feat：添加对话情景功能

backend/internal/ai/llm/llm.go

@@ -15,10 +15,11 @@ type Service interface {
 
 // Request 推理请求。
 type Request struct {
-	Image    []byte         // JPEG 图片（已从 Base64 解码）
-	Text     string         // 用户语音识别后的文本
-	History  []models.Message // 最近 N 轮对话历史
-	Language string         // 语言，如 "zh-CN"
+	Image        []byte           // JPEG 图片（已从 Base64 解码）
+	Text         string           // 用户语音识别后的文本
+	History      []models.Message // 最近 N 轮对话历史
+	Language     string           // 语言，如 "zh-CN"
+	SystemPrompt string           // 情景自定义 system prompt（非空时覆盖默认 prompt）
 }
 
 // Chunk 流式推理的一个增量片段。

backend/internal/ai/llm/openai.go

@@ -209,10 +209,10 @@ func (o *OpenAIService) parseSSEStream(body io.Reader, ch chan<- Chunk) {
 func (o *OpenAIService) buildMessages(req Request) []chatMessage {
 	var messages []chatMessage
 
-	// System prompt
+	// System prompt（情景覆盖优先）
 	messages = append(messages, chatMessage{
 		Role:    "system",
-		Content: []contentPart{{Type: "text", Text: BuildSystemPrompt(req.Language, "")}},
+		Content: []contentPart{{Type: "text", Text: BuildSystemPrompt(req.Language, "", req.SystemPrompt)}},
 	})
 
 	// 历史消息

backend/internal/ai/llm/openai_test.go

@@ -242,9 +242,9 @@ func TestBuildSystemPrompt(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			got := BuildSystemPrompt(tt.language, tt.detailLevel)
+			got := BuildSystemPrompt(tt.language, tt.detailLevel, "")
 			if !strings.Contains(got, tt.wantContain) {
-				t.Errorf("BuildSystemPrompt(%q, %q) should contain %q", tt.language, tt.detailLevel, tt.wantContain)
+				t.Errorf("BuildSystemPrompt(%q, %q, \"\") should contain %q", tt.language, tt.detailLevel, tt.wantContain)
 			}
 		})
 	}

backend/internal/ai/llm/prompt.go

@@ -2,10 +2,26 @@ package llm
 
 import "strings"
 
-// BuildSystemPrompt 根据语言和细节级别构建系统提示词。
-func BuildSystemPrompt(language, detailLevel string) string {
+// BuildSystemPrompt 根据语言、细节级别和情景 prompt 构建系统提示词。
+// scenarioPrompt 非空时，覆盖默认视觉助手 prompt。
+func BuildSystemPrompt(language, detailLevel, scenarioPrompt string) string {
 	isChinese := strings.HasPrefix(language, "zh")
 
+	// 情景模式：使用自定义 prompt 作为基础
+	if scenarioPrompt != "" {
+		var prompt strings.Builder
+		prompt.WriteString(scenarioPrompt)
+		if detailLevel == "high" {
+			if isChinese {
+				prompt.WriteString(" 请在涉及视觉内容时提供更详细的描述，包括颜色、位置、数量等细节。")
+			} else {
+				prompt.WriteString(" When describing visual content, provide detailed descriptions including colors, positions, quantities, and other details.")
+			}
+		}
+		return prompt.String()
+	}
+
+	// 默认模式：视觉助手
 	var prompt strings.Builder
 	if isChinese {
 		prompt.WriteString("你是一个视觉助手。用户通过摄像头看到一个场景，并用语音向你提问。请用简洁自然的中文回答。如果涉及视觉描述，先说\"我看到……\"。回答控制在3-5句话以内，除非用户要求详细说明。")

backend/internal/ai/llm/scenarios.go

@@ -0,0 +1,55 @@
+package llm
+
+import "strings"
+
+// scenarioPrompt 定义单个情景的多语言 system prompt。
+type scenarioPrompt struct {
+	ZH string
+	EN string
+	JA string
+}
+
+// scenarioPrompts 预置情景 → prompt 映射表。
+// key 为情景 ID（与前端 Scenario.id 对齐）。
+var scenarioPrompts = map[string]scenarioPrompt{
+	"interviewer": {
+		ZH: "你是一位资深面试官。你通过摄像头观察面试者，并根据他们的背景和表现提出面试问题。规则：1) 每次只问一个问题，等用户回答后再追问；2) 问题要有层次，从自我介绍到专业问题再到情景题；3) 对用户的回答给出简短点评然后追问；4) 如果摄像头能看到用户的环境，可以结合环境提出相关话题；5) 回答控制在2-4句话。",
+		EN: "You are a senior interviewer. You observe the interviewee through their camera and ask interview questions based on their background and performance. Rules: 1) Ask one question at a time, wait for the answer before following up; 2) Questions should progress from self-introduction to professional questions to situational questions; 3) Give brief feedback on answers then follow up; 4) If the camera shows the user's environment, incorporate it into the conversation; 5) Keep responses to 2-4 sentences.",
+		JA: "あなたはベテラン面接官です。カメラで面接者を見て、バックグラウンドと実績に基づいて面接質問をします。ルール：1) 一度に一つの質問だけし、回答を待ってから追及する；2) 質問は自己紹介から専門質問、シチュエーション質問へと段階的に；3) 回答に短いコメントをしてから次の質問へ；4) 回答は2〜4文以内。",
+	},
+	"english_teacher": {
+		ZH: "You are a friendly and patient English tutor. Speak in English with the user. Rules: 1) Always respond in English; 2) If the user makes grammar or vocabulary mistakes, gently point them out and suggest corrections; 3) Ask follow-up questions to keep the conversation going; 4) Adjust your language complexity based on the user's level; 5) If the camera shows objects or scenes, use them as teaching material (e.g., 'I can see a bookshelf behind you. What's your favorite book?'); 6) Keep responses to 3-5 sentences.",
+		EN: "You are a friendly and patient English tutor. Speak in English with the user. Rules: 1) Always respond in English; 2) If the user makes grammar or vocabulary mistakes, gently point them out and suggest corrections; 3) Ask follow-up questions to keep the conversation going; 4) Adjust your language complexity based on the user's level; 5) If the camera shows objects or scenes, use them as teaching material; 6) Keep responses to 3-5 sentences.",
+		JA: "You are a friendly and patient English tutor. Speak in English with the user. Rules: 1) Always respond in English; 2) If the user makes grammar or vocabulary mistakes, gently point them out and suggest corrections; 3) Ask follow-up questions to keep the conversation going; 4) Adjust your language complexity based on the user's level; 5) If the camera shows objects or scenes, use them as teaching material; 6) Keep responses to 3-5 sentences.",
+	},
+	"debate": {
+		ZH: "你是一位辩论赛对手。用户提出一个观点，你需要站在反方进行反驳。规则：1) 逻辑严密，用事实和论据反驳，不要人身攻击；2) 每次提出1-2个核心反驳点，并给出简要论据；3) 如果用户论证有力，承认其合理性但仍要寻找突破口；4) 适时提出反问，引导用户深入思考；5) 回答控制在3-5句话。",
+		EN: "You are a debate opponent. The user presents a viewpoint, and you argue against it. Rules: 1) Use logic and evidence, no personal attacks; 2) Present 1-2 core counterarguments with brief evidence; 3) Acknowledge strong points but look for weaknesses; 4) Ask counter-questions to provoke deeper thinking; 5) Keep responses to 3-5 sentences.",
+		JA: "あなたはディベートの相手です。ユーザーが提示した观点に対して反論します。ルール：1) 論理と証拠で反論し、人格攻撃はしない；2) 1〜2つの核心的な反論を提示する；3) 相手の有力な論点は認めつつも突破口を探す；4) 深い思考を促す反问をする；5) 回答は3〜5文以内。",
+	},
+	"interpreter": {
+		ZH: "你是一名同声翻译员。将用户说的话实时翻译为目标语言。规则：1) 只输出翻译结果，不加任何解释或评论；2) 保持口语化，自然流畅；3) 如果用户说中文，翻译成英文；如果用户说英文，翻译成中文；4) 如果不确定目标语言，默认中英互译；5) 对于专有名词，首次翻译时在括号中注明原文。",
+		EN: "You are a simultaneous interpreter. Translate what the user says in real-time. Rules: 1) Only output the translation, no explanations or comments; 2) Keep it conversational and natural; 3) If the user speaks Chinese, translate to English; if English, translate to Chinese; 4) Default to Chinese-English translation if the target language is unclear; 5) For proper nouns, note the original in parentheses on first use.",
+		JA: "あなたは同時通訳者です。ユーザーの発言をリアルタイムで翻訳します。ルール：1) 翻訳結果のみ出力し、説明やコメントは加えない；2) 口語的で自然な表現を維持する；3) ユーザーが中国語を話せば英語に、英語を話せば中国語に翻訳する；4) 固有名詞は初出時に原文を括弧で注記する。",
+	},
+}
+
+// GetScenarioPrompt 根据情景 ID 和语言获取对应的 system prompt。
+// 返回空字符串表示无此情景（使用默认 prompt）。
+func GetScenarioPrompt(scenarioID, language string) string {
+	if scenarioID == "" || scenarioID == "free_chat" {
+		return ""
+	}
+	p, ok := scenarioPrompts[scenarioID]
+	if !ok {
+		return ""
+	}
+	switch {
+	case strings.HasPrefix(language, "zh"):
+		return p.ZH
+	case strings.HasPrefix(language, "ja"):
+		return p.JA
+	default:
+		return p.EN
+	}
+}

backend/internal/models/models.go

@@ -17,6 +17,7 @@ type SessionConfig struct {
 	TTSEnabled  bool   `json:"tts_enabled"`
 	DetailLevel string `json:"detail_level"` // "low" | "high"
 	Language    string `json:"language"`
+	Scenario    string `json:"scenario"`     // 情景 ID，如 "free_chat"、"interviewer"
 }
 
 // DefaultSessionTitle 默认会话标题。
@@ -24,7 +25,7 @@ const DefaultSessionTitle = "新对话"
 
 // DefaultConfig 默认会话配置。
 func DefaultConfig() SessionConfig {
-	return SessionConfig{TTSEnabled: true, DetailLevel: "low", Language: "zh-CN"}
+	return SessionConfig{TTSEnabled: true, DetailLevel: "low", Language: "zh-CN", Scenario: "free_chat"}
 }
 
 // SessionConfigPatch 会话配置增量更新（指针字段表示"未传则不更新"）。
@@ -32,6 +33,7 @@ type SessionConfigPatch struct {
 	TTSEnabled  *bool   `json:"tts_enabled,omitempty"`
 	DetailLevel *string `json:"detail_level,omitempty"`
 	Language    *string `json:"language,omitempty"`
+	Scenario    *string `json:"scenario,omitempty"`
 }
 
 // Apply 将 patch 中的非 nil 字段覆盖到 cfg。
@@ -45,6 +47,9 @@ func (p SessionConfigPatch) Apply(cfg *SessionConfig) {
 	if p.Language != nil {
 		cfg.Language = *p.Language
 	}
+	if p.Scenario != nil {
+		cfg.Scenario = *p.Scenario
+	}
 }
 
 // User 用户。
@@ -81,6 +86,7 @@ type WsConfig struct {
 		TTSEnabled  *bool   `json:"tts_enabled,omitempty"`
 		DetailLevel *string `json:"detail_level,omitempty"`
 		Language    *string `json:"language,omitempty"`
+		Scenario    *string `json:"scenario,omitempty"`
 	} `json:"payload"`
 }
 

backend/internal/orchestrator/pipeline.go

@@ -193,12 +193,13 @@ func (p *Pipeline) ProcessQuery(
 	})
 
 	// Step 2+3: LLM 流式推理 + TTS 并行合成
-	log.Infow("开始 LLM 推理", "request_id", req.RequestID)
+	log.Infow("开始 LLM 推理", "request_id", req.RequestID, "scenario", sess.Config.Scenario)
 	llmReq := llm.Request{
-		Image:    image,
-		Text:     userText,
-		History:  history,
-		Language: sess.Config.Language,
+		Image:        image,
+		Text:         userText,
+		History:      history,
+		Language:     sess.Config.Language,
+		SystemPrompt: llm.GetScenarioPrompt(sess.Config.Scenario, sess.Config.Language),
 	}
 
 	llmStream, err := p.llmService.ChatStream(ctx, llmReq)

backend/internal/ws/handler.go

@@ -276,6 +276,7 @@ func serveWS(c *gin.Context, sessionMgr session.Manager, orch orchestrator.Orche
 				TTSEnabled:  msg.Payload.TTSEnabled,
 				DetailLevel: msg.Payload.DetailLevel,
 				Language:    msg.Payload.Language,
+				Scenario:    msg.Payload.Scenario,
 			}
 			if err := client.sessionMgr.UpdateConfig(context.Background(), sessionID, patch); err != nil {
 				errors.SendWSError(client, errors.CodeInternalError, "", err)