquireQueueLock(accountId, requestId = null, timeoutMs = null) {
+    const cfg = await this.getConfig()
+
+    if (!cfg.enabled) {
+      return { acquired: true, requestId: requestId || uuidv4(), skipped: true }
+    }
+
+    const reqId = requestId || uuidv4()
+    const timeout = timeoutMs || cfg.timeoutMs
+    const startTime = Date.now()
+    let retryCount = 0
+
+    logger.debug(`📬 User message queue: attempting to acquire lock for account ${accountId}`, {
+      requestId: reqId,
+      timeoutMs: timeout
+    })
+
+    while (Date.now() - startTime < timeout) {
+      const result = await redis.acquireUserMessageLock(
+        accountId,
+        reqId,
+        cfg.lockTtlMs,
+        cfg.delayMs
+      )
+
+      // 检测 Redis 错误，立即返回系统错误而非继续轮询
+      if (result.redisError) {
+        logger.error(`📬 User message queue: Redis error while acquiring lock`, {
+          accountId,
+          requestId: reqId,
+          errorMessage: result.errorMessage
+        })
+        return {
+          acquired: false,
+          requestId: reqId,
+          error: 'queue_backend_error',
+          errorMessage: result.errorMessage
+        }
+      }
+
+      if (result.acquired) {
+        logger.debug(`📬 User message queue: lock acquired for account ${accountId}`, {
+          requestId: reqId,
+          waitedMs: Date.now() - startTime,
+          retries: retryCount
+        })
+        return { acquired: true, requestId: reqId }
+      }
+
+      // 需要等待
+      if (result.waitMs > 0) {
+        // 需要延迟（上一个请求刚完成）
+        await this._sleep(Math.min(result.waitMs, timeout - (Date.now() - startTime)))
+      } else {
+        // 锁被占用，使用指数退避轮询等待
+        const basePollInterval = Math.min(
+          POLL_INTERVAL_BASE_MS * Math.pow(POLL_BACKOFF_FACTOR, retryCount),
+          POLL_INTERVAL_MAX_MS
+        )
+        // 添加 ±15% 随机抖动，避免高并发下的周期性碰撞
+        const jitter = basePollInterval * (0.85 + Math.random() * 0.3)
+        const pollInterval = Math.min(jitter, POLL_INTERVAL_MAX_MS)
+        await this._sleep(pollInterval)
+        retryCount++
+      }
+    }
+
+    // 超时
+    logger.warn(`📬 User message queue: timeout waiting for lock`, {
+      accountId,
+      requestId: reqId,
+      timeoutMs: timeout
+    })
+
+    return {
+      acquired: false,
+      requestId: reqId,
+      error: 'queue_timeout'
+    }
+  }
+
+  /**
+   * 释放账户队列锁
+   * @param {string} accountId - 账户ID
+   * @param {string} requestId - 请求ID
+   * @returns {Promise<boolean>}
+   */
+  async releaseQueueLock(accountId, requestId) {
+    if (!accountId || !requestId) {
+      return false
+    }
+
+    const released = await redis.releaseUserMessageLock(accountId, requestId)
+
+    if (released) {
+      logger.debug(`📬 User message queue: lock released for account ${accountId}`, {
+        requestId
+      })
+    } else {
+      logger.warn(`📬 User message queue: failed to release lock (not owner?)`, {
+        accountId,
+        requestId
+      })
+    }
+
+    return released
+  }
+
+  /**
+   * 获取队列统计信息
+   * @param {string} accountId - 账户ID
+   * @returns {Promise<Object>}
+   */
+  async getQueueStats(accountId) {
+    return await redis.getUserMessageQueueStats(accountId)
+  }
+
+  /**
+   * 服务启动时清理所有残留的队列锁
+   * 防止服务重启后旧锁阻塞新请求
+   * @returns {Promise<number>} 清理的锁数量
+   */
+  async cleanupStaleLocks() {
+    try {
+      const accountIds = await redis.scanUserMessageQueueLocks()
+      let cleanedCount = 0
+
+      for (const accountId of accountIds) {
+        try {
+          await redis.forceReleaseUserMessageLock(accountId)
+          cleanedCount++
+          logger.debug(`📬 User message queue: cleaned stale lock for account ${accountId}`)
+        } catch (error) {
+          logger.error(
+            `📬 User message queue: failed to clean lock for account ${accountId}:`,
+            error
+          )
+        }
+      }
+
+      if (cleanedCount > 0) {
+        logger.info(`📬 User message queue: cleaned ${cleanedCount} stale lock(s) on startup`)
+      }
+
+      return cleanedCount
+    } catch (error) {
+      logger.error('📬 User message queue: failed to cleanup stale locks on startup:', error)
+      return 0
+    }
+  }
+
+  /**
+   * 启动定时清理任务
+   * 始终启动，每次执行时检查配置以支持运行时动态启用/禁用
+   */
+  startCleanupTask() {
+    if (this.cleanupTimer) {
+      return
+    }
+
+    this.cleanupTimer = setInterval(async () => {
+      // 每次运行时检查配置，以便在运行时动态启用/禁用
+      const currentConfig = await this.getConfig()
+      if (!currentConfig.enabled) {
+        logger.debug('📬 User message queue: cleanup skipped (feature disabled)')
+        return
+      }
+      await this._cleanupOrphanLocks()
+    }, CLEANUP_INTERVAL_MS)
+
+    logger.info('📬 User message queue: cleanup task started')
+  }
+
+  /**
+   * 停止定时清理任务
+   */
+  stopCleanupTask() {
+    if (this.cleanupTimer) {
+      clearInterval(this.cleanupTimer)
+      this.cleanupTimer = null
+      logger.info('📬 User message queue: cleanup task stopped')
+    }
+  }
+
+  /**
+   * 清理孤儿锁
+   * 检测异常情况：锁存在但没有设置过期时间（lockTtlRaw === -1）
+   * 正常情况下所有锁都应该有 TTL，Redis 会自动过期
+   * @private
+   */
+  async _cleanupOrphanLocks() {
+    try {
+      const accountIds = await redis.scanUserMessageQueueLocks()
+
+      for (const accountId of accountIds) {
+        const stats = await redis.getUserMessageQueueStats(accountId)
+
+        // 检测异常情况：锁存在（isLocked=true）但没有过期时间（lockTtlRaw=-1）
+        // 正常创建的锁都带有 PX 过期时间，如果没有说明是异常状态
+        if (stats.isLocked && stats.lockTtlRaw === -1) {
+          logger.warn(
+            `📬 User message queue: cleaning up orphan lock without TTL for account ${accountId}`,
+            { lockHolder: stats.lockHolder }
+          )
+          await redis.forceReleaseUserMessageLock(accountId)
+        }
+      }
+    } catch (error) {
+      logger.error('📬 User message queue: cleanup task error:', error)
+    }
+  }
+
+  /**
+   * 睡眠辅助函数
+   * @param {number} ms - 毫秒
+   * @private
+   */
+  _sleep(ms) {
+    return new Promise((resolve) => setTimeout(resolve, ms))
+  }
+}
+
+module.exports = new UserMessageQueueService()

src/utils/contents.js

@@ -84,6 +84,11 @@ const PROMPT_DEFINITIONS = {
     title: 'Claude Code Compact System Prompt',
     text: 'You are a helpful AI assistant tasked with summarizing conversations.'
   },
+  exploreAgentSystemPrompt: {
+    category: 'system',
+    title: 'Claude Code Explore Agent System Prompt',
+    text: "You are a file search specialist for Claude Code, Anthropic's official CLI for Claude."
+  },
   outputStyleInsightsPrompt: {
     category: 'output_style',
     title: 'Output Style Insights Addendum',

src/utils/headerFilter.js

@@ -52,50 +52,38 @@ function filterForOpenAI(headers) {
 
 /**
  * 为 Claude/Anthropic API 过滤 headers
- * 在原有逻辑基础上添加 CDN headers 到敏感列表
+ * 使用白名单模式，只允许指定的 headers 通过
  */
 function filterForClaude(headers) {
-  const sensitiveHeaders = [
+  // 白名单模式：只允许以下 headers
-    'content-type',
+  const allowedHeaders = [
-    'user-agent',
-    'x-api-key',
-    'authorization',
-    'x-authorization',
-    'host',
-    'content-length',
-    'connection',
-    'proxy-authorization',
-    'content-encoding',
-    'transfer-encoding',
-    ...cdnHeaders // 添加 CDN headers
-  ]
-
-  const browserHeaders = [
-    'origin',
-    'referer',
-    'sec-fetch-mode',
-    'sec-fetch-site',
-    'sec-fetch-dest',
-    'sec-ch-ua',
-    'sec-ch-ua-mobile',
-    'sec-ch-ua-platform',
-    'accept-language',
-    'accept-encoding',
     'accept',
-    'cache-control',
+    'x-stainless-retry-count',
-    'pragma',
+    'x-stainless-timeout',
-    'anthropic-dangerous-direct-browser-access'
+    'x-stainless-lang',
+    'x-stainless-package-version',
+    'x-stainless-os',
+    'x-stainless-arch',
+    'x-stainless-runtime',
+    'x-stainless-runtime-version',
+    'x-stainless-helper-method',
+    'anthropic-dangerous-direct-browser-access',
+    'anthropic-version',
+    'x-app',
+    'anthropic-beta',
+    'accept-language',
+    'sec-fetch-mode',
+    'accept-encoding',
+    'user-agent',
+    'content-type',
+    'connection'
   ]
 
-  const allowedHeaders = ['x-request-id', 'anthropic-version', 'anthropic-beta']
-
   const filtered = {}
   Object.keys(headers || {}).forEach((key) => {
     const lowerKey = key.toLowerCase()
     if (allowedHeaders.includes(lowerKey)) {
       filtered[key] = headers[key]
-    } else if (!sensitiveHeaders.includes(lowerKey) && !browserHeaders.includes(lowerKey)) {
-      filtered[key] = headers[key]
     }
   })
 

src/utils/logger.js

@@ -137,6 +137,7 @@ const createLogFormat = (colorize = false) => {
 
 const logFormat = createLogFormat(false)
 const consoleFormat = createLogFormat(true)
+const isTestEnv = process.env.NODE_ENV === 'test' || process.env.JEST_WORKER_ID
 
 // 📁 确保日志目录存在并设置权限
 if (!fs.existsSync(config.logging.dirname)) {
@@ -159,18 +160,20 @@ const createRotateTransport = (filename, level = null) => {
     transport.level = level
   }
 
-  // 监听轮转事件
+  // 监听轮转事件（测试环境关闭以避免 Jest 退出后输出）
-  transport.on('rotate', (oldFilename, newFilename) => {
+  if (!isTestEnv) {
-    console.log(`📦 Log rotated: ${oldFilename} -> ${newFilename}`)
+    transport.on('rotate', (oldFilename, newFilename) => {
-  })
+      console.log(`📦 Log rotated: ${oldFilename} -> ${newFilename}`)
+    })
 
-  transport.on('new', (newFilename) => {
+    transport.on('new', (newFilename) => {
-    console.log(`📄 New log file created: ${newFilename}`)
+      console.log(`📄 New log file created: ${newFilename}`)
-  })
+    })
 
-  transport.on('archive', (zipFilename) => {
+    transport.on('archive', (zipFilename) => {
-    console.log(`🗜️ Log archived: ${zipFilename}`)
+      console.log(`🗜️ Log archived: ${zipFilename}`)
-  })
+    })
+  }
 
   return transport
 }

src/validators/clients/claudeCodeValidator.js

@@ -65,8 +65,9 @@ class ClaudeCodeValidator {
       const { bestScore } = bestSimilarityByTemplates(rawText)
       if (bestScore < threshold) {
         logger.error(
-          `Claude system prompt similarity below threshold: score=${bestScore.toFixed(4)}, threshold=${threshold}, prompt=${rawText}`
+          `Claude system prompt similarity below threshold: score=${bestScore.toFixed(4)}, threshold=${threshold}`
         )
+        logger.warn(`Claude system prompt detail: ${rawText}`)
         return false
       }
     }

tests/userMessageQueue.test.js

@@ -0,0 +1,434 @@
+/**
+ * 用户消息队列服务测试
+ * 测试消息类型检测、队列串行行为、延迟间隔、超时处理和功能开关
+ */
+
+const redis = require('../src/models/redis')
+const userMessageQueueService = require('../src/services/userMessageQueueService')
+
+describe('UserMessageQueueService', () => {
+  describe('isUserMessageRequest', () => {
+    it('should return true when last message role is user', () => {
+      const requestBody = {
+        messages: [
+          { role: 'user', content: 'Hello' },
+          { role: 'assistant', content: 'Hi there' },
+          { role: 'user', content: 'How are you?' }
+        ]
+      }
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(true)
+    })
+
+    it('should return false when last message role is assistant', () => {
+      const requestBody = {
+        messages: [
+          { role: 'user', content: 'Hello' },
+          { role: 'assistant', content: 'Hi there' }
+        ]
+      }
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(false)
+    })
+
+    it('should return false when last message contains tool_result', () => {
+      const requestBody = {
+        messages: [
+          { role: 'user', content: 'Hello' },
+          { role: 'assistant', content: 'Let me check that' },
+          {
+            role: 'user',
+            content: [
+              {
+                type: 'tool_result',
+                tool_use_id: 'test-id',
+                content: 'Tool result'
+              }
+            ]
+          }
+        ]
+      }
+      // tool_result 消息虽然 role 是 user，但不是真正的用户消息
+      // 应该返回 false，不进入用户消息队列
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(false)
+    })
+
+    it('should return false when last message contains multiple tool_results', () => {
+      const requestBody = {
+        messages: [
+          { role: 'user', content: 'Run multiple tools' },
+          {
+            role: 'user',
+            content: [
+              {
+                type: 'tool_result',
+                tool_use_id: 'tool-1',
+                content: 'Result 1'
+              },
+              {
+                type: 'tool_result',
+                tool_use_id: 'tool-2',
+                content: 'Result 2'
+              }
+            ]
+          }
+        ]
+      }
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(false)
+    })
+
+    it('should return true when user message has array content with text type', () => {
+      const requestBody = {
+        messages: [
+          {
+            role: 'user',
+            content: [
+              {
+                type: 'text',
+                text: 'Hello, this is a user message'
+              }
+            ]
+          }
+        ]
+      }
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(true)
+    })
+
+    it('should return true when user message has mixed text and image content', () => {
+      const requestBody = {
+        messages: [
+          {
+            role: 'user',
+            content: [
+              {
+                type: 'text',
+                text: 'What is in this image?'
+              },
+              {
+                type: 'image',
+                source: { type: 'base64', media_type: 'image/png', data: '...' }
+              }
+            ]
+          }
+        ]
+      }
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(true)
+    })
+
+    it('should return false when messages is empty', () => {
+      const requestBody = { messages: [] }
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(false)
+    })
+
+    it('should return false when messages is not an array', () => {
+      const requestBody = { messages: 'not an array' }
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(false)
+    })
+
+    it('should return false when messages is undefined', () => {
+      const requestBody = {}
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(false)
+    })
+
+    it('should return false when requestBody is null', () => {
+      expect(userMessageQueueService.isUserMessageRequest(null)).toBe(false)
+    })
+
+    it('should return false when requestBody is undefined', () => {
+      expect(userMessageQueueService.isUserMessageRequest(undefined)).toBe(false)
+    })
+
+    it('should return false when last message has no role', () => {
+      const requestBody = {
+        messages: [{ content: 'Hello' }]
+      }
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(false)
+    })
+
+    it('should handle single user message', () => {
+      const requestBody = {
+        messages: [{ role: 'user', content: 'Hello' }]
+      }
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(true)
+    })
+
+    it('should handle single assistant message', () => {
+      const requestBody = {
+        messages: [{ role: 'assistant', content: 'Hello' }]
+      }
+      expect(userMessageQueueService.isUserMessageRequest(requestBody)).toBe(false)
+    })
+  })
+
+  describe('getConfig', () => {
+    it('should return config with expected properties', async () => {
+      const config = await userMessageQueueService.getConfig()
+      expect(config).toHaveProperty('enabled')
+      expect(config).toHaveProperty('delayMs')
+      expect(config).toHaveProperty('timeoutMs')
+      expect(config).toHaveProperty('lockTtlMs')
+      expect(typeof config.enabled).toBe('boolean')
+      expect(typeof config.delayMs).toBe('number')
+      expect(typeof config.timeoutMs).toBe('number')
+      expect(typeof config.lockTtlMs).toBe('number')
+    })
+  })
+
+  describe('isEnabled', () => {
+    it('should return boolean', async () => {
+      const enabled = await userMessageQueueService.isEnabled()
+      expect(typeof enabled).toBe('boolean')
+    })
+  })
+
+  describe('acquireQueueLock', () => {
+    afterEach(() => {
+      jest.restoreAllMocks()
+    })
+
+    it('should acquire lock immediately when no lock exists', async () => {
+      jest.spyOn(userMessageQueueService, 'getConfig').mockResolvedValue({
+        enabled: true,
+        delayMs: 200,
+        timeoutMs: 30000,
+        lockTtlMs: 120000
+      })
+      jest.spyOn(redis, 'acquireUserMessageLock').mockResolvedValue({
+        acquired: true,
+        waitMs: 0
+      })
+
+      const result = await userMessageQueueService.acquireQueueLock('acct-1', 'req-1')
+
+      expect(result.acquired).toBe(true)
+      expect(result.requestId).toBe('req-1')
+      expect(result.error).toBeUndefined()
+    })
+
+    it('should skip lock acquisition when queue disabled', async () => {
+      jest.spyOn(userMessageQueueService, 'getConfig').mockResolvedValue({
+        enabled: false,
+        delayMs: 200,
+        timeoutMs: 30000,
+        lockTtlMs: 120000
+      })
+      const acquireSpy = jest.spyOn(redis, 'acquireUserMessageLock')
+
+      const result = await userMessageQueueService.acquireQueueLock('acct-1')
+
+      expect(result.acquired).toBe(true)
+      expect(result.skipped).toBe(true)
+      expect(acquireSpy).not.toHaveBeenCalled()
+    })
+
+    it('should generate requestId when not provided', async () => {
+      jest.spyOn(userMessageQueueService, 'getConfig').mockResolvedValue({
+        enabled: true,
+        delayMs: 200,
+        timeoutMs: 30000,
+        lockTtlMs: 120000
+      })
+      jest.spyOn(redis, 'acquireUserMessageLock').mockResolvedValue({
+        acquired: true,
+        waitMs: 0
+      })
+
+      const result = await userMessageQueueService.acquireQueueLock('acct-1')
+
+      expect(result.acquired).toBe(true)
+      expect(result.requestId).toBeDefined()
+      expect(result.requestId.length).toBeGreaterThan(0)
+    })
+
+    it('should wait and retry when lock is held by another request', async () => {
+      jest.spyOn(userMessageQueueService, 'getConfig').mockResolvedValue({
+        enabled: true,
+        delayMs: 200,
+        timeoutMs: 1000,
+        lockTtlMs: 120000
+      })
+
+      let callCount = 0
+      jest.spyOn(redis, 'acquireUserMessageLock').mockImplementation(async () => {
+        callCount++
+        if (callCount < 3) {
+          return { acquired: false, waitMs: -1 } // lock held
+        }
+        return { acquired: true, waitMs: 0 }
+      })
+
+      // Mock sleep to speed up test
+      jest.spyOn(userMessageQueueService, '_sleep').mockResolvedValue(undefined)
+
+      const result = await userMessageQueueService.acquireQueueLock('acct-1', 'req-1')
+
+      expect(result.acquired).toBe(true)
+      expect(callCount).toBe(3)
+    })
+
+    it('should respect delay when previous request just completed', async () => {
+      jest.spyOn(userMessageQueueService, 'getConfig').mockResolvedValue({
+        enabled: true,
+        delayMs: 200,
+        timeoutMs: 1000,
+        lockTtlMs: 120000
+      })
+
+      let callCount = 0
+      jest.spyOn(redis, 'acquireUserMessageLock').mockImplementation(async () => {
+        callCount++
+        if (callCount === 1) {
+          return { acquired: false, waitMs: 150 } // need to wait 150ms for delay
+        }
+        return { acquired: true, waitMs: 0 }
+      })
+
+      const sleepSpy = jest.spyOn(userMessageQueueService, '_sleep').mockResolvedValue(undefined)
+
+      const result = await userMessageQueueService.acquireQueueLock('acct-1', 'req-1')
+
+      expect(result.acquired).toBe(true)
+      expect(sleepSpy).toHaveBeenCalledWith(150) // Should wait for delay
+    })
+
+    it('should timeout and return error when wait exceeds timeout', async () => {
+      jest.spyOn(userMessageQueueService, 'getConfig').mockResolvedValue({
+        enabled: true,
+        delayMs: 200,
+        timeoutMs: 100, // very short timeout
+        lockTtlMs: 120000
+      })
+
+      jest.spyOn(redis, 'acquireUserMessageLock').mockResolvedValue({
+        acquired: false,
+        waitMs: -1 // always held
+      })
+
+      // Use real timers for timeout test but mock sleep to be instant
+      jest.spyOn(userMessageQueueService, '_sleep').mockImplementation(async () => {
+        // Simulate time passing
+        await new Promise((resolve) => setTimeout(resolve, 60))
+      })
+
+      const result = await userMessageQueueService.acquireQueueLock('acct-1', 'req-1', 100)
+
+      expect(result.acquired).toBe(false)
+      expect(result.error).toBe('queue_timeout')
+    })
+  })
+
+  describe('releaseQueueLock', () => {
+    afterEach(() => {
+      jest.restoreAllMocks()
+    })
+
+    it('should release lock successfully when holding the lock', async () => {
+      jest.spyOn(redis, 'releaseUserMessageLock').mockResolvedValue(true)
+
+      const result = await userMessageQueueService.releaseQueueLock('acct-1', 'req-1')
+
+      expect(result).toBe(true)
+      expect(redis.releaseUserMessageLock).toHaveBeenCalledWith('acct-1', 'req-1')
+    })
+
+    it('should return false when not holding the lock', async () => {
+      jest.spyOn(redis, 'releaseUserMessageLock').mockResolvedValue(false)
+
+      const result = await userMessageQueueService.releaseQueueLock('acct-1', 'req-1')
+
+      expect(result).toBe(false)
+    })
+
+    it('should return false when accountId is missing', async () => {
+      const releaseSpy = jest.spyOn(redis, 'releaseUserMessageLock')
+
+      const result = await userMessageQueueService.releaseQueueLock(null, 'req-1')
+
+      expect(result).toBe(false)
+      expect(releaseSpy).not.toHaveBeenCalled()
+    })
+
+    it('should return false when requestId is missing', async () => {
+      const releaseSpy = jest.spyOn(redis, 'releaseUserMessageLock')
+
+      const result = await userMessageQueueService.releaseQueueLock('acct-1', null)
+
+      expect(result).toBe(false)
+      expect(releaseSpy).not.toHaveBeenCalled()
+    })
+  })
+
+  describe('queue serialization behavior', () => {
+    afterEach(() => {
+      jest.restoreAllMocks()
+    })
+
+    it('should allow different accounts to acquire locks simultaneously', async () => {
+      jest.spyOn(userMessageQueueService, 'getConfig').mockResolvedValue({
+        enabled: true,
+        delayMs: 200,
+        timeoutMs: 30000,
+        lockTtlMs: 120000
+      })
+      jest.spyOn(redis, 'acquireUserMessageLock').mockResolvedValue({
+        acquired: true,
+        waitMs: 0
+      })
+
+      const [result1, result2] = await Promise.all([
+        userMessageQueueService.acquireQueueLock('acct-1', 'req-1'),
+        userMessageQueueService.acquireQueueLock('acct-2', 'req-2')
+      ])
+
+      expect(result1.acquired).toBe(true)
+      expect(result2.acquired).toBe(true)
+    })
+
+    it('should serialize requests for same account', async () => {
+      jest.spyOn(userMessageQueueService, 'getConfig').mockResolvedValue({
+        enabled: true,
+        delayMs: 50,
+        timeoutMs: 5000,
+        lockTtlMs: 120000
+      })
+
+      const lockState = { held: false, holderId: null }
+
+      jest
+        .spyOn(redis, 'acquireUserMessageLock')
+        .mockImplementation(async (accountId, requestId) => {
+          if (!lockState.held) {
+            lockState.held = true
+            lockState.holderId = requestId
+            return { acquired: true, waitMs: 0 }
+          }
+          return { acquired: false, waitMs: -1 }
+        })
+
+      jest
+        .spyOn(redis, 'releaseUserMessageLock')
+        .mockImplementation(async (accountId, requestId) => {
+          if (lockState.holderId === requestId) {
+            lockState.held = false
+            lockState.holderId = null
+            return true
+          }
+          return false
+        })
+
+      jest.spyOn(userMessageQueueService, '_sleep').mockResolvedValue(undefined)
+
+      // First request acquires lock
+      const result1 = await userMessageQueueService.acquireQueueLock('acct-1', 'req-1')
+      expect(result1.acquired).toBe(true)
+
+      // Second request should fail to acquire (lock held)
+      const acquirePromise = userMessageQueueService.acquireQueueLock('acct-1', 'req-2', 200)
+
+      // Release first lock
+      await userMessageQueueService.releaseQueueLock('acct-1', 'req-1')
+
+      // Now second request should acquire
+      const result2 = await acquirePromise
+      expect(result2.acquired).toBe(true)
+    })
+  })
+})

web/admin-spa/package-lock.json

@@ -1157,7 +1157,6 @@
       "resolved": "https://registry.npmmirror.com/@types/lodash-es/-/lodash-es-4.17.12.tgz",
       "integrity": "sha512-0NgftHUcV4v34VhXm8QBSftKVXtbkBG3ViCjs6+eJ5a6y6Mi/jiFGPc1sC7QK+9BFhWrURE3EOggmWaSxL9OzQ==",
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "@types/lodash": "*"
       }
@@ -1352,7 +1351,6 @@
       "integrity": "sha512-NZyJarBfL7nWwIq+FDL6Zp/yHEhePMNnnJ0y3qfieCrmNvYct8uvtiV41UvlSe6apAfk0fY1FbWx+NwfmpvtTg==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "bin": {
         "acorn": "bin/acorn"
       },
@@ -1589,7 +1587,6 @@
         }
       ],
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "caniuse-lite": "^1.0.30001726",
         "electron-to-chromium": "^1.5.173",
@@ -3063,15 +3060,13 @@
       "version": "4.17.21",
       "resolved": "https://registry.npmmirror.com/lodash/-/lodash-4.17.21.tgz",
       "integrity": "sha512-v2kDEe57lecTulaDIuNTPy3Ry4gLGJ6Z1O3vE1krgXZNrsQ+LFTGHVxVjcXPs17LhbZVGedAJv8XZ1tvj5FvSg==",
-      "license": "MIT",
+      "license": "MIT"
-      "peer": true
     },
     "node_modules/lodash-es": {
       "version": "4.17.21",
       "resolved": "https://registry.npmmirror.com/lodash-es/-/lodash-es-4.17.21.tgz",
       "integrity": "sha512-mKnC+QJ9pWVzv+C4/U3rRsHapFfHvQFoFB92e52xeyGMcX6/OlIl78je1u8vePzYZSkkogMPJ2yjxxsb89cxyw==",
-      "license": "MIT",
+      "license": "MIT"
-      "peer": true
     },
     "node_modules/lodash-unified": {
       "version": "1.0.3",
@@ -3623,7 +3618,6 @@
         }
       ],
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "nanoid": "^3.3.11",
         "picocolors": "^1.1.1",
@@ -3770,7 +3764,6 @@
       "integrity": "sha512-I7AIg5boAr5R0FFtJ6rCfD+LFsWHp81dolrFD8S79U9tb8Az2nGrJncnMSnys+bpQJfRUzqs9hnA81OAA3hCuQ==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "bin": {
         "prettier": "bin/prettier.cjs"
       },
@@ -4035,7 +4028,6 @@
       "integrity": "sha512-33xGNBsDJAkzt0PvninskHlWnTIPgDtTwhg0U38CUoNP/7H6wI2Cz6dUeoNPbjdTdsYTGuiFFASuUOWovH0SyQ==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "@types/estree": "1.0.8"
       },
@@ -4533,7 +4525,6 @@
       "integrity": "sha512-5gTmgEY/sqK6gFXLIsQNH19lWb4ebPDLA4SdLP7dsWkIXHWlG66oPuVvXSGFPppYZz8ZDZq0dYYrbHfBCVUb1Q==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "engines": {
         "node": ">=12"
       },
@@ -4924,7 +4915,6 @@
       "integrity": "sha512-qO3aKv3HoQC8QKiNSTuUM1l9o/XX3+c+VTgLHbJWHZGeTPVAg2XwazI9UWzoxjIJCGCV2zU60uqMzjeLZuULqA==",
       "dev": true,
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "esbuild": "^0.21.3",
         "postcss": "^8.4.43",
@@ -5125,7 +5115,6 @@
       "resolved": "https://registry.npmmirror.com/vue/-/vue-3.5.18.tgz",
       "integrity": "sha512-7W4Y4ZbMiQ3SEo+m9lnoNpV9xG7QVMLa+/0RFwwiAVkeYoyGXqWE85jabU4pllJNUzqfLShJ5YLptewhCWUgNA==",
       "license": "MIT",
-      "peer": true,
       "dependencies": {
         "@vue/compiler-dom": "3.5.18",
         "@vue/compiler-sfc": "3.5.18",

web/admin-spa/src/views/AccountsView.vue

@@ -2049,7 +2049,7 @@ const bindingCounts = ref({}) // 轻量级绑定计数，用于显示"绑定: X
 const accountGroups = ref([])
 const groupFilter = ref('all')
 const platformFilter = ref('all')
-const statusFilter = ref('normal') // 状态过滤 (normal/rateLimited/other/all)
+const statusFilter = ref('all') // 状态过滤 (normal/rateLimited/other/all)
 const searchKeyword = ref('')
 const PAGE_SIZE_STORAGE_KEY = 'accountsPageSize'
 const getInitialPageSize = () => {
@@ -2804,11 +2804,12 @@ const loadAccounts = async (forceReload = false) => {
     let openaiResponsesRaw = []
 
     const appendAccounts = (platform, data) => {
-      if (!data || data.length === 0) return
+      const list = Array.isArray(data) ? data : []
+      if (list.length === 0) return
 
       switch (platform) {
         case 'claude': {
-          const items = data.map((acc) => {
+          const items = list.map((acc) => {
             const boundApiKeysCount = counts.claudeAccountId?.[acc.id] || 0
             return { ...acc, platform: 'claude', boundApiKeysCount }
           })
@@ -2816,7 +2817,7 @@ const loadAccounts = async (forceReload = false) => {
           break
         }
         case 'claude-console': {
-          const items = data.map((acc) => {
+          const items = list.map((acc) => {
             const boundApiKeysCount = counts.claudeConsoleAccountId?.[acc.id] || 0
             return { ...acc, platform: 'claude-console', boundApiKeysCount }
           })
@@ -2824,12 +2825,12 @@ const loadAccounts = async (forceReload = false) => {
           break
         }
         case 'bedrock': {
-          const items = data.map((acc) => ({ ...acc, platform: 'bedrock', boundApiKeysCount: 0 }))
+          const items = list.map((acc) => ({ ...acc, platform: 'bedrock', boundApiKeysCount: 0 }))
           allAccounts.push(...items)
           break
         }
         case 'gemini': {
-          const items = data.map((acc) => {
+          const items = list.map((acc) => {
             const boundApiKeysCount = counts.geminiAccountId?.[acc.id] || 0
             return { ...acc, platform: 'gemini', boundApiKeysCount }
           })
@@ -2837,7 +2838,7 @@ const loadAccounts = async (forceReload = false) => {
           break
         }
         case 'openai': {
-          const items = data.map((acc) => {
+          const items = list.map((acc) => {
             const boundApiKeysCount = counts.openaiAccountId?.[acc.id] || 0
             return { ...acc, platform: 'openai', boundApiKeysCount }
           })
@@ -2845,7 +2846,7 @@ const loadAccounts = async (forceReload = false) => {
           break
         }
         case 'azure_openai': {
-          const items = data.map((acc) => {
+          const items = list.map((acc) => {
             const boundApiKeysCount = counts.azureOpenaiAccountId?.[acc.id] || 0
             return { ...acc, platform: 'azure_openai', boundApiKeysCount }
           })
@@ -2853,16 +2854,16 @@ const loadAccounts = async (forceReload = false) => {
           break
         }
         case 'openai-responses': {
-          openaiResponsesRaw = data
+          openaiResponsesRaw = list
           break
         }
         case 'ccr': {
-          const items = data.map((acc) => ({ ...acc, platform: 'ccr', boundApiKeysCount: 0 }))
+          const items = list.map((acc) => ({ ...acc, platform: 'ccr', boundApiKeysCount: 0 }))
           allAccounts.push(...items)
           break
         }
         case 'droid': {
-          const items = data.map((acc) => {
+          const items = list.map((acc) => {
             const boundApiKeysCount = counts.droidAccountId?.[acc.id] || acc.boundApiKeysCount || 0
             return { ...acc, platform: 'droid', boundApiKeysCount }
           })
@@ -2870,7 +2871,7 @@ const loadAccounts = async (forceReload = false) => {
           break
         }
         case 'gemini-api': {
-          const items = data.map((acc) => {
+          const items = list.map((acc) => {
             const boundApiKeysCount = counts.geminiAccountId?.[`api:${acc.id}`] || 0
             return { ...acc, platform: 'gemini-api', boundApiKeysCount }
           })

web/admin-spa/src/views/SettingsView.vue

@@ -36,6 +36,18 @@
             <i class="fas fa-bell mr-2"></i>
             通知设置
           </button>
+          <button
+            :class="[
+              'border-b-2 pb-2 text-sm font-medium transition-colors',
+              activeSection === 'claude'
+                ? 'border-blue-500 text-blue-600 dark:border-blue-400 dark:text-blue-400'
+                : 'border-transparent text-gray-500 hover:border-gray-300 hover:text-gray-700 dark:text-gray-400 dark:hover:text-gray-300'
+            ]"
+            @click="activeSection = 'claude'"
+          >
+            <i class="fas fa-robot mr-2"></i>
+            Claude 转发
+          </button>
         </nav>
       </div>
 
@@ -629,6 +641,276 @@
             </button>
           </div>
         </div>
+
+        <!-- Claude 转发配置部分 -->
+        <div v-show="activeSection === 'claude'">
+          <!-- 加载状态 -->
+          <div v-if="claudeConfigLoading" class="py-12 text-center">
+            <div class="loading-spinner mx-auto mb-4"></div>
+            <p class="text-gray-500 dark:text-gray-400">正在加载配置...</p>
+          </div>
+
+          <div v-else>
+            <!-- Claude Code 客户端限制 -->
+            <div
+              class="mb-6 rounded-lg bg-white/80 p-6 shadow-lg backdrop-blur-sm dark:bg-gray-800/80"
+            >
+              <div class="flex items-center justify-between">
+                <div>
+                  <div class="flex items-center">
+                    <div
+                      class="mr-3 flex h-10 w-10 items-center justify-center rounded-full bg-gradient-to-br from-orange-500 to-amber-600 text-white shadow-lg"
+                    >
+                      <i class="fas fa-terminal"></i>
+                    </div>
+                    <div>
+                      <h2 class="text-lg font-semibold text-gray-800 dark:text-gray-200">
+                        仅允许 Claude Code 客户端
+                      </h2>
+                      <p class="mt-1 text-sm text-gray-600 dark:text-gray-400">
+                        启用后，所有
+                        <code class="rounded bg-gray-100 px-1 dark:bg-gray-700"
+                          >/api/v1/messages</code
+                        >
+                        和
+                        <code class="rounded bg-gray-100 px-1 dark:bg-gray-700"
+                          >/claude/v1/messages</code
+                        >
+                        端点将强制验证 Claude Code CLI 客户端
+                      </p>
+                    </div>
+                  </div>
+                </div>
+                <label class="relative inline-flex cursor-pointer items-center">
+                  <input
+                    v-model="claudeConfig.claudeCodeOnlyEnabled"
+                    class="peer sr-only"
+                    type="checkbox"
+                    @change="saveClaudeConfig"
+                  />
+                  <div
+                    class="peer h-6 w-11 rounded-full bg-gray-200 after:absolute after:left-[2px] after:top-[2px] after:h-5 after:w-5 after:rounded-full after:border after:border-gray-300 after:bg-white after:transition-all after:content-[''] peer-checked:bg-orange-500 peer-checked:after:translate-x-full peer-checked:after:border-white peer-focus:outline-none peer-focus:ring-4 peer-focus:ring-orange-300 dark:border-gray-600 dark:bg-gray-700 dark:peer-focus:ring-orange-800"
+                  ></div>
+                </label>
+              </div>
+              <div class="mt-4 rounded-lg bg-amber-50 p-4 dark:bg-amber-900/20">
+                <div class="flex">
+                  <i class="fas fa-info-circle mt-0.5 text-amber-500"></i>
+                  <div class="ml-3">
+                    <p class="text-sm text-amber-700 dark:text-amber-300">
+                      此设置与 API Key 级别的客户端限制是 <strong>OR 逻辑</strong>：全局启用或 API
+                      Key 设置中启用，都会执行 Claude Code 验证。
+                    </p>
+                  </div>
+                </div>
+              </div>
+            </div>
+
+            <!-- 全局会话绑定 -->
+            <div
+              class="mb-6 rounded-lg bg-white/80 p-6 shadow-lg backdrop-blur-sm dark:bg-gray-800/80"
+            >
+              <div class="flex items-center justify-between">
+                <div>
+                  <div class="flex items-center">
+                    <div
+                      class="mr-3 flex h-10 w-10 items-center justify-center rounded-full bg-gradient-to-br from-purple-500 to-indigo-600 text-white shadow-lg"
+                    >
+                      <i class="fas fa-link"></i>
+                    </div>
+                    <div>
+                      <h2 class="text-lg font-semibold text-gray-800 dark:text-gray-200">
+                        强制会话绑定
+                      </h2>
+                      <p class="mt-1 text-sm text-gray-600 dark:text-gray-400">
+                        启用后，系统会将原始会话 ID 绑定到首次使用的账户，确保上下文的一致性
+                      </p>
+                    </div>
+                  </div>
+                </div>
+                <label class="relative inline-flex cursor-pointer items-center">
+                  <input
+                    v-model="claudeConfig.globalSessionBindingEnabled"
+                    class="peer sr-only"
+                    type="checkbox"
+                    @change="saveClaudeConfig"
+                  />
+                  <div
+                    class="peer h-6 w-11 rounded-full bg-gray-200 after:absolute after:left-[2px] after:top-[2px] after:h-5 after:w-5 after:rounded-full after:border after:border-gray-300 after:bg-white after:transition-all after:content-[''] peer-checked:bg-purple-500 peer-checked:after:translate-x-full peer-checked:after:border-white peer-focus:outline-none peer-focus:ring-4 peer-focus:ring-purple-300 dark:border-gray-600 dark:bg-gray-700 dark:peer-focus:ring-purple-800"
+                  ></div>
+                </label>
+              </div>
+
+              <!-- 绑定配置详情（仅在启用时显示） -->
+              <div v-if="claudeConfig.globalSessionBindingEnabled" class="mt-6 space-y-4">
+                <!-- 绑定有效期 -->
+                <div>
+                  <label class="block text-sm font-medium text-gray-700 dark:text-gray-300">
+                    <i class="fas fa-clock mr-2 text-gray-400"></i>
+                    绑定有效期（天）
+                  </label>
+                  <input
+                    v-model.number="claudeConfig.sessionBindingTtlDays"
+                    class="mt-1 block w-full max-w-xs rounded-lg border border-gray-300 bg-white px-3 py-2 shadow-sm focus:border-purple-500 focus:outline-none focus:ring-2 focus:ring-purple-500/20 dark:border-gray-500 dark:bg-gray-700 dark:text-white sm:text-sm"
+                    max="365"
+                    min="1"
+                    placeholder="30"
+                    type="number"
+                    @change="saveClaudeConfig"
+                  />
+                  <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                    会话绑定到账户后的有效时间，过期后会自动解除绑定
+                  </p>
+                </div>
+
+                <!-- 错误提示消息 -->
+                <div>
+                  <label class="block text-sm font-medium text-gray-700 dark:text-gray-300">
+                    <i class="fas fa-exclamation-triangle mr-2 text-gray-400"></i>
+                    旧会话污染提示
+                  </label>
+                  <textarea
+                    v-model="claudeConfig.sessionBindingErrorMessage"
+                    class="mt-1 block w-full rounded-lg border border-gray-300 bg-white px-3 py-2 shadow-sm focus:border-purple-500 focus:outline-none focus:ring-2 focus:ring-purple-500/20 dark:border-gray-500 dark:bg-gray-700 dark:text-white sm:text-sm"
+                    placeholder="你的本地session已污染，请清理后使用。"
+                    rows="2"
+                    @change="saveClaudeConfig"
+                  ></textarea>
+                  <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                    当检测到为旧的sessionId且未在系统中有调度记录时提示，返回给客户端的错误消息
+                  </p>
+                </div>
+              </div>
+
+              <div class="mt-4 rounded-lg bg-purple-50 p-4 dark:bg-purple-900/20">
+                <div class="flex">
+                  <i class="fas fa-lightbulb mt-0.5 text-purple-500"></i>
+                  <div class="ml-3">
+                    <p class="text-sm text-purple-700 dark:text-purple-300">
+                      <strong>工作原理：</strong>系统会提取请求中的原始 session ID （来自
+                      <code class="rounded bg-purple-100 px-1 dark:bg-purple-800"
+                        >metadata.user_id</code
+                      >）， 并将其与首次调度的账户绑定。后续使用相同 session ID
+                      的请求将自动路由到同一账户。
+                    </p>
+                    <p class="mt-2 text-sm text-purple-700 dark:text-purple-300">
+                      <strong>新会话识别：</strong>如果绑定会话历史中没有该sessionId但请求中
+                      <code class="rounded bg-purple-100 px-1 dark:bg-purple-800"
+                        >messages.length > 1</code
+                      >， 系统会认为这是一个污染的会话并拒绝请求。
+                    </p>
+                  </div>
+                </div>
+              </div>
+            </div>
+
+            <!-- 用户消息串行队列 -->
+            <div
+              class="mb-6 rounded-lg bg-white/80 p-6 shadow-lg backdrop-blur-sm dark:bg-gray-800/80"
+            >
+              <div class="flex items-center justify-between">
+                <div>
+                  <div class="flex items-center">
+                    <div
+                      class="mr-3 flex h-10 w-10 items-center justify-center rounded-full bg-gradient-to-br from-cyan-500 to-teal-600 text-white shadow-lg"
+                    >
+                      <i class="fas fa-list-ol"></i>
+                    </div>
+                    <div>
+                      <h2 class="text-lg font-semibold text-gray-800 dark:text-gray-200">
+                        用户消息串行队列
+                      </h2>
+                      <p class="mt-1 text-sm text-gray-600 dark:text-gray-400">
+                        启用后，同一账户的用户消息请求将串行执行，并在请求之间添加延迟，防止触发上游限流
+                      </p>
+                    </div>
+                  </div>
+                </div>
+                <label class="relative inline-flex cursor-pointer items-center">
+                  <input
+                    v-model="claudeConfig.userMessageQueueEnabled"
+                    class="peer sr-only"
+                    type="checkbox"
+                    @change="saveClaudeConfig"
+                  />
+                  <div
+                    class="peer h-6 w-11 rounded-full bg-gray-200 after:absolute after:left-[2px] after:top-[2px] after:h-5 after:w-5 after:rounded-full after:border after:border-gray-300 after:bg-white after:transition-all after:content-[''] peer-checked:bg-teal-500 peer-checked:after:translate-x-full peer-checked:after:border-white peer-focus:outline-none peer-focus:ring-4 peer-focus:ring-teal-300 dark:border-gray-600 dark:bg-gray-700 dark:peer-focus:ring-teal-800"
+                  ></div>
+                </label>
+              </div>
+
+              <!-- 队列配置详情（仅在启用时显示） -->
+              <div v-if="claudeConfig.userMessageQueueEnabled" class="mt-6 space-y-4">
+                <!-- 请求间隔 -->
+                <div>
+                  <label class="block text-sm font-medium text-gray-700 dark:text-gray-300">
+                    <i class="fas fa-hourglass-half mr-2 text-gray-400"></i>
+                    请求间隔（毫秒）
+                  </label>
+                  <input
+                    v-model.number="claudeConfig.userMessageQueueDelayMs"
+                    class="mt-1 block w-full max-w-xs rounded-lg border border-gray-300 bg-white px-3 py-2 shadow-sm focus:border-teal-500 focus:outline-none focus:ring-2 focus:ring-teal-500/20 dark:border-gray-500 dark:bg-gray-700 dark:text-white sm:text-sm"
+                    max="10000"
+                    min="0"
+                    placeholder="200"
+                    type="number"
+                    @change="saveClaudeConfig"
+                  />
+                  <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                    同一账户的用户消息请求之间的最小间隔时间（0-10000毫秒）
+                  </p>
+                </div>
+
+                <!-- 队列超时 -->
+                <div>
+                  <label class="block text-sm font-medium text-gray-700 dark:text-gray-300">
+                    <i class="fas fa-stopwatch mr-2 text-gray-400"></i>
+                    队列超时（毫秒）
+                  </label>
+                  <input
+                    v-model.number="claudeConfig.userMessageQueueTimeoutMs"
+                    class="mt-1 block w-full max-w-xs rounded-lg border border-gray-300 bg-white px-3 py-2 shadow-sm focus:border-teal-500 focus:outline-none focus:ring-2 focus:ring-teal-500/20 dark:border-gray-500 dark:bg-gray-700 dark:text-white sm:text-sm"
+                    max="300000"
+                    min="1000"
+                    placeholder="30000"
+                    type="number"
+                    @change="saveClaudeConfig"
+                  />
+                  <p class="mt-1 text-xs text-gray-500 dark:text-gray-400">
+                    请求在队列中等待的最大时间，超时将返回 503 错误（1000-300000毫秒）
+                  </p>
+                </div>
+              </div>
+
+              <div class="mt-4 rounded-lg bg-teal-50 p-4 dark:bg-teal-900/20">
+                <div class="flex">
+                  <i class="fas fa-info-circle mt-0.5 text-teal-500"></i>
+                  <div class="ml-3">
+                    <p class="text-sm text-teal-700 dark:text-teal-300">
+                      <strong>工作原理：</strong>系统检测请求中最后一条消息的
+                      <code class="rounded bg-teal-100 px-1 dark:bg-teal-800">role</code>
+                      是否为
+                      <code class="rounded bg-teal-100 px-1 dark:bg-teal-800">user</code
+                      >。用户消息请求需要排队串行执行，而工具调用结果、助手消息续传等不受此限制。
+                    </p>
+                  </div>
+                </div>
+              </div>
+            </div>
+
+            <!-- 配置更新信息 -->
+            <div
+              v-if="claudeConfig.updatedAt"
+              class="rounded-lg bg-gray-50 p-4 text-sm text-gray-500 dark:bg-gray-700/50 dark:text-gray-400"
+            >
+              <i class="fas fa-history mr-2"></i>
+              最后更新：{{ formatDateTime(claudeConfig.updatedAt) }}
+              <span v-if="claudeConfig.updatedBy" class="ml-2">
+                由 <strong>{{ claudeConfig.updatedBy }}</strong> 修改
+              </span>
+            </div>
+          </div>
+        </div>
       </div>
     </div>
   </div>
@@ -1274,6 +1556,20 @@ const webhookConfig = ref({
   }
 })
 
+// Claude 转发配置
+const claudeConfigLoading = ref(false)
+const claudeConfig = ref({
+  claudeCodeOnlyEnabled: false,
+  globalSessionBindingEnabled: false,
+  sessionBindingErrorMessage: '你的本地session已污染，请清理后使用。',
+  sessionBindingTtlDays: 30,
+  userMessageQueueEnabled: true,
+  userMessageQueueDelayMs: 200,
+  userMessageQueueTimeoutMs: 30000,
+  updatedAt: null,
+  updatedBy: null
+})
+
 // 平台表单相关
 const showAddPlatformModal = ref(false)
 const editingPlatform = ref(null)
@@ -1311,6 +1607,8 @@ const sectionWatcher = watch(activeSection, async (newSection) => {
   if (!isMounted.value) return
   if (newSection === 'webhook') {
     await loadWebhookConfig()
+  } else if (newSection === 'claude') {
+    await loadClaudeConfig()
   }
 })
 
@@ -1522,6 +1820,73 @@ const saveWebhookConfig = async () => {
   }
 }
 
+// 加载 Claude 转发配置
+const loadClaudeConfig = async () => {
+  if (!isMounted.value) return
+  claudeConfigLoading.value = true
+  try {
+    const response = await apiClient.get('/admin/claude-relay-config', {
+      signal: abortController.value.signal
+    })
+    if (response.success && isMounted.value) {
+      claudeConfig.value = {
+        claudeCodeOnlyEnabled: response.config?.claudeCodeOnlyEnabled ?? false,
+        globalSessionBindingEnabled: response.config?.globalSessionBindingEnabled ?? false,
+        sessionBindingErrorMessage:
+          response.config?.sessionBindingErrorMessage || '你的本地session已污染，请清理后使用。',
+        sessionBindingTtlDays: response.config?.sessionBindingTtlDays ?? 30,
+        userMessageQueueEnabled: response.config?.userMessageQueueEnabled ?? true,
+        userMessageQueueDelayMs: response.config?.userMessageQueueDelayMs ?? 200,
+        userMessageQueueTimeoutMs: response.config?.userMessageQueueTimeoutMs ?? 30000,
+        updatedAt: response.config?.updatedAt || null,
+        updatedBy: response.config?.updatedBy || null
+      }
+    }
+  } catch (error) {
+    if (error.name === 'AbortError') return
+    if (!isMounted.value) return
+    showToast('获取 Claude 转发配置失败', 'error')
+    console.error(error)
+  } finally {
+    if (isMounted.value) {
+      claudeConfigLoading.value = false
+    }
+  }
+}
+
+// 保存 Claude 转发配置
+const saveClaudeConfig = async () => {
+  if (!isMounted.value) return
+  try {
+    const payload = {
+      claudeCodeOnlyEnabled: claudeConfig.value.claudeCodeOnlyEnabled,
+      globalSessionBindingEnabled: claudeConfig.value.globalSessionBindingEnabled,
+      sessionBindingErrorMessage: claudeConfig.value.sessionBindingErrorMessage,
+      sessionBindingTtlDays: claudeConfig.value.sessionBindingTtlDays,
+      userMessageQueueEnabled: claudeConfig.value.userMessageQueueEnabled,
+      userMessageQueueDelayMs: claudeConfig.value.userMessageQueueDelayMs,
+      userMessageQueueTimeoutMs: claudeConfig.value.userMessageQueueTimeoutMs
+    }
+
+    const response = await apiClient.put('/admin/claude-relay-config', payload, {
+      signal: abortController.value.signal
+    })
+    if (response.success && isMounted.value) {
+      claudeConfig.value = {
+        ...claudeConfig.value,
+        updatedAt: response.config?.updatedAt || new Date().toISOString(),
+        updatedBy: response.config?.updatedBy || null
+      }
+      showToast('Claude 转发配置已保存', 'success')
+    }
+  } catch (error) {
+    if (error.name === 'AbortError') return
+    if (!isMounted.value) return
+    showToast('保存 Claude 转发配置失败', 'error')
+    console.error(error)
+  }
+}
+
 // 验证 URL
 const validateUrl = () => {
   // Bark和SMTP平台不需要验证URL