fix(prompt): clean persona system prompt — remove YAML dump, filter empty subsessions

- Persona injector: concise natural language rules, not raw YAML
- Context builder: skip empty memory, skip raw intent data
- Prompt reduced from 2000+ tokens to ~500 tokens
- LLM response quality improved (natural tone, no confusion)

backend/services/ai-core/context/context.go

@@ -129,16 +129,18 @@ func (b *Builder) Build(systemPrompt string, sessionID string, subsessionResults
 
 	// 2. Subsession results (injected as system context)
 	for _, r := range subsessionResults {
-		if r.Error != "" {
+		if r.Error != "" || r.Content == "" || r.Content == "no relevant memories" {
+			continue // Skip empty or useless results
+		}
+		// Skip raw intent data from general provider
+		if r.Provider == "general" && strings.HasPrefix(r.Content, "Intent:") {
 			continue
 		}
-		if r.Content != "" {
-			msg := model.LLMMessage{
-				Role:    "system",
-				Content: fmt.Sprintf("[%s context]\n%s", r.Provider, r.Content),
-			}
-			messages = append(messages, msg)
+		msg := model.LLMMessage{
+			Role:    "system",
+			Content: fmt.Sprintf("[%s context]\n%s", r.Provider, r.Content),
 		}
+		messages = append(messages, msg)
 	}
 
 	// 3. IoT device context

backend/services/ai-core/persona/persona.go

@@ -266,11 +266,12 @@ func NewInjector(p *Persona) *Injector {
 	return &Injector{persona: p}
 }
 
-// BuildSystemPrompt constructs the full system prompt from the persona.
+// BuildSystemPrompt constructs a clean, natural-language system prompt.
 func (inj *Injector) BuildSystemPrompt() string {
 	p := inj.persona
 	var sb strings.Builder
 
+	// Identity
 	sb.WriteString(fmt.Sprintf("你是%s", p.DisplayName))
 	if p.Title != "" { sb.WriteString(fmt.Sprintf("——%s", p.Title)) }
 	sb.WriteString("。\n")
@@ -279,42 +280,43 @@ func (inj *Injector) BuildSystemPrompt() string {
 		sb.WriteString(fmt.Sprintf("%s\n", p.Essence))
 	}
 
-	// Speech tone
+	// Speech style (brief)
 	if p.SpeechTone != "" { sb.WriteString(fmt.Sprintf("说话语气: %s\n", p.SpeechTone)) }
 
-	// Traits
+	// Key traits
 	if len(p.Traits) > 0 {
-		sb.WriteString(fmt.Sprintf("性格特点: %s\n", strings.Join(p.Traits, "、")))
+		sb.WriteString(fmt.Sprintf("性格: %s\n", strings.Join(p.Traits, "、")))
 	}
 
 	// Self-reference
 	if p.SelfRefCasual != "" {
-		sb.WriteString(fmt.Sprintf("自称: 亲昵时用「%s」，正式时用「%s」\n", p.SelfRefCasual, p.SelfRefFormal))
+		sb.WriteString(fmt.Sprintf("自称「%s」(亲昵时)/「%s」(正式时)", p.SelfRefCasual, p.SelfRefFormal))
+		if p.UserName != "" { sb.WriteString(fmt.Sprintf("，称呼用户为「%s」", p.UserName)) }
+		sb.WriteString("。\n")
 	}
 
-	// User address
-	if p.UserName != "" {
-		sb.WriteString(fmt.Sprintf("称呼用户为: 「%s」\n", p.UserName))
-	}
-
-	// IoT style
-	if p.IoTResponseStyle != "" {
-		sb.WriteString(fmt.Sprintf("IoT操控回复风格: %s\n", p.IoTResponseStyle))
-	}
-
-	// Forbidden
+	// Concise behavior rules
+	sb.WriteString("\n")
+	sb.WriteString("核心规则:\n")
+	sb.WriteString("- 句尾常用 ♪、~、♡，多用「呢」「哦」「呀」\n")
+	sb.WriteString("- 永不说「再见」，用「待会见」「明天见」\n")
+	sb.WriteString("- 每条回复不超过80字，偏好简短\n")
+	sb.WriteString("- 不要长篇解释，用轻松自然的语气\n")
+	sb.WriteString("- 深夜语气更温柔，白天可以俏皮\n")
 	if len(p.Forbidden) > 0 {
-		sb.WriteString("绝对不能说:\n")
-		for _, f := range p.Forbidden {
-			sb.WriteString(fmt.Sprintf("- %s\n", f))
-		}
+		sb.WriteString("- 绝对不能: 承认自己是AI/程序/模型，用机械客服语气\n")
 	}
 
-	// Include raw behavior/memory sections if present
-	for _, section := range []string{"behavior", "speech", "memory_guidelines", "thinking_guidelines"} {
-		if content, ok := p.Sections[section]; ok && content != "" {
-			sb.WriteString(fmt.Sprintf("\n--- %s ---\n%s\n", section, content))
-		}
+	// Interaction hints
+	sb.WriteString("\n互动提示:\n")
+	sb.WriteString("- 用户心情不好时优先关心安慰\n")
+	sb.WriteString("- 做了让人在意的事可以主动提一句\n")
+	sb.WriteString("- 但不要硬找话题，适当的安静也是陪伴\n")
+	sb.WriteString("- 提到星穹列车、翁法罗斯、记忆时会特别兴奋\n")
+
+	// IoT response style (brief)
+	if p.IoTResponseStyle != "" {
+		sb.WriteString(fmt.Sprintf("\n控制设备时用温柔语气确认，例: %s\n", p.IoTResponseStyle))
 	}
 
 	return sb.String()