fix: 后台思考身份混淆 + 静默模式视觉理解 + QQ卡片解析 + 仪表盘状态修复

- 后台思考对话历史增加标签说明，严格区分群聊中不同发送者
- 静默观察模式传入图片URL并预处理，供后台思考参考
- 视觉+OCR双模型结果合并格式优化，避免LLM误认为多张图片
- QQ卡片消息(CQ:json)正确解析标题/类型，不再丢失为[JSON]
- 进程管理器stop()在进程为null时重置pid/startTime，消除矛盾状态

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend/ai-core/cmd/main.go

@@ -841,9 +841,19 @@ func handleChat(
 		if thinker != nil {
 			thinker.RecordUserMessage(req.SessionID)
 		}
-		ctxBuilder.CacheMessage(req.SessionID, model.RoleUser, req.Message)
+		// 图片预处理：静默观察时也分析图片内容，供后台思考使用
+		message := req.Message
+		if len(req.Images) > 0 {
+			startTime := time.Now()
+			augmented := orch.PreprocessImages(r.Context(), message, req.Images)
+			if augmented != message {
+				message = augmented
+				log.Printf("[silent] 图片预处理耗时: %%v", time.Since(startTime))
+			}
+		}
+		ctxBuilder.CacheMessage(req.SessionID, model.RoleUser, message)
 		// 从观察到的群聊消息中提取记忆。
-		orch.ExtractMemoriesOnly(r.Context(), req.UserID, req.SessionID, req.Message)
+		orch.ExtractMemoriesOnly(r.Context(), req.UserID, req.SessionID, message)
 		if thinker != nil {
 			thinker.TriggerPostChatThink()
 		}

backend/ai-core/internal/background/thinker.go

@@ -1177,8 +1177,11 @@ func (t *Thinker) buildThinkingUserPrompt(
 
 	// 对话历史
 	var lastUserMsg string
+	lastUserIsAdmin := false
 	if len(convHistory) > 0 {
 		sb.WriteString("\n【最近的对话】\n")
+		sb.WriteString(fmt.Sprintf("（标签说明：每条消息前的 [名字] 标识了说话者。只有 [%s] 才是%s。其他名字是群聊中的其他成员，不是%s。请严格根据标签区分不同的人，不要张冠李戴。）\n",
+			t.adminNickname, t.adminNickname, t.adminNickname))
 		msgCount := 0
 		for _, msg := range convHistory {
 			if msg.Role == model.RoleUser || msg.Role == model.RoleAssistant {
@@ -1187,6 +1190,8 @@ func (t *Thinker) buildThinkingUserPrompt(
 					roleLabel = "昔涟"
 				} else if strings.Contains(msg.Content, t.adminNickname+"/") {
 					roleLabel = t.adminNickname
+				} else if name := extractGroupSender(msg.Content); name != "" {
+					roleLabel = name
 				}
 				content := msg.Content
 				runes := []rune(content)
@@ -1197,18 +1202,19 @@ func (t *Thinker) buildThinkingUserPrompt(
 				msgCount++
 				if msg.Role == model.RoleUser {
 					lastUserMsg = msg.Content
+					lastUserIsAdmin = roleLabel == t.adminNickname
 				}
 			}
 		}
 		if msgCount == 0 {
-			sb.WriteString("（暂无对话历史）\n")
+		sb.WriteString("（暂无对话历史）\n")
 		}
 	} else {
 		sb.WriteString("\n【最近的对话】\n（暂无对话历史）\n")
 	}
 
-	// 关键：强调根据对话历史判断用户当前状态
+	// 关键：强调根据对话历史判断当前状态
-	if lastUserMsg != "" {
+	if lastUserMsg != "" && lastUserIsAdmin {
 		sb.WriteString(fmt.Sprintf("\n🔍 **重要**：开拓者最后说的是「%s」。请认真判断：他现在是不是在休息/睡觉/忙？如果是，不要输出【主动消息】指令行。\n", lastUserMsg))
 	}
 
@@ -1821,6 +1827,27 @@ func (t *Thinker) expandMemoryKeywords(ctx context.Context, message string) []st
 	return keywords
 }
 
+// extractGroupSender extracts the sender name from a group message prefix.
+// Group messages have the format: [群聊 GROUPID] SENDERNAME (UID)：\ncontent
+// Returns empty string if the message doesn't match the group format.
+func extractGroupSender(content string) string {
+	if !strings.HasPrefix(content, "[群聊 ") {
+		return ""
+	}
+	// Find "] " which ends the group label
+	bracketEnd := strings.Index(content, "] ")
+	if bracketEnd < 0 {
+		return ""
+	}
+	rest := content[bracketEnd+2:]
+	// Find " (" which precedes the UID
+	parenIdx := strings.Index(rest, " (")
+	if parenIdx < 0 {
+		return ""
+	}
+	return rest[:parenIdx]
+}
+
 // lastUserMessage extracts the last user message from conversation history.
 func lastUserMessage(history []model.LLMMessage) string {
 	for i := len(history) - 1; i >= 0; i-- {

backend/ai-core/internal/orchestrator/orchestrator.go

@@ -180,7 +180,7 @@ func (o *Orchestrator) ProcessInput(
 		// 0.5 图片预处理: 使用视觉模型分析图片，将描述注入消息
 	if len(params.Images) > 0 && o.visionProvider != nil {
 		startTime := time.Now()
-		augmented := o.preprocessImages(ctx, params.Message, params.Images)
+		augmented := o.PreprocessImages(ctx, params.Message, params.Images)
 		if augmented != params.Message {
 			params.Message = augmented
 			logger.Printf("[orchestrator] 图片预处理耗时: %v, 原消息=%d字, 增强后=%d字",
@@ -736,25 +736,20 @@ func (o *Orchestrator) CacheMessage(sessionID string, role model.Role, content s
 	}
 }
 
-// cacheAssistantMessage caches the assistant response, tagging it with the recipient
+// cacheAssistantMessage caches the assistant response.
-// in group chats so dialog history shows who the AI was addressing.
 func (o *Orchestrator) cacheAssistantMessage(params ProcessParams, fullContent string) {
 	if o.contextBuilder == nil {
 		return
 	}
-	cached := fullContent
+	o.contextBuilder.CacheMessage(params.SessionID, model.RoleAssistant, fullContent)
-	if params.ChannelType == "group" && params.Nickname != "" {
-		cached = fmt.Sprintf("[回复 %s]\n%s", params.Nickname, fullContent)
-	}
-	o.contextBuilder.CacheMessage(params.SessionID, model.RoleAssistant, cached)
 }
 
-// preprocessImages uses vision and OCR models to analyze images and augments the user message.
+// PreprocessImages uses vision and OCR models to analyze images and augments the user message.
 // When both vision and OCR providers are available (and are different models), they are called
 // in parallel and both results are passed to the chat model for autonomous judgment.
 // For standalone images (no text): generates a comprehensive description as the message.
 // For text+images: appends image descriptions as contextual annotations.
-func (o *Orchestrator) preprocessImages(ctx context.Context, message string, images []string) string {
+func (o *Orchestrator) PreprocessImages(ctx context.Context, message string, images []string) string {
 	visionPromptBase := "请详细描述这张图片的内容，包括场景、物体、人物、文字（如有）、颜色、氛围等所有视觉信息。"
 	ocrPromptBase := `请逐字逐句完整提取图片中的所有文字内容，保持原有格式和排版。如果图片中没有文字，请回复"无文字"。`
 
@@ -809,7 +804,7 @@ func (o *Orchestrator) preprocessImages(ctx context.Context, message string, ima
 		var combined string
 		switch {
 		case visionDesc != "" && ocrDesc != "":
-			combined = fmt.Sprintf("[视觉分析]: %s\n[文字提取(OCR)]: %s", visionDesc, ocrDesc)
+			combined = fmt.Sprintf("视觉描述：%s\n（图中文字：%s）", visionDesc, ocrDesc)
 		case visionDesc != "":
 			combined = visionDesc
 		case ocrDesc != "":
@@ -831,7 +826,7 @@ func (o *Orchestrator) preprocessImages(ctx context.Context, message string, ima
 
 	augmented := message
 	for i, desc := range descriptions {
-		augmented += fmt.Sprintf("\n\n[图片%d的视觉分析]: %s", i+1, desc)
+		augmented += fmt.Sprintf("\n\n[图片%d分析结果]: %s", i+1, desc)
 	}
 	return augmented
 }

backend/ai-core/internal/orchestrator/synthesizer.go

@@ -216,7 +216,7 @@ func (s *Synthesizer) buildSynthesizeMessages(params SynthesizeParams) []model.L
 	if params.ChannelType == "group" {
 		messages = append(messages, model.LLMMessage{
 			Role:    model.RoleSystem,
-			Content: "【群聊上下文】这条消息来自QQ群聊。消息前缀 [群聊 群号] 昵称 (QQ号) 标注了真实发送者。你不是在和开拓者一对一私聊，而是在群聊中和不同成员交流。请根据消息前缀中的发送者名字称呼对方，不同的人有不同的名字。只在对你说话或延续已有对话时才回复。",
+			Content: "【群聊上下文】这条消息来自QQ群聊。消息前缀 [群聊 群号] 昵称 (QQ号) 标注了真实发送者。你不是在和开拓者一对一私聊，而是在群聊中和不同成员交流。请根据当前这条消息前缀中的发送者名字来称呼对方——即使你之前在历史对话中称呼过别人，也不要把之前用的称呼套在当前发送者身上。不同的人有不同的名字。只在对你说话或延续已有对话时才回复。",
 		})
 	}
 

backend/platform-bridge/cmd/main.go

@@ -180,7 +180,7 @@ func main() {
 		case isAdmin && !isBotMentioned && shouldAdminBeSilent(msg, router):
 			msg.RouteType = "silent"
 			namespace := buildMemoryNamespace(msg.Platform, msg.ChannelType, msg.ChannelID)
-			response, routeErr = forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, nil, videoURLs, voiceURLs, isAdmin)
+			response, routeErr = forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, imageURLs, videoURLs, voiceURLs, isAdmin)
 
 		case isAdmin:
 			msg.RouteType = "normal"
@@ -195,12 +195,12 @@ func main() {
 			// the admin already gets QQ's native @notification. Observe silently.
 			msg.RouteType = "silent"
 			namespace := buildMemoryNamespace(msg.Platform, msg.ChannelType, msg.ChannelID)
-			response, routeErr = forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, nil, videoURLs, voiceURLs, isAdmin)
+			response, routeErr = forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, imageURLs, videoURLs, voiceURLs, isAdmin)
 
 		case isSilent:
 			msg.RouteType = "silent"
 			namespace := buildMemoryNamespace(msg.Platform, msg.ChannelType, msg.ChannelID)
-			silentResponse, silentErr := forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, nil, videoURLs, voiceURLs, isAdmin)
+			silentResponse, silentErr := forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, imageURLs, videoURLs, voiceURLs, isAdmin)
 			if silentErr != nil {
 				msgLogger.Log(logging.LogEntry{
 					Timestamp:  time.Now(),
@@ -769,23 +769,49 @@ func parseSSEAndAccumulate(body string) string {
 	return strings.Join(deltas, "")
 }
 
-// splitContent splits text by ♪ (sentence-break marker), then by \n\n within each segment.
+// splitContent splits text into separate chat messages.
-// Non-empty segments are each wrapped as a chat message; empty input returns a single empty message.
+// It first splits by \n\n (message separator), then within each message
+// optionally splits further by ♪ (sentence-break marker).
+// Very short segments (< 10 chars) are merged with their neighbors to avoid
+// one-word messages followed by a wall of text.
 func splitContent(text string) []bridge.ResponseMessage {
-	// First split by ♪ sentence-break marker.
+	// Step 1: split by \n\n (message-level separator) always.
-	var rawParts []string
+	rawParts := strings.Split(text, "\n\n")
-	if strings.Contains(text, "♪") {
-		rawParts = strings.Split(text, "♪")
-	} else {
-		rawParts = strings.Split(text, "\n\n")
-	}
 	var parts []string
 	for _, p := range rawParts {
 		p = strings.TrimSpace(p)
-		if p != "" {
+		if p == "" {
+			continue
+		}
+		// Step 2: within each \n\n segment, split by ♪ if present.
+		if strings.Contains(p, "♪") {
+			for _, sub := range strings.Split(p, "♪") {
+				sub = strings.TrimSpace(sub)
+				if sub != "" {
+					parts = append(parts, sub)
+				}
+			}
+		} else {
 			parts = append(parts, p)
 		}
 	}
+
+	// Step 3: merge very short segments with neighbors.
+	const minRunes = 8
+	var merged []string
+	for _, part := range parts {
+		if len(merged) > 0 && len([]rune(merged[len(merged)-1])) < minRunes {
+			// Previous segment is too short: merge current into it.
+			merged[len(merged)-1] = merged[len(merged)-1] + "\n" + part
+		} else if len(merged) > 0 && len([]rune(part)) < minRunes {
+			// Current segment is too short: merge it into previous.
+			merged[len(merged)-1] = merged[len(merged)-1] + "\n" + part
+		} else {
+			merged = append(merged, part)
+		}
+	}
+	parts = merged
+
 	var msgs []bridge.ResponseMessage
 	for _, part := range parts {
 		msgs = append(msgs, bridge.ResponseMessage{

backend/platform-bridge/internal/adapter/qq/adapter.go

@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"net/http"
+	"net/url"
 	"regexp"
 	"strings"
 	"sync"
@@ -506,6 +507,55 @@ func (a *Adapter) ReadMessages(ctx context.Context, msgCh chan<- *OBv11Message)
 	}
 }
 
+// parseJSONCardTitle extracts a human-readable title from a QQ JSON card.
+// The data is the raw JSON string from the "data" field of a json-type message segment.
+func parseJSONCardTitle(data string) string {
+	var card struct {
+		App    string `json:"app"`
+		Prompt string `json:"prompt"`
+		Title  string `json:"title"`
+		Desc   string `json:"desc"`
+		Meta   struct {
+			Detail1 struct {
+				Title string `json:"title"`
+				Desc  string `json:"desc"`
+			} `json:"detail_1"`
+			News struct {
+				Title string `json:"title"`
+				Desc  string `json:"desc"`
+			} `json:"news"`
+			Music struct {
+				Title string `json:"title"`
+				Desc  string `json:"desc"`
+			} `json:"music"`
+		} `json:"meta"`
+	}
+	if err := json.Unmarshal([]byte(data), &card); err != nil {
+		return "[卡片消息]"
+	}
+
+	// Prefer prompt (e.g. "[分享]标题"), then meta titles, then top-level title.
+	if card.Prompt != "" {
+		return "[卡片] " + card.Prompt
+	}
+	if card.Meta.Detail1.Title != "" {
+		return "[卡片] " + card.Meta.Detail1.Title
+	}
+	if card.Meta.News.Title != "" {
+		return "[卡片] " + card.Meta.News.Title
+	}
+	if card.Meta.Music.Title != "" {
+		return "[卡片] " + card.Meta.Music.Title
+	}
+	if card.Title != "" {
+		return "[卡片] " + card.Title
+	}
+	if card.Desc != "" {
+		return "[卡片] " + card.Desc
+	}
+	return "[卡片消息]"
+}
+
 // cqSimplifyMap maps CQ code types to simplified Chinese labels.
 var cqSimplifyMap = map[string]string{
 	"image":  "[图片]",
@@ -528,6 +578,15 @@ func simplifyCQCodes(s string) string {
 				break
 			}
 		}
+		if typ == "json" {
+			// Parse data= field from [CQ:json,data=URL_ENCODED_JSON]
+			if dataVal := extractCQParam(match, "data"); dataVal != "" {
+				if decoded, err := url.QueryUnescape(dataVal); err == nil {
+					return parseJSONCardTitle(decoded)
+				}
+			}
+			return "[卡片消息]"
+		}
 		if label, ok := cqSimplifyMap[typ]; ok {
 			return label
 		}
@@ -535,6 +594,24 @@ func simplifyCQCodes(s string) string {
 	})
 }
 
+// extractCQParam extracts a named parameter value from a CQ code string.
+// e.g. extractCQParam("[CQ:json,data=hello%20world]", "data") → "hello%20world"
+func extractCQParam(cqCode, paramName string) string {
+	prefix := paramName + "="
+	idx := strings.Index(cqCode, prefix)
+	if idx < 0 {
+		return ""
+	}
+	val := cqCode[idx+len(prefix):]
+	// Value ends at "," or "]"
+	for i, c := range val {
+		if c == ',' || c == ']' {
+			return val[:i]
+		}
+	}
+	return val
+}
+
 // extractText retrieves plain text from an OBv11 message.
 // CQ codes are converted to human-readable form where applicable (e.g. [CQ:at,qq=xxx] → @xxx).
 func extractText(msg *OBv11Message) string {
@@ -576,6 +653,16 @@ func extractText(msg *OBv11Message) string {
 				case "reply":
 					// Reply is handled separately in ToUnified with reply text.
 					text += "[回复]"
+				case "json":
+					if data, ok := s["data"].(map[string]interface{}); ok {
+						if inner, ok := data["data"].(string); ok && inner != "" {
+							text += parseJSONCardTitle(inner)
+						} else {
+							text += "[卡片消息]"
+						}
+					} else {
+						text += "[卡片消息]"
+					}
 				}
 			}
 		}

ethend/src/process-manager.js

@@ -360,9 +360,10 @@ class ProcessManager extends EventEmitter {
 
     const procInfo = this.processes.get(serviceId);
     if (!procInfo.process) {
-      // 可能已经崩溃了，重置状态
       procInfo.status = 'stopped';
-      return { success: true, message: `${svc.name} 未在运行` };
+      procInfo.pid = null;
+      procInfo.startTime = null;
+      return { success: true, message: `${svc.name} 已停止` };
     }
 
     return new Promise((resolve) => {