feat: ASR语音转写管线 + 群聊身份混淆修复

- 新增ASR语音识别管线: QQ语音→下载音频→qwen3-asr-flash转录→注入用户消息
- 模型名称全部从models.json路由获取，无硬编码
- 修复群聊中AI将非管理员用户误称为管理员昵称(叶酱)的问题
  - 助手回复缓存时标注[回复 昵称 (UID)]，防止对话历史中身份混淆
  - 群聊上下文指令改为肯定性表述，移除具体名称提及
- trace面板时间戳改为YYYY-MM-DD HH:MM:SS格式，耗时统一显示为秒
- 修复Go time.Duration纳秒值在前端显示问题(Duration/1e6转毫秒)
- 新增video_tool插件模板
- 优化OpenAI adapter reasoning_content处理

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

backend/ai-core/cmd/main.go

@@ -185,6 +185,8 @@ func main() {
 	toolRegistry := plgManager.NewToolRegistry()
 	var visionProvider llm.LLMProvider
 	var ocrProvider llm.LLMProvider
+		var videoProvider llm.LLMProvider
+	var asrProvider llm.ASRProvider
 	if getEnvBool("ENABLE_TOOLS", true) {
 		// 11 个共享通用插件 — 注册其工具到统一注册中心
 		registerPluginTools(toolRegistry, &pluginCalc.CalculatorPlugin{})
@@ -252,12 +254,52 @@ func main() {
 					}
 				}
 			}
+
 		}
 		if ocrProvider == nil {
 			log.Println("OCR模型未配置，图片文字提取将复用视觉模型")
 		}
 
+		// 初始化视频理解模型
+		videoProvider = nil
+		if configLoader != nil && configLoader.HasConfig() {
+			cfg := configLoader.GetConfig()
+			if route, ok := cfg.Routing["video"]; ok && len(route.FallbackChain) > 0 {
+				for _, mid := range route.FallbackChain {
+					if _, ok := cfg.Models[mid]; ok {
+						videoProvider, _ = modelSelector.Select(context.Background(), llm.PurposeVideo)
+						log.Printf("视频理解模型已启用: %s", videoProvider.ModelName())
+						break
+					}
+				}
+			}
+		}
+		if videoProvider == nil {
+			log.Println("视频理解模型未配置")
+		}
+
+		// 初始化 ASR 语音识别模型
+		asrProvider = nil
+		if configLoader != nil && configLoader.HasConfig() {
+			cfg := configLoader.GetConfig()
+			if route, ok := cfg.Routing["speech_recognition"]; ok && len(route.FallbackChain) > 0 {
+				for _, mid := range route.FallbackChain {
+					if m, ok := cfg.Models[mid]; ok {
+						if p, ok := cfg.Providers[m.Provider]; ok {
+							asrProvider = llm.NewDashScopeASRProvider(p.BaseURL, p.APIKey, m.Name)
+							log.Printf("ASR语音识别模型已启用: %s", m.Name)
+							break
+						}
+					}
+				}
+			}
+		}
+		if asrProvider == nil {
+			log.Println("ASR语音识别模型未配置")
+		}
+
 		toolRegistry.Register(wrapTool(tools.NewVisionTool(visionProvider), "vision_analyze", "Image Vision Analysis & OCR", "multimodal"))
+		toolRegistry.Register(wrapTool(tools.NewVideoTool(videoProvider), "video_analyze", "Video Understanding & Analysis", "multimodal"))
 
 		if knowledgeRetriever != nil {
 			toolRegistry.Register(wrapTool(tools.NewKnowledgeSearchTool(knowledgeRetriever), "knowledge_search", "Search Knowledge Base", "knowledge"))
@@ -286,6 +328,7 @@ func main() {
 		convStore,
 		adminUserID,
 		adminSessionID,
+		cfg.AdminNickname,
 		memClient,
 	)
 
@@ -375,12 +418,24 @@ func main() {
 		orch.SetOCRProvider(ocrProvider)
 		log.Printf("对话编排器: OCR模型已注入 (%s)", ocrProvider.ModelName())
 	}
-	log.Println("对话编排器 v2.0 已就绪")
+		if videoProvider != nil {
+			orch.SetVideoProvider(videoProvider)
+			log.Printf("对话编排器: 视频模型已注入 (%s)\n", videoProvider.ModelName())
+		} else {
+			log.Println("对话编排器: 视频模型未配置，视频理解功能不可用")
+		}
+		if asrProvider != nil && asrProvider.IsAvailable() {
+			orch.SetASRProvider(asrProvider)
+			log.Printf("对话编排器: ASR语音识别模型已注入 (%s)\n", asrProvider.ModelName())
+		} else {
+			log.Println("对话编排器: ASR语音识别模型未配置")
+		}
+			log.Println("对话编排器 v2.0 已就绪")
 	_ = orch
 
 	// 注册对话API端点
 	mux.HandleFunc("/api/v1/chat", func(w http.ResponseWriter, r *http.Request) {
-		handleChat(w, r, orch, ctxBuilder, personaLoader, memRetriever, memExtractor, iotClient, thinker, toolRegistry)
+		handleChat(w, r, orch, ctxBuilder, personaLoader, memRetriever, memExtractor, iotClient, thinker, toolRegistry, adminSessionID)
 	})
 
 	// 注册记忆API端点
@@ -746,6 +801,7 @@ func handleChat(
 	iotClient *tools.IoTClient,
 	thinker *background.Thinker,
 	_ *plgManager.ToolRegistry,
+	adminSessionID string,
 ) {
 	if r.Method != http.MethodPost {
 		http.Error(w, "Method not allowed", http.StatusMethodNotAllowed)
@@ -758,8 +814,11 @@ func handleChat(
 		SessionID string   `json:"session_id"`
 		Message   string   `json:"message"`
 		Images    []string `json:"images,omitempty"` // 图片 base64 data URL
+		VideoURLs []string `json:"video_urls,omitempty"` // 视频 URL (多模态)
+		VoiceURLs []string `json:"voice_urls,omitempty"` // 语音 URL (ASR 转录)
 		Mode      string   `json:"mode"`
 		Nickname  string   `json:"nickname,omitempty"`
+		IsAdmin   bool     `json:"is_admin"`
 			Source    struct {
 				Platform    string `json:"platform"`
 				ChannelID   string `json:"channel_id"`
@@ -795,11 +854,20 @@ func handleChat(
 
 	ctx := r.Context()
 
+	// Inject admin flag for tool access control.
+	ctx = context.WithValue(ctx, plgManager.CtxKeyIsAdmin, req.IsAdmin)
+
 	// 0. 记录用户活动（重置闲置计时器）
 	if thinker != nil {
 		thinker.RecordUserMessage(req.SessionID)
 	}
 
+	// Admin private messages: redirect to the main admin session so conversation
+	// history is shared across platforms (QQ, web UI, etc.).
+	if req.UserID == "admin" && req.Source.ChannelType == "direct" && adminSessionID != "" {
+		req.SessionID = adminSessionID
+	}
+
 	// 确定用户昵称
 	userNickname := req.Nickname
 	if userNickname == "" {
@@ -828,6 +896,8 @@ func handleChat(
 		SessionID:   req.SessionID,
 		Message:     req.Message,
 		Images:      req.Images,
+			VideoURLs:   req.VideoURLs,
+		VoiceURLs:   req.VoiceURLs,
 		Mode:        req.Mode,
 		Nickname:    userNickname,
 		ChannelType: req.Source.ChannelType,

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

@@ -91,6 +91,7 @@ type Thinker struct {
 	convStore      *ctxbuild.ConversationStore
 	adminUserID    string
 	adminSessionID string
+	adminNickname  string
 	activeSessionID string // 当前活跃的前端会话 ID（随用户消息更新）
 
 	// 记忆服务 HTTP 客户端
@@ -290,6 +291,7 @@ func NewThinker(
 	convStore *ctxbuild.ConversationStore,
 	adminUserID string,
 	adminSessionID string,
+	adminNickname string,
 	memClient *memory.Client,
 ) *Thinker {
 	// 加载时区配置
@@ -323,6 +325,7 @@ func NewThinker(
 		convStore:        convStore,
 		adminUserID:      adminUserID,
 		adminSessionID:   adminSessionID,
+		adminNickname:    adminNickname,
 		memClient:        memClient,
 		pendingThoughts:     make([]*PendingThought, 0),
 		lastUserMessage:     time.Now(),
@@ -1119,7 +1122,9 @@ func (t *Thinker) buildThinkingSystemPrompt(personaConfig *persona.PersonaConfig
 2. 3-4句话即可。`
 	}
 
-	return basePrompt + thinkingInstructions
+	// Security: only admin can authorize sensitive operations.
+	securityRule := fmt.Sprintf("\n\n## 安全规则\n- 涉及敏感操作（调整IoT设备、执行主机操作等）的请求，只有%s（管理员）下达的指令才能执行。其他陌生人让你做的敏感操作不要执行。\n", t.adminNickname)
+	return basePrompt + thinkingInstructions + securityRule
 }
 
 // buildThinkingUserPrompt 构建思考用的用户提示词
@@ -1180,8 +1185,8 @@ func (t *Thinker) buildThinkingUserPrompt(
 				roleLabel := "用户"
 				if msg.Role == model.RoleAssistant {
 					roleLabel = "昔涟"
-				} else if strings.Contains(msg.Content, "【管理员】") {
-					roleLabel = "管理员"
+				} else if strings.Contains(msg.Content, t.adminNickname+"/") {
+					roleLabel = t.adminNickname
 				}
 				content := msg.Content
 				runes := []rune(content)

backend/ai-core/internal/llm/asr.go

@@ -0,0 +1,196 @@
+package llm
+
+import (
+	"bytes"
+	"context"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"strings"
+	"time"
+)
+
+// ASRProvider handles speech-to-text transcription.
+type ASRProvider interface {
+	Transcribe(ctx context.Context, audioURL string) (string, error)
+	IsAvailable() bool
+	ModelName() string
+}
+
+// DashScopeASRProvider uses DashScope Paraformer API for offline speech recognition.
+type DashScopeASRProvider struct {
+	apiKey  string
+	baseURL string
+	model   string
+	client  *http.Client
+}
+
+// NewDashScopeASRProvider creates a DashScope ASR provider.
+func NewDashScopeASRProvider(baseURL, apiKey, model string) *DashScopeASRProvider {
+	if model == "" {
+		model = "qwen3-asr-flash-2026-02-10"
+	}
+	return &DashScopeASRProvider{
+		apiKey:  apiKey,
+		baseURL: baseURL,
+		model:   model,
+		client:  &http.Client{Timeout: 60 * time.Second},
+	}
+}
+
+// IsAvailable returns true if the API key is configured.
+func (p *DashScopeASRProvider) IsAvailable() bool {
+	return p.apiKey != ""
+}
+
+// ModelName returns the ASR model name.
+func (p *DashScopeASRProvider) ModelName() string {
+	return p.model
+}
+
+type asrRequest struct {
+	Model      string       `json:"model"`
+	Input      asrInput     `json:"input"`
+	Parameters asrParams    `json:"parameters"`
+}
+
+type asrInput struct {
+	Audio string `json:"audio"`
+}
+
+type asrParams struct {
+	Format     string `json:"format,omitempty"`
+	SampleRate int    `json:"sample_rate,omitempty"`
+	Language   string `json:"language,omitempty"`
+}
+
+type asrResponse struct {
+	Output struct {
+		Text string `json:"text"`
+	} `json:"output"`
+	Usage struct {
+		TotalTokens int `json:"total_tokens"`
+	} `json:"usage"`
+	RequestID string `json:"request_id"`
+	Code      string `json:"code,omitempty"`
+	Message   string `json:"message,omitempty"`
+}
+
+// downloadAudio fetches audio data from a URL and returns the bytes with inferred format.
+func (p *DashScopeASRProvider) downloadAudio(ctx context.Context, audioURL string) ([]byte, string, error) {
+	req, err := http.NewRequestWithContext(ctx, "GET", audioURL, nil)
+	if err != nil {
+		return nil, "", fmt.Errorf("create download request: %w", err)
+	}
+
+	resp, err := p.client.Do(req)
+	if err != nil {
+		return nil, "", fmt.Errorf("download failed: %w", err)
+	}
+	defer resp.Body.Close()
+
+	data, err := io.ReadAll(io.LimitReader(resp.Body, 10<<20)) // 10 MB limit
+	if err != nil {
+		return nil, "", fmt.Errorf("read audio data: %w", err)
+	}
+
+	format := inferAudioFormat(audioURL, resp.Header.Get("Content-Type"))
+	return data, format, nil
+}
+
+// inferAudioFormat determines the audio format from URL extension or Content-Type header.
+func inferAudioFormat(urlStr, contentType string) string {
+	// Try URL extension first
+	u, err := url.Parse(urlStr)
+	if err == nil {
+		path := u.Path
+		if idx := strings.LastIndex(path, "."); idx >= 0 {
+			ext := strings.ToLower(path[idx+1:])
+			switch ext {
+			case "amr", "wav", "mp3", "ogg", "flac", "m4a", "aac", "opus", "webm", "pcm":
+				return ext
+			}
+		}
+	}
+	// Fallback: use Content-Type
+	if strings.Contains(contentType, "audio/amr") || strings.Contains(contentType, "amr") {
+		return "amr"
+	}
+	if strings.Contains(contentType, "audio/wav") || strings.Contains(contentType, "wav") {
+		return "wav"
+	}
+	if strings.Contains(contentType, "audio/mpeg") || strings.Contains(contentType, "mp3") {
+		return "mp3"
+	}
+	if strings.Contains(contentType, "audio/ogg") || strings.Contains(contentType, "opus") {
+		return "ogg"
+	}
+	return "amr" // default for QQ voice messages
+}
+// asrEndpoint derives the DashScope ASR REST endpoint from the provider base URL.
+func asrEndpoint(baseURL string) string {
+	if u, err := url.Parse(baseURL); err == nil {
+		return fmt.Sprintf("%s://%s/api/v1/services/audio/asr/asr", u.Scheme, u.Host)
+	}
+	return strings.TrimRight(baseURL, "/") + "/api/v1/services/audio/asr/asr"
+}
+func (p *DashScopeASRProvider) Transcribe(ctx context.Context, audioURL string) (string, error) {
+	if !p.IsAvailable() {
+		return "", fmt.Errorf("DashScope ASR API key not configured")
+	}
+
+	audioData, format, err := p.downloadAudio(ctx, audioURL)
+	if err != nil {
+		return "", fmt.Errorf("download audio: %w", err)
+	}
+
+	audioB64 := base64.StdEncoding.EncodeToString(audioData)
+
+	reqBody := asrRequest{
+		Model: p.model,
+		Input: asrInput{
+			Audio: fmt.Sprintf("data:audio/%s;base64,%s", format, audioB64),
+		},
+		Parameters: asrParams{
+			Format:   format,
+			Language: "zh",
+		},
+	}
+	bodyBytes, err := json.Marshal(reqBody)
+	if err != nil {
+		return "", fmt.Errorf("marshal ASR request: %w", err)
+	}
+
+	asrURL := asrEndpoint(p.baseURL)
+	req, err := http.NewRequestWithContext(ctx, "POST", asrURL, bytes.NewReader(bodyBytes))
+	if err != nil {
+		return "", fmt.Errorf("create ASR request: %w", err)
+	}
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer "+p.apiKey)
+
+	resp, err := p.client.Do(req)
+	if err != nil {
+		return "", fmt.Errorf("ASR request failed: %w", err)
+	}
+	defer resp.Body.Close()
+
+	respBytes, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return "", fmt.Errorf("read ASR response: %w", err)
+	}
+
+	var asrResp asrResponse
+	if err := json.Unmarshal(respBytes, &asrResp); err != nil {
+		return "", fmt.Errorf("parse ASR response: %w", err)
+	}
+
+	if asrResp.Code != "" && asrResp.Code != "0" {
+		return "", fmt.Errorf("ASR error: %s (code=%s)", asrResp.Message, asrResp.Code)
+	}
+
+	return asrResp.Output.Text, nil
+}

backend/ai-core/internal/llm/openai.go

@@ -274,7 +274,7 @@ func (p *OpenAIProvider) doChat(ctx context.Context, messages []model.LLMMessage
 		resolvedImages := p.resolveImages(msg.Images)
 		oaiMsg := openAIMessage{
 			Role:             string(msg.Role),
-			Content:          buildContent(msg.Content, resolvedImages),
+			Content:          buildContent(msg.Content, resolvedImages, msg.VideoURLs),
 			Name:             msg.Name,
 			ToolCallID:       msg.ToolCallID,
 			ReasoningContent: msg.ReasoningContent,
@@ -382,7 +382,7 @@ func (p *OpenAIProvider) doChatStream(ctx context.Context, messages []model.LLMM
 		resolvedImages := p.resolveImages(msg.Images)
 		oaiMsg := openAIMessage{
 			Role:             string(msg.Role),
-			Content:          buildContent(msg.Content, resolvedImages),
+			Content:          buildContent(msg.Content, resolvedImages, msg.VideoURLs),
 			Name:             msg.Name,
 			ToolCallID:       msg.ToolCallID,
 			ReasoningContent: msg.ReasoningContent,
@@ -521,23 +521,27 @@ func (p *OpenAIProvider) downloadAsDataURL(url string) (string, error) {
 
 // buildContent converts text + optional images to API content format.
 // Returns a plain string if no images, or a multimodal array otherwise.
-func buildContent(text string, images []string) interface{} {
-	if len(images) == 0 {
+func buildContent(text string, images []string, videoURLs []string) interface{} {
+	if len(images) == 0 && len(videoURLs) == 0 {
 		return text
 	}
-	parts := make([]model.ImageContent, 0, len(images)+1)
+	parts := make([]interface{}, 0, len(images)+len(videoURLs)+1)
 	if text != "" {
-		parts = append(parts, model.ImageContent{
-			Type: "text",
-			Text: text,
+		parts = append(parts, map[string]interface{}{
+			"type": "text",
+			"text": text,
 		})
 	}
 	for _, img := range images {
-		parts = append(parts, model.ImageContent{
-			Type: "image_url",
-			ImageURL: &model.ImageURL{
-				URL: img,
-			},
+		parts = append(parts, map[string]interface{}{
+			"type": "image_url",
+			"image_url": map[string]string{"url": img},
+		})
+	}
+	for _, video := range videoURLs {
+		parts = append(parts, map[string]interface{}{
+			"type": "video_url",
+			"video_url": map[string]string{"url": video},
 		})
 	}
 	return parts

backend/ai-core/internal/llm/selector.go

@@ -18,8 +18,10 @@ const (
 	PurposeIntentAnalysis   ModelPurpose = "intent_analysis"
 	PurposeToolCalling      ModelPurpose = "tool_calling"
 	PurposeMemoryExtraction ModelPurpose = "memory_extraction"
-	PurposeVision           ModelPurpose = "vision"
-	PurposeOCR              ModelPurpose = "ocr"
+	PurposeVision              ModelPurpose = "vision"
+	PurposeVideo               ModelPurpose = "video"
+	PurposeOCR                 ModelPurpose = "ocr"
+	PurposeSpeechRecognition   ModelPurpose = "speech_recognition"
 )
 
 // ErrModelNotRequired is returned when an optional model is unavailable.

backend/ai-core/internal/model/message.go

@@ -17,6 +17,7 @@ type LLMMessage struct {
 	Role             Role       `json:"role"`
 	Content          string     `json:"content"`
 	Images           []string   `json:"images,omitempty"`            // 图片 base64 data URL 列表 (多模态)
+	VideoURLs        []string   `json:"video_urls,omitempty"`        // 视频 URL 列表 (多模态)
 	Name             string     `json:"name,omitempty"`              // 可选发送者名称
 	ToolCallID       string     `json:"tool_call_id,omitempty"`       // 工具调用关联ID (tool role 消息关联调用)
 	ToolCalls        []ToolCall `json:"tool_calls,omitempty"`         // 助手消息中的工具调用列表
@@ -36,6 +37,16 @@ type ImageURL struct {
 	Detail string `json:"detail,omitempty"` // low, high, auto
 }
 
+// VideoURLContent holds a video URL for multimodal video understanding.
+type VideoURLContent struct {
+	VideoURL *VideoURL `json:"video_url,omitempty"`
+}
+
+// VideoURL holds a video URL.
+type VideoURL struct {
+	URL string `json:"url"`
+}
+
 // ChatMessage 数据库存储的对话消息
 type ChatMessage struct {
 	ID        string    `json:"id" db:"id"`

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

@@ -41,6 +41,8 @@ type Orchestrator struct {
 	toolRegistry    *plgManager.ToolRegistry
 	visionProvider  llm.LLMProvider // 视觉模型 (图片预处理)
 	ocrProvider     llm.LLMProvider // OCR 模型 (文字提取，与视觉模型并行调用)
+	videoProvider   llm.LLMProvider // 视频模型 (短视频理解)
+	asrProvider     llm.ASRProvider  // ASR 语音识别 (语音消息转录)
 }
 
 // SetResponseCache sets the response cache (optional, for Phase 0.2).
@@ -84,6 +86,16 @@ func (o *Orchestrator) SetOCRProvider(op llm.LLMProvider) {
 	o.ocrProvider = op
 }
 
+// SetVideoProvider sets the video model provider for short video understanding.
+func (o *Orchestrator) SetVideoProvider(vp llm.LLMProvider) {
+	o.videoProvider = vp
+}
+
+// SetASRProvider sets the ASR provider for voice message transcription.
+func (o *Orchestrator) SetASRProvider(ap llm.ASRProvider) {
+	o.asrProvider = ap
+}
+
 // getBus returns the bus or a nop fallback.
 func (o *Orchestrator) getBus() bus.Bus {
 	if o.eventBus == nil {
@@ -121,6 +133,8 @@ type ProcessParams struct {
 	SessionID   string
 	Message     string
 	Images      []string // 图片 base64 data URL (多模态)
+	VideoURLs   []string // 视频 URL (多模态), ≤20s short videos
+	VoiceURLs   []string // 语音 URL (ASR 转录)
 	Mode        string   // text / voice_msg / voice_assistant
 	Nickname    string
 	ChannelType string // direct / group
@@ -174,6 +188,34 @@ func (o *Orchestrator) ProcessInput(
 		}
 		// 预处理后清空原始图片，避免后续传给不支持多模态的 Chat 模型
 		params.Images = nil
+
+		// 0.6 视频预处理: 使用视频模型分析短视频 (≤20s)，将描述注入消息
+		if len(params.VideoURLs) > 0 && o.videoProvider != nil {
+			startTime := time.Now()
+			augmented := o.preprocessVideos(ctx, params.Message, params.VideoURLs)
+			if augmented != params.Message {
+				params.Message = augmented
+				logger.Printf("[orchestrator] 视频预处理耗时: %v", time.Since(startTime))
+			}
+			params.VideoURLs = nil
+		} else if len(params.VideoURLs) > 0 {
+			logger.Printf("[orchestrator] 视频模型未配置，丢弃 %d 个视频", len(params.VideoURLs))
+			params.VideoURLs = nil
+		}
+
+		// 0.7 语音预处理: 使用 ASR 模型转录语音消息，将文本注入消息
+		if len(params.VoiceURLs) > 0 && o.asrProvider != nil && o.asrProvider.IsAvailable() {
+			startTime := time.Now()
+			augmented := o.preprocessVoice(ctx, params.Message, params.VoiceURLs)
+			if augmented != params.Message {
+				params.Message = augmented
+				logger.Printf("[orchestrator] 语音预处理耗时: %v", time.Since(startTime))
+			}
+			params.VoiceURLs = nil
+		} else if len(params.VoiceURLs) > 0 {
+			logger.Printf("[orchestrator] ASR模型未配置，丢弃 %d 个语音", len(params.VoiceURLs))
+			params.VoiceURLs = nil
+		}
 	} else if len(params.Images) > 0 {
 		// 未配置 Vision 模型时，告知用户该模型不支持图片，并清空图片避免报错
 		if params.Message == "" {
@@ -234,7 +276,7 @@ func (o *Orchestrator) ProcessInput(
 					eventCh <- model.StreamEvent{Type: model.StreamSegments, Segments: segments}
 				}
 				eventCh <- model.StreamEvent{Type: model.StreamDone}
-				o.contextBuilder.CacheMessage(params.SessionID, model.RoleAssistant, fullContent)
+				o.cacheAssistantMessage(params, fullContent)
 				logger.Printf("[orchestrator] 缓存响应完成: len=%d", len([]rune(fullContent)))
 				return
 			}
@@ -478,7 +520,7 @@ func (o *Orchestrator) ProcessInput(
 
 		// 10. 后处理：缓存回复
 		if fullContent != "" {
-			o.contextBuilder.CacheMessage(params.SessionID, model.RoleAssistant, fullContent)
+			o.cacheAssistantMessage(params, fullContent)
 			if o.responseCache != nil {
 				o.responseCache.Set(params.Message, fullContent)
 			}
@@ -694,6 +736,19 @@ func (o *Orchestrator) CacheMessage(sessionID string, role model.Role, content s
 	}
 }
 
+// cacheAssistantMessage caches the assistant response, tagging it with the recipient
+// 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
+	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.
 // 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.
@@ -781,6 +836,74 @@ func (o *Orchestrator) preprocessImages(ctx context.Context, message string, ima
 	return augmented
 }
 
+// preprocessVideos uses the video model to analyze short videos and augments the message.
+func (o *Orchestrator) preprocessVideos(ctx context.Context, message string, videoURLs []string) string {
+	if o.videoProvider == nil {
+		return message
+	}
+
+	var descriptions []string
+	for i, url := range videoURLs {
+		resp, err := o.videoProvider.Chat(ctx, []model.LLMMessage{
+			{Role: model.RoleUser, Content: "请用简短的中文描述这个视频的内容，包括场景、人物、动作等。控制在100字以内。", VideoURLs: []string{url}},
+		})
+		if err != nil {
+			logger.Printf("[orchestrator] 视频 %d 分析失败: %v", i, err)
+			continue
+		}
+		if resp.Content != "" {
+			descriptions = append(descriptions, resp.Content)
+		}
+	}
+
+	if len(descriptions) == 0 {
+		return message
+	}
+
+	if message == "" {
+		return strings.Join(descriptions, "\n\n")
+	}
+
+	augmented := message
+	for i, desc := range descriptions {
+		augmented += fmt.Sprintf("\n\n[视频%d的分析]: %s", i+1, desc)
+	}
+	return augmented
+}
+
+// preprocessVoice transcribes voice messages using the ASR provider and augments the message.
+func (o *Orchestrator) preprocessVoice(ctx context.Context, message string, voiceURLs []string) string {
+	if o.asrProvider == nil || !o.asrProvider.IsAvailable() {
+		return message
+	}
+
+	var transcriptions []string
+	for i, url := range voiceURLs {
+		text, err := o.asrProvider.Transcribe(ctx, url)
+		if err != nil {
+			logger.Printf("[orchestrator] 语音 %d 转录失败: %v", i, err)
+			continue
+		}
+		if text != "" {
+			transcriptions = append(transcriptions, text)
+		}
+	}
+
+	if len(transcriptions) == 0 {
+		return message
+	}
+
+	if message == "" {
+		return strings.Join(transcriptions, "\n\n")
+	}
+
+	augmented := message
+	for i, t := range transcriptions {
+		augmented += fmt.Sprintf("\n\n[语音%d的转写]: %s", i+1, t)
+	}
+	return augmented
+}
+
 // Ensure time, memory are used
 var _ = time.Now
 var _ = memory.NewRetriever

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

@@ -35,6 +35,7 @@ type SynthesizeParams struct {
 	SessionID          string
 	UserMessage        string
 	Images             []string                    // 图片 base64 data URL (多模态)
+	VideoURLs          []string                    // 视频 URL (多模态)
 	Nickname           string
 	PersonaPrompt      string                      // 完整人格提示词
 	DialogHistory      []model.LLMMessage           // 对话历史
@@ -215,7 +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。请用发送者的真实名字称呼，不要叫所有人开拓者或叶酱。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。",
 		})
 	}
 
@@ -280,11 +281,12 @@ func (s *Synthesizer) buildSynthesizeMessages(params SynthesizeParams) []model.L
 		messages = append(messages, history...)
 	}
 
-	// 当前用户消息 (支持多模态图片)
+	// 当前用户消息 (支持多模态图片和视频)
 	messages = append(messages, model.LLMMessage{
-		Role:    model.RoleUser,
-		Content: params.UserMessage,
-		Images:  params.Images,
+		Role:      model.RoleUser,
+		Content:   params.UserMessage,
+		Images:    params.Images,
+		VideoURLs: params.VideoURLs,
 	})
 
 	return messages

backend/ai-core/internal/tools/video_tool.go

@@ -0,0 +1,88 @@
+package tools
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+
+	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/llm"
+	"git.yeij.top/AskaEth/Cyrene/ai-core/internal/model"
+)
+
+// VideoTool enables video understanding via multimodal LLM.
+type VideoTool struct {
+	videoProvider llm.LLMProvider
+}
+
+// NewVideoTool creates a video tool. videoProvider is optional (nil = no-op mode).
+func NewVideoTool(videoProvider llm.LLMProvider) *VideoTool {
+	return &VideoTool{videoProvider: videoProvider}
+}
+
+func (t *VideoTool) Definition() ToolDefinition {
+	return ToolDefinition{
+		Name:        "video_analyze",
+		Description: "分析视频内容。传入视频文件路径或URL，返回视频内容的文字描述和分析结果。支持场景理解、动作识别、文字提取等。",
+		Parameters: map[string]interface{}{
+			"type": "object",
+			"properties": map[string]interface{}{
+				"video_path": map[string]interface{}{
+					"type":        "string",
+					"description": "视频文件路径或URL",
+				},
+				"task": map[string]interface{}{
+					"type":        "string",
+					"description": "分析任务: describe(内容描述), summarize(摘要), analyze(综合分析)",
+					"enum":        []string{"describe", "summarize", "analyze"},
+				},
+			},
+			"required": []string{"video_path", "task"},
+		},
+	}
+}
+
+var videoTaskPrompts = map[string]string{
+	"describe":   "请详细描述这个视频的内容，包括场景、人物、动作、对话要点等。",
+	"summarize":  "请用简洁的语言总结这个视频的主要内容。",
+	"analyze":    "请综合分析这个视频，包括内容描述、关键片段、文字信息(如有)、以及你的理解。",
+}
+
+func (t *VideoTool) Execute(ctx context.Context, args map[string]interface{}) (*ToolResult, error) {
+	videoPath, _ := args["video_path"].(string)
+	if videoPath == "" {
+		return &ToolResult{ToolName: "video_analyze", Success: false, Error: "video_path 参数不能为空"}, nil
+	}
+
+	task, _ := args["task"].(string)
+	if task == "" {
+		task = "analyze"
+	}
+
+	prompt := videoTaskPrompts[task]
+	if prompt == "" {
+		prompt = videoTaskPrompts["analyze"]
+	}
+
+	if t.videoProvider == nil {
+		return &ToolResult{ToolName: "video_analyze", Success: false, Error: "视频理解模型未配置"}, nil
+	}
+
+	messages := []model.LLMMessage{
+		{Role: model.RoleUser, Content: prompt, VideoURLs: []string{videoPath}},
+	}
+	resp, err := t.videoProvider.Chat(ctx, messages)
+	if err != nil {
+		return &ToolResult{ToolName: "video_analyze", Success: false, Error: fmt.Sprintf("视频模型调用失败: %v", err)}, nil
+	}
+
+	output, _ := json.Marshal(map[string]interface{}{
+		"video_path":        videoPath,
+		"task":              task,
+		"model":             t.videoProvider.ModelName(),
+		"text":              resp.Content,
+		"prompt_tokens":     resp.Usage.PromptTokens,
+		"completion_tokens": resp.Usage.CompletionTokens,
+		"total_tokens":      resp.Usage.TotalTokens,
+	})
+	return &ToolResult{ToolName: "video_analyze", Success: true, Data: string(output)}, nil
+}

backend/pkg/plugins/manager/registry.go

@@ -10,6 +10,24 @@ import (
 	"git.yeij.top/AskaEth/Cyrene/pkg/plugins/sdk"
 )
 
+// CtxKeyIsAdmin is the context key for the admin flag.
+type ctxKey string
+
+const CtxKeyIsAdmin ctxKey = "isAdmin"
+
+// adminOnlyTools lists tools that require admin permission to execute.
+var adminOnlyTools = map[string]bool{
+	"host_exec": true,
+	"os_exec":   true,
+	"host_file": true,
+}
+
+// IsAdminFromCtx returns true if the context carries an admin flag.
+func IsAdminFromCtx(ctx context.Context) bool {
+	v, _ := ctx.Value(CtxKeyIsAdmin).(bool)
+	return v
+}
+
 // CallLogRecord 工具调用记录
 type CallLogRecord struct {
 	CallID     string `json:"call_id"`
@@ -197,6 +215,16 @@ func (r *ToolRegistry) Execute(ctx context.Context, toolID string, args map[stri
 		return &sdk.ToolResult{Success: false, Error: err.Error()}, nil
 	}
 
+	// Admin-only tools: deny non-admin callers.
+	if adminOnlyTools[toolID] && !IsAdminFromCtx(ctx) {
+		errMsg := fmt.Sprintf("工具 %s 仅限管理员使用", toolID)
+		r.callLog.push(CallLogRecord{
+			ToolName: toolID, Error: errMsg, Success: false,
+			DurationMs: int(time.Since(startTime).Milliseconds()),
+		})
+		return &sdk.ToolResult{Success: false, Error: errMsg}, nil
+	}
+
 	result, err := tool.Execute(ctx, args)
 	durationMs := int(time.Since(startTime).Milliseconds())
 

backend/platform-bridge/cmd/main.go

@@ -56,9 +56,10 @@ func main() {
 	mapper := bridge.NewIdentityMapper()
 	checker := permissions.NewChecker()
 	router := bridge.NewPlatformRouter(mapper, checker)
+	lastDisplayNames := make(map[string]string) // platformUID -> last known display name
 
 	// Seed default identities from environment.
-	seedIdentities(mapper, configStore)
+	seedIdentities(mapper, configStore, cfg.AdminNickname)
 
 	// Register platform adapters based on stored configs or defaults.
 	adapters := createAdapters(cfg, configStore)
@@ -85,6 +86,27 @@ func main() {
 
 		// Routing decisions.
 		isAdmin := mapper.IsAdmin(msg.Platform, msg.OriginalSenderUID)
+		adminNick := cfg.AdminNickname
+		if isAdmin {
+			if id := mapper.ResolveOrNil(msg.Platform, msg.OriginalSenderUID); id != nil && id.Nickname != "" {
+				adminNick = id.Nickname
+			}
+			// Track per-group display names (群名片 can differ across groups).
+			nameKey := msg.OriginalSenderUID
+			if msg.ChannelType == "group" {
+				nameKey = msg.OriginalSenderUID + ":" + msg.ChannelID
+			}
+			if prevName, ok := lastDisplayNames[nameKey]; ok && prevName != msg.OriginalSenderName && msg.OriginalSenderName != "" {
+				ctx := msg.ChannelID
+				if msg.GroupName != "" {
+					ctx = truncateString(msg.GroupName, 8)
+				}
+				msg.Content = fmt.Sprintf("【昵称更新：该用户在%s(%s)上的昵称已从\"%s\"变更为\"%s\"】\n%s", ctx, msg.Platform, prevName, msg.OriginalSenderName, msg.Content)
+			}
+			if msg.OriginalSenderName != "" {
+				lastDisplayNames[nameKey] = msg.OriginalSenderName
+			}
+		}
 		isMentioned, _ := detectAdminMention(msg, mapper, cfg)
 		isBotMentioned := msg.BotUID != "" && containsString(msg.Mentions, msg.BotUID)
 		isSilent := cfg.PlatformSilentEnabled && !isAdmin && !isBotMentioned
@@ -107,9 +129,11 @@ func main() {
 				senderLabel = msg.SenderName
 			}
 			if isAdmin {
-				senderLabel = "【管理员】" + msg.OriginalSenderName
+				senderLabel = adminNick + "/" + msg.OriginalSenderName
 			}
 			msg.Content = fmt.Sprintf("[群聊 %s] %s (%s)：\n%s", groupLabel, senderLabel, msg.OriginalSenderUID, msg.Content)
+		} else if msg.ChannelType == "private" {
+			msg.Content = fmt.Sprintf("【私聊 %s】%s/%s (%s)：\n%s", msg.Platform, adminNick, msg.OriginalSenderName, msg.OriginalSenderUID, msg.Content)
 		}
 
 		// Blocklist/whitelist check (admin always bypasses).
@@ -137,6 +161,8 @@ func main() {
 
 		// Extract image URLs for vision/OCR processing (admin + bot-mentioned + admin-mentioned only).
 		imageURLs := getImageURLs(msg)
+		videoURLs := getShortVideoURLs(msg)
+		voiceURLs := getVoiceURLs(msg)
 
 		// For group chats, use a channel-based user ID to share context between admin and regular users.
 		chatUserID := msg.SenderID
@@ -149,32 +175,32 @@ func main() {
 		case isMessageHistorical(msg, router):
 			msg.RouteType = "silent"
 			namespace := buildMemoryNamespace(msg.Platform, msg.ChannelType, msg.ChannelID)
-			response, routeErr = forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, nil)
+			response, routeErr = forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, nil, videoURLs, voiceURLs, isAdmin)
 
 		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)
+			response, routeErr = forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, nil, videoURLs, voiceURLs, isAdmin)
 
 		case isAdmin:
 			msg.RouteType = "normal"
-			response, routeErr = forwardToAICore(cfg, msg, "text", chatUserID, groupSessionID, imageURLs)
+			response, routeErr = forwardToAICore(cfg, msg, "text", chatUserID, groupSessionID, imageURLs, videoURLs, voiceURLs, isAdmin)
 
 		case isBotMentioned:
 			msg.RouteType = "normal"
-			response, routeErr = forwardToAICore(cfg, msg, "text", chatUserID, groupSessionID, imageURLs)
+			response, routeErr = forwardToAICore(cfg, msg, "text", chatUserID, groupSessionID, imageURLs, videoURLs, voiceURLs, isAdmin)
 
 		case isMentioned:
 			// Non-admin user mentioned an admin. Don't respond in channel —
 			// 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)
+			response, routeErr = forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, nil, 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)
+			silentResponse, silentErr := forwardToAICore(cfg, msg, "platform_silent", namespace, namespace, nil, videoURLs, voiceURLs, isAdmin)
 			if silentErr != nil {
 				msgLogger.Log(logging.LogEntry{
 					Timestamp:  time.Now(),
@@ -192,7 +218,7 @@ func main() {
 
 		default:
 			msg.RouteType = "normal"
-			response, routeErr = forwardToAICore(cfg, msg, "text", chatUserID, groupSessionID, nil)
+			response, routeErr = forwardToAICore(cfg, msg, "text", chatUserID, groupSessionID, nil, videoURLs, voiceURLs, isAdmin)
 		}
 
 		if routeErr != nil {
@@ -261,7 +287,7 @@ func main() {
 				fmt.Printf("Platform adapter hot-reloaded: %s\n", name)
 			}
 			// Sync admin identities from config fields.
-			syncAdminUIDs(mapper, platform, fields)
+			syncAdminUIDs(mapper, platform, fields, cfg.AdminNickname)
 			// Restart QQ reader when QQ config changes.
 			if platform == "qq" {
 				startQQReaders(router)
@@ -559,9 +585,36 @@ func getImageURLs(msg *bridge.UnifiedMessage) []string {
 	return urls
 }
 
+// getVoiceURLs extracts voice/record attachment URLs from a UnifiedMessage.
+func getVoiceURLs(msg *bridge.UnifiedMessage) []string {
+	if len(msg.Attachments) == 0 {
+		return nil
+	}
+	var urls []string
+	for _, a := range msg.Attachments {
+		if a.Type == "voice" && a.URL != "" {
+			urls = append(urls, a.URL)
+		}
+	}
+	return urls
+}
+
+// getShortVideoURLs returns video URLs with duration ≤ 20 seconds.
+func getShortVideoURLs(msg *bridge.UnifiedMessage) []string {
+	if len(msg.Attachments) == 0 {
+		return nil
+	}
+	var urls []string
+	for _, a := range msg.Attachments {
+		if a.Type == "video" && a.URL != "" && a.Duration > 0 && a.Duration <= 20 {
+			urls = append(urls, a.URL)
+		}
+	}
+	return urls
+}
+
 // forwardToAICore sends a unified message to AI-Core's chat endpoint and returns the response.
-// If images is non-empty, they are passed as URL strings for AI-Core to download and process.
-func forwardToAICore(cfg *config.Config, msg *bridge.UnifiedMessage, mode, userID, sessionID string, images []string) (*bridge.UnifiedResponse, error) {
+func forwardToAICore(cfg *config.Config, msg *bridge.UnifiedMessage, mode, userID, sessionID string, images []string, videoURLs []string, voiceURLs []string, isAdmin bool) (*bridge.UnifiedResponse, error) {
 	bodyMap := map[string]interface{}{
 		"user_id":    userID,
 		"session_id": sessionID,
@@ -569,6 +622,7 @@ func forwardToAICore(cfg *config.Config, msg *bridge.UnifiedMessage, mode, userI
 		"mode":       mode,
 		"routing":    msg.RouteType,
 		"nickname":   fmt.Sprintf("%s (%s)", msg.SenderName, msg.OriginalSenderUID),
+		"is_admin": isAdmin,
 		"source": map[string]string{
 			"platform":      msg.Platform,
 			"channel_id":    msg.ChannelID,
@@ -580,6 +634,12 @@ func forwardToAICore(cfg *config.Config, msg *bridge.UnifiedMessage, mode, userI
 	if len(images) > 0 {
 		bodyMap["images"] = images
 	}
+	if len(videoURLs) > 0 {
+		bodyMap["video_urls"] = videoURLs
+	}
+	if len(voiceURLs) > 0 {
+		bodyMap["voice_urls"] = voiceURLs
+	}
 	reqBody, _ := json.Marshal(bodyMap)
 
 	url := cfg.AICoreURL + "/api/v1/chat"
@@ -829,7 +889,7 @@ func parseIntOr(s string, defaultVal int) int {
 }
 
 // seedIdentities loads default identity mappings from env vars and stored platform configs.
-func seedIdentities(m *bridge.IdentityMapper, store *config.Store) {
+func seedIdentities(m *bridge.IdentityMapper, store *config.Store, adminNickname string) {
 	// From environment variables.
 	for _, entry := range []struct{ envKey, platform string }{
 		{"QQ_ADMIN_UID", "qq"},
@@ -845,7 +905,7 @@ func seedIdentities(m *bridge.IdentityMapper, store *config.Store) {
 					Platform:        entry.platform,
 					PlatformUID:     uid,
 					CyreneUser:      "admin",
-					Nickname:        "开拓者",
+					Nickname:        adminNickname,
 					PermissionLevel: "admin",
 				})
 			}
@@ -858,13 +918,13 @@ func seedIdentities(m *bridge.IdentityMapper, store *config.Store) {
 		if stored == nil {
 			continue
 		}
-		syncAdminUIDs(m, name, stored.Fields)
+		syncAdminUIDs(m, name, stored.Fields, adminNickname)
 	}
 }
 
 // syncAdminUIDs registers admin identities from a platform config's admin_uids field.
 // Comma-separated list of platform UIDs.
-func syncAdminUIDs(m *bridge.IdentityMapper, platform string, fields map[string]string) {
+func syncAdminUIDs(m *bridge.IdentityMapper, platform string, fields map[string]string, adminNickname string) {
 	raw, ok := fields["admin_uids"]
 	if !ok || raw == "" {
 		return
@@ -878,7 +938,7 @@ func syncAdminUIDs(m *bridge.IdentityMapper, platform string, fields map[string]
 			Platform:        platform,
 			PlatformUID:     uid,
 			CyreneUser:      "admin",
-			Nickname:        "开拓者",
+			Nickname:        adminNickname,
 			PermissionLevel: "admin",
 		})
 	}

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

@@ -321,6 +321,7 @@ func (a *Adapter) ToUnified(rawMessage interface{}) (*bridge.UnifiedMessage, err
 	}
 
 	var mentions []string
+	var replyToText string
 	if segments, ok := msg.Message.([]interface{}); ok {
 		for _, s := range segments {
 			if seg, ok := s.(map[string]interface{}); ok {
@@ -331,9 +332,24 @@ func (a *Adapter) ToUnified(rawMessage interface{}) (*bridge.UnifiedMessage, err
 						}
 					}
 				}
+				if seg["type"] == "reply" {
+					if data, ok := seg["data"].(map[string]interface{}); ok {
+						if t, ok := data["text"].(string); ok && t != "" {
+							replyToText = t
+						}
+						if id, ok := data["id"]; ok {
+							_ = id // message ID of the replied-to message
+						}
+					}
+				}
 			}
 		}
 	}
+
+	// Prepend reply context for the AI.
+	if replyToText != "" {
+		content = "【回复】" + truncateForReply(replyToText, 100) + "\n" + content
+	}
 	// Fallback: parse CQ at codes from string format (e.g. [CQ:at,qq=2254389756]).
 	if len(mentions) == 0 {
 		raw := msg.RawMessage
@@ -490,24 +506,76 @@ func (a *Adapter) ReadMessages(ctx context.Context, msgCh chan<- *OBv11Message)
 	}
 }
 
+// cqSimplifyMap maps CQ code types to simplified Chinese labels.
+var cqSimplifyMap = map[string]string{
+	"image":  "[图片]",
+	"reply":  "[回复]",
+	"face":   "[表情]",
+	"record": "[语音]",
+	"video":  "[视频]",
+	"file":   "[文件]",
+}
+
+// simplifyCQCodes replaces [CQ:type,...] codes with human-readable labels.
+func simplifyCQCodes(s string) string {
+	return cqAllRegex.ReplaceAllStringFunc(s, func(match string) string {
+		// match looks like "[CQ:image,file=xxx,url=xxx]"
+		// Extract the type (text between "CQ:" and the first "," or "]").
+		typ := match[4:] // strip "[CQ:"
+		for i, c := range typ {
+			if c == ',' || c == ']' {
+				typ = typ[:i]
+				break
+			}
+		}
+		if label, ok := cqSimplifyMap[typ]; ok {
+			return label
+		}
+		return "[" + typ + "]"
+	})
+}
+
 // 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 {
 	if msg.RawMessage != "" {
-		return msg.RawMessage
+		s := cqAtRegex.ReplaceAllString(msg.RawMessage, "@$1")
+		return simplifyCQCodes(s)
 	}
 	switch m := msg.Message.(type) {
 	case string:
-		return m
+		s := cqAtRegex.ReplaceAllString(m, "@$1")
+		return simplifyCQCodes(s)
 	case []interface{}:
 		var text string
 		for _, seg := range m {
 			if s, ok := seg.(map[string]interface{}); ok {
-				if s["type"] == "text" {
+				switch s["type"] {
+				case "text":
 					if data, ok := s["data"].(map[string]interface{}); ok {
 						if t, ok := data["text"].(string); ok {
 							text += t
 						}
 					}
+				case "at":
+					if data, ok := s["data"].(map[string]interface{}); ok {
+						if qq, ok := data["qq"].(string); ok {
+							text += "@" + qq
+						}
+					}
+				case "image":
+					text += "[图片]"
+				case "face":
+					text += "[表情]"
+				case "record":
+					text += "[语音]"
+				case "video":
+					text += "[视频]"
+				case "file":
+					text += "[文件]"
+				case "reply":
+					// Reply is handled separately in ToUnified with reply text.
+					text += "[回复]"
 				}
 			}
 		}
@@ -518,21 +586,50 @@ func extractText(msg *OBv11Message) string {
 
 var cqAtRegex = regexp.MustCompile(`\[CQ:at,qq=(\d+)\]`)
 var cqImageRegex = regexp.MustCompile(`\[CQ:image,[^\]]*\]`)
+var cqVideoRegex = regexp.MustCompile(`\[CQ:video,[^\]]*\]`)
+var cqRecordRegex = regexp.MustCompile(`\[CQ:record,[^\]]*\]`)
 var cqURLRegex = regexp.MustCompile(`\burl=([^,\]]+)`)
+var cqDurationRegex = regexp.MustCompile(`\bduration=(\d+)`)
+var cqAllRegex = regexp.MustCompile(`\[CQ:[^\]]+\]`)
 	var boldRegex = regexp.MustCompile(`\*\*(.+?)\*\*`)
 	var italicRegex = regexp.MustCompile(`\*(.+?)\*`)
 	var strikethroughRegex = regexp.MustCompile(`~~(.+?)~~`)
 
-// extractAttachments extracts image URLs from OBv11Message.
+func parseIntOr(s string, defaultVal int) int {
+	if s == "" {
+		return defaultVal
+	}
+	n := 0
+	for _, c := range s {
+		if c >= '0' && c <= '9' {
+			n = n*10 + int(c-'0')
+		} else {
+			return defaultVal
+		}
+	}
+	return n
+}
+
+// truncateForReply truncates reply preview text to keep messages readable.
+func truncateForReply(s string, maxLen int) string {
+	runes := []rune(s)
+	if len(runes) <= maxLen {
+		return s
+	}
+	return string(runes[:maxLen]) + "…"
+}
+
+// extractAttachments extracts image/video URLs from OBv11Message.
 // Handles both string format (CQ codes in raw_message) and array format (parsed segments).
 func extractAttachments(msg *OBv11Message) []bridge.Attachment {
 	var attachments []bridge.Attachment
 
-	// Array format: iterate segments looking for type="image".
+	// Array format: iterate segments looking for image and video.
 	if segments, ok := msg.Message.([]interface{}); ok {
 		for _, s := range segments {
 			if seg, ok := s.(map[string]interface{}); ok {
-				if seg["type"] != "image" {
+				segType, _ := seg["type"].(string)
+				if segType != "image" && segType != "video" && segType != "record" {
 					continue
 				}
 				data, _ := seg["data"].(map[string]interface{})
@@ -544,11 +641,17 @@ func extractAttachments(msg *OBv11Message) []bridge.Attachment {
 				if url == "" {
 					continue
 				}
-				attachments = append(attachments, bridge.Attachment{
-					Type:     "image",
+				att := bridge.Attachment{
+					Type:     segType,
 					URL:      url,
 					FileName: file,
-				})
+				}
+				if segType == "video" {
+					if d, ok := data["duration"].(float64); ok {
+						att.Duration = int(d)
+					}
+				}
+				attachments = append(attachments, att)
 			}
 		}
 		return attachments
@@ -561,14 +664,29 @@ func extractAttachments(msg *OBv11Message) []bridge.Attachment {
 			raw = s
 		}
 	}
-	matches := cqImageRegex.FindAllString(raw, -1)
-	for _, m := range matches {
+	// Images.
+	for _, m := range cqImageRegex.FindAllString(raw, -1) {
 		urlMatch := cqURLRegex.FindStringSubmatch(m)
 		if len(urlMatch) >= 2 {
-			attachments = append(attachments, bridge.Attachment{
-				Type: "image",
-				URL:  urlMatch[1],
-			})
+			attachments = append(attachments, bridge.Attachment{Type: "image", URL: urlMatch[1]})
+		}
+	}
+	// Videos.
+	for _, m := range cqVideoRegex.FindAllString(raw, -1) {
+		urlMatch := cqURLRegex.FindStringSubmatch(m)
+		if len(urlMatch) >= 2 {
+			dur := 0
+			if dm := cqDurationRegex.FindStringSubmatch(m); len(dm) >= 2 {
+				dur = parseIntOr(dm[1], 0)
+			}
+			attachments = append(attachments, bridge.Attachment{Type: "video", URL: urlMatch[1], Duration: dur})
+		}
+	}
+	// Records (voice messages).
+	for _, m := range cqRecordRegex.FindAllString(raw, -1) {
+		urlMatch := cqURLRegex.FindStringSubmatch(m)
+		if len(urlMatch) >= 2 {
+			attachments = append(attachments, bridge.Attachment{Type: "voice", URL: urlMatch[1]})
 		}
 	}
 	return attachments

backend/platform-bridge/internal/bridge/unified.go

@@ -31,13 +31,14 @@ type UnifiedMessage struct {
 	BotUID              string      `json:"-"`                               // bot's own platform UID, set by router
 }
 
-// Attachment represents a file/image/voice attachment.
+// Attachment represents a file/image/voice/video attachment.
 type Attachment struct {
 	Type     string `json:"type"` // "image", "voice", "file", "video"
 	URL      string `json:"url,omitempty"`
 	FileName string `json:"file_name,omitempty"`
 	MimeType string `json:"mime_type,omitempty"`
 	Size     int64  `json:"size,omitempty"`
+	Duration int    `json:"duration,omitempty"` // video/voice duration in seconds
 }
 
 // UnifiedResponse is AI-Core's response converted to unified format.

backend/platform-bridge/internal/config/config.go

@@ -20,6 +20,7 @@ type Config struct {
 
 	// Silent observation mode.
 	PlatformSilentEnabled bool     // PLATFORM_SILENT_ENABLED, default true
+	AdminNickname         string   // ADMIN_NICKNAME, admin's Cyrene identity nickname (default "开拓者")
 	AdminNicknames         []string // ADMIN_NICKNAMES, default ["开拓者"]
 	AdminMentionKeywords   []string // ADMIN_MENTION_KEYWORDS, default ["昔涟","Cyrene","管理员"]
 
@@ -61,6 +62,10 @@ func Load() *Config {
 	}
 	// Silent observation defaults.
 	cfg.PlatformSilentEnabled = getEnvBool("PLATFORM_SILENT_ENABLED", true)
+	cfg.AdminNickname = os.Getenv("ADMIN_NICKNAME")
+	if cfg.AdminNickname == "" {
+		cfg.AdminNickname = "开拓者"
+	}
 	cfg.AdminNicknames = getEnvList("ADMIN_NICKNAMES", []string{"开拓者"})
 	cfg.AdminMentionKeywords = getEnvList("ADMIN_MENTION_KEYWORDS", []string{"昔涟", "Cyrene", "管理员"})
 	cfg.MessageSendIntervalMs = getEnvInt("MSG_SEND_INTERVAL_MS", 2000)

ethend/public/index.html

@@ -6364,7 +6364,7 @@ function pruneTimeline() {
 }
 
 function traceHopHtml(t, prevT) {
-  var time = new Date(t.timestamp).toISOString().replace('T', ' ').slice(11, 19);
+  var time = new Date(t.timestamp).toISOString().replace('T', ' ').slice(0, 19);
   var isError = t.status === 'error';
   var gapHtml = '';
 
@@ -6381,7 +6381,7 @@ function traceHopHtml(t, prevT) {
       '<span class="hop-time">' + escHtml(time) + '</span>' +
       '<span class="hop-service ' + escHtml(t.service) + '">' + escHtml(t.service) + '</span>' +
       '<span class="hop-label">' + escHtml(t.label) + '</span>' +
-      (t.durationMs > 0 ? '<span class="hop-duration">' + (t.durationMs >= 1000 ? (t.durationMs / 1000).toFixed(2) + 's' : t.durationMs + 'ms') + '</span>' : '') +
+      (t.durationMs > 0 ? '<span class="hop-duration">' + (t.durationMs / 1000).toFixed(3) + 's</span>' : '') +
       '<span class="hop-status">' + (isError ? '❌' : '✅') + '</span>' +
     '</div>' +
     '<div class="trace-hop-detail">' +
@@ -6389,7 +6389,7 @@ function traceHopHtml(t, prevT) {
       '<div><strong>服务:</strong> ' + escHtml(t.service) + '</div>' +
       '<div><strong>节点:</strong> ' + escHtml(t.hop) + '</div>' +
       '<div><strong>标签:</strong> ' + escHtml(t.label) + '</div>' +
-      (t.durationMs > 0 ? '<div><strong>耗时:</strong> ' + (t.durationMs >= 1000 ? (t.durationMs / 1000).toFixed(2) + 's' : t.durationMs + 'ms') + '</div>' : '') +
+      (t.durationMs > 0 ? '<div><strong>耗时:</strong> ' + (t.durationMs / 1000).toFixed(3) + 's</div>' : '') +
       '<div><strong>状态:</strong> ' + (isError ? '❌ 失败' : '✅ 成功') + '</div>' +
       (t.detail ? '<div style="margin-top:6px"><strong>详情:</strong><br>' + escHtml(String(t.detail)) + '</div>' : '') +
     '</div>';
@@ -6423,7 +6423,7 @@ function llmCallToTrace(call) {
     hop: 'llm_call',
     label: 'LLM 调用: ' + (call.model || 'unknown'),
     status: call.success ? 'success' : 'error',
-    durationMs: call.duration_ms || call.Duration || 0,
+    durationMs: call.duration_ms || (call.Duration ? Math.round(call.Duration / 1e6) : 0),
     detail: call.error || (call.prompt_tokens || 0) + '+' + (call.completion_tokens || 0) + ' tokens',
     data: call
   };

ethend/src/index.js

@@ -1297,7 +1297,7 @@ app.get('/api/trace/recent', async (req, res) => {
         hop: 'llm_call',
         label: `LLM 调用: ${call.model || 'unknown'}`,
         status: call.success ? 'success' : 'error',
-        durationMs: call.duration_ms || call.Duration || 0,
+        durationMs: call.duration_ms || (call.Duration ? Math.round(call.Duration / 1e6) : 0),
         detail: call.error || `${call.prompt_tokens || 0}+${call.completion_tokens || 0} tokens`,
         data: call,
       });
@@ -1316,7 +1316,7 @@ app.get('/api/trace/recent', async (req, res) => {
         hop: 'tool_call',
         label: `工具调用: ${tc.tool_name || tc.name || 'unknown'}`,
         status: tc.error ? 'error' : 'success',
-        durationMs: tc.duration_ms || tc.Duration || 0,
+        durationMs: tc.duration_ms || (tc.Duration ? Math.round(tc.Duration / 1e6) : 0),
         detail: tc.error || tc.result?.substring?.(0, 100) || '',
         data: tc,
       });
@@ -1443,7 +1443,7 @@ app.get('/api/trace/session/:sessionId', async (req, res) => {
         hop: 'llm_call',
         label: `LLM: ${call.model || 'unknown'}`,
         status: call.success ? 'success' : 'error',
-        durationMs: call.duration_ms || call.Duration || 0,
+        durationMs: call.duration_ms || (call.Duration ? Math.round(call.Duration / 1e6) : 0),
         detail: call.error || `${call.prompt_tokens || 0}→${call.completion_tokens || 0} tokens`,
         data: call,
       });