feat: 全链路优化 — 死锁修复、MD3主题、上下文持久化、群聊自然化、打字状态、知识库

**死锁根因修复**
- periodicThinkLoop:1015 orphaned lock → 删除（字段已原子化）
- RecordUserMessage 隔离为 recordMu
- atomic.Int64 替换 lastUserMessage/lastThinkTime 等

**MD3 / Android 17 主题**
- 毛玻璃卡片 (backdrop-filter)
- MD3 色彩令牌 (pink primary #f472b6)
- icons.js 独立矢量图标库 + 运行时 emoji 替换
- 无边框卡片、圆角按钮、阴影层次

**上下文持久化**
- AddMessage → saveToDB 异步写 PostgreSQL
- LoadFromDB 恢复 (admin-session-main + 懒加载)
- LLMMessage.Timestamp 字段

**群聊与适配器**
- group_ambient 模式: 非@消息让 LLM 自己判断是否插话
- 戳一戳动作消息总是回复
- NapCat 打字状态 (set_input_status, 最小3秒显示)
- HTTP API 配置 (http_url/http_token)

**知识库 & 防编造**
- knowledge.CanHandle 对 chat 意图也触发
- 关键词预筛选避免无关 embedding 调用
- persona + synthesizer 三重诚实规则
- 工具结果持久化到会话历史

**平台桥接器**
- detached:true Go进程独立存活
- ethend 重启自动接管已运行服务
- stop() 接管模式 taskkill/F/ PID
- Windows netstat 替代 fuser 获取 PID
- 重复适配器种子逻辑修复
- 失败转发日志 Direction: error

**崩溃诊断**
- crashlog 包 (Recover + WrapHTTP + LLMCall)
- /api/v1/debug/goroutines 端点
- thinker 操作日志 + 30s stats
- 日志写入 logs/ 目录持久化

Co-Authored-By: Claude <noreply@anthropic.com>

backend/ai-core/cmd/main.go

@@ -9,6 +9,7 @@ import (
 	"log"
 	"net/http"
 	"os"
+	"runtime"
 	"os/signal"
 	"path/filepath"
 	"strconv"
@@ -20,6 +21,7 @@ import (
 
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/background"
 	aiConfig "git.yeij.top/AskaEth/Cyrene/ai-core/internal/config"
+	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/crashlog"
 	ctxbuild "git.yeij.top/AskaEth/Cyrene/ai-core/internal/context"
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/host"
 	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/llm"
@@ -448,15 +450,78 @@ func main() {
 	)
 	orch.SetToolRegistry(toolRegistry)
 	// 设置工具结果主动推送回调 — 通用，不绑定特定工具
-	orch.SetToolResultPusher(func(sessionID, userID, toolName, result string) {
-		if thinker == nil {
+	orch.SetToolResultPusher(func(sessionID, userID, toolName, result string, params orchestrator.SynthesizeParams) {
+		if thinker == nil || orch == nil {
 			return
 		}
-		// 通过 thinker 的主动消息机制推送
 		if userID == "" {
 			userID = adminUserID
 		}
-		thinker.TriggerReminderMessage(userID, sessionID, fmt.Sprintf("🔧 %s 执行完成：%s", toolName, result))
+
+		// 异步执行跟进：触发 LLM 生成回复并推送到原消息渠道
+		go func() {
+			var toolResult map[string]interface{}
+			if err := json.Unmarshal([]byte(result), &toolResult); err != nil {
+				toolResult = map[string]interface{}{"output": result}
+			}
+			output, _ := toolResult["output"].(string)
+			if output == "" {
+				output = result
+			}
+
+			followUpMsg := fmt.Sprintf("【系统消息】后台工具 %s 执行完成。结果：\n%s\n\n请基于以上结果生成回复发送给用户。", toolName, output)
+
+			// 持久化工具结果到会话历史（重启后不丢失）
+			ctxBuilder.CacheMessage(sessionID, model.RoleSystem,
+				fmt.Sprintf("[工具 %s 执行结果]\n%s", toolName, output))
+
+			ctx, cancel := context.WithTimeout(context.Background(), 90*time.Second)
+			defer cancel()
+			eventCh, err := orch.ProcessInput(ctx, orchestrator.ProcessParams{
+				UserID:      userID,
+				SessionID:   sessionID,
+				Message:     followUpMsg,
+				Mode:        "text",
+				Nickname:    params.Nickname,
+				ChannelType: params.ChannelType,
+				ChannelID:   params.ChannelID,
+				AdapterName: params.AdapterName,
+			})
+			if err != nil {
+				log.Printf("[tool-followup] ProcessInput 失败: %v", err)
+				return
+			}
+
+			var sb strings.Builder
+			for event := range eventCh {
+				if event.Type == model.StreamDelta {
+					sb.WriteString(event.Delta)
+				}
+			}
+			followUpResponse := sb.String()
+			if followUpResponse == "" {
+				return
+			}
+
+			// 推送到原平台渠道
+			if params.ChannelType != "" && params.ChannelID != "" && params.AdapterName != "" {
+				target := background.ProactiveTarget{
+					Platform: params.AdapterName,
+					ChatType: "private", // send-proactive 要求 private/group，不是 direct
+				}
+				if params.ChannelType == "group" {
+					target.ChatType = "group"
+					target.GroupID = params.ChannelID
+				} else {
+					target.UserID = strings.TrimPrefix(params.ChannelID, "private_")
+				}
+				log.Printf("[tool-followup] 推送跟进到 platform=%s chat=%s channel=%s len=%d",
+					target.Platform, target.ChatType, params.ChannelID, len(followUpResponse))
+				thinker.PushPlatformMessage(target, followUpResponse)
+			} else {
+				thinker.TriggerReminderMessage(userID, sessionID, followUpResponse)
+			}
+		}()
 	})
 
 	if visionProvider != nil {
@@ -593,6 +658,21 @@ func main() {
 		w.Write([]byte(`{"status":"ok"}`))
 	})
 
+	mux.HandleFunc("/api/v1/debug/lock-holder", func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		by, at := thinker.LockHolder()
+		json.NewEncoder(w).Encode(map[string]interface{}{
+			"locked_by": by,
+			"locked_at": at.Format(time.RFC3339),
+			"held_for":  time.Since(at).String(),
+		})
+	})
+	mux.HandleFunc("/api/v1/debug/goroutines", func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/plain")
+		buf := make([]byte, 1024*1024)
+		n := runtime.Stack(buf, true)
+		w.Write(buf[:n])
+	})
 	mux.HandleFunc("/api/v1/health", func(w http.ResponseWriter, r *http.Request) {
 		w.Header().Set("Content-Type", "application/json")
 		w.Write([]byte(`{"status":"ok","service":"ai-core","model":"` + chatAdapter.ModelName() + `"}`))
@@ -622,6 +702,7 @@ func main() {
 		w.Header().Set("Connection", "keep-alive")
 		w.Header().Set("Access-Control-Allow-Origin", "*")
 
+		
 		flusher, ok := w.(http.Flusher)
 		if !ok {
 			http.Error(w, "streaming not supported", http.StatusInternalServerError)
@@ -714,10 +795,20 @@ func main() {
 		json.NewEncoder(w).Encode(result)
 	})
 
-	// 启动HTTP服务
+	// Debug: 全链路 HTTP 请求日志
+	debugMux := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		start := time.Now()
+		crashlog.WrapHTTP(mux).ServeHTTP(w, r)
+		elapsed := time.Since(start)
+		if elapsed > 2*time.Second || r.URL.Path != "/api/v1/health" {
+			log.Printf("[http] %s %s %v", r.Method, r.URL.Path, elapsed.Round(time.Millisecond))
+		}
+	})
+
+	// 启动HTTP服务（全局 panic 恢复 + 崩溃日志）
 	srv := &http.Server{
 		Addr:    ":" + cfg.Port,
-		Handler: mux,
+		Handler: debugMux,
 	}
 
 	go func() {
@@ -727,6 +818,21 @@ func main() {
 		}
 	}()
 
+	// Debug: 每30秒输出内存和goroutine统计
+	go func() {
+		ticker := time.NewTicker(30 * time.Second)
+		defer ticker.Stop()
+		var m runtime.MemStats
+		for range ticker.C {
+			runtime.ReadMemStats(&m)
+			log.Printf("[stats] goroutines=%d heap=%dMB sys=%dMB gc=%d",
+				runtime.NumGoroutine(),
+				m.HeapAlloc/1024/1024,
+				m.Sys/1024/1024,
+				m.NumGC)
+		}
+	}()
+
 	// 优雅关闭
 	quit := make(chan os.Signal, 1)
 	signal.Notify(quit, syscall.SIGINT, syscall.SIGTERM)
@@ -1013,7 +1119,8 @@ func handleChat(
 		return
 	}
 
-	ctx := r.Context()
+	ctx, cancel := context.WithTimeout(r.Context(), 120*time.Second)
+	defer cancel()
 
 	// Inject admin flag for tool access control.
 	ctx = context.WithValue(ctx, plgManager.CtxKeyIsAdmin, req.IsAdmin)
@@ -1035,7 +1142,7 @@ func handleChat(
 
 	// Admin private messages: redirect to the main admin session so conversation
 	// history is shared across platforms (OBv11, web UI, etc.).
-	if req.UserID == "admin" && req.Source.ChannelType == "direct" && adminSessionID != "" {
+	if req.IsAdmin && req.Source.ChannelType == "direct" && adminSessionID != "" {
 		req.SessionID = adminSessionID
 	}
 
@@ -1051,6 +1158,9 @@ func handleChat(
 	w.Header().Set("Connection", "keep-alive")
 	w.Header().Set("X-Accel-Buffering", "no")
 
+	// Trace: message received
+	AddTraceEvent("msg_received", req.SessionID, req.UserID, "收到消息: "+req.Message, "success", fmt.Sprintf("platform=%s", req.Source.Platform), 0, nil)
+
 	flusher, ok := w.(http.Flusher)
 	if !ok {
 		http.Error(w, "Streaming not supported", http.StatusInternalServerError)
@@ -1058,7 +1168,12 @@ func handleChat(
 	}
 
 	// 1.5 缓存用户消息到会话历史（在 Orchestrator 之前，确保顺序正确：user → assistant）
-	ctxBuilder.CacheMessage(req.SessionID, model.RoleUser, req.Message)
+	// 管理员主会话聚合多平台消息，加平台标签让 LLM 知道消息来源
+	userMsgForCache := req.Message
+	if req.IsAdmin && req.SessionID == adminSessionID && req.Source.AdapterName != "" {
+		userMsgForCache = fmt.Sprintf("[来自 %s] %s", req.Source.AdapterName, req.Message)
+	}
+	ctxBuilder.CacheMessage(req.SessionID, model.RoleUser, userMsgForCache)
 
 	// 2. 调用 Orchestrator 处理（替代原有的线性处理流程）
 	// Orchestrator 内部处理：意图分析 → 子会话分派 → 结果汇总 → 综合生成回复