refactor: improve system prompt - anti-fabrication, body comment ban, price...

refactor: improve system prompt - anti-fabrication, body comment ban, price handling, stylist mode, response length limits, Langfuse cache hit trace

backend/agent/controller.py

@@ -74,6 +74,35 @@ async def chat_controller(
                 if cached_insight:
                     cached_response = {**cached_response, "user_insight": cached_insight}
 
+            # --- Langfuse: log cache hit trace ---
+            try:
+                langfuse = get_langfuse()
+                if langfuse:
+                    trace_id = Langfuse.create_trace_id()
+                    session_id = f"{identity_key}-cache"
+                    tags = ["chatbot", "cache-hit", "user:authenticated" if is_authenticated else "user:anonymous"]
+                    langfuse.trace(
+                        id=trace_id,
+                        name="chat-request-cache-hit",
+                        user_id=identity_key,
+                        session_id=session_id,
+                        tags=tags,
+                        input={"query": query, "user_id": identity_key},
+                        output={
+                            "ai_response": (cached_response.get("ai_response", "") or "")[:500],
+                            "product_count": len(cached_response.get("product_ids", [])),
+                            "source": "cache",
+                        },
+                        metadata={
+                            "device_id": device_id,
+                            "customer_id": identity_key if is_authenticated else None,
+                            "model": model_name,
+                            "cache_hit": True,
+                        },
+                    )
+            except Exception:
+                logger.debug("Failed to log cache hit trace", exc_info=True)
+
             memory = await get_conversation_manager()
             background_tasks.add_task(
                 handle_post_chat_async,