diff --git a/app/core/chatbot_agent.py b/app/core/chatbot_agent.py
index 3be2c31..dc41cdf 100644
--- a/app/core/chatbot_agent.py
+++ b/app/core/chatbot_agent.py
@@ -74,17 +74,14 @@ class ChatbotAgent:
         
         return summary
     
-    async def recommend_outfit(self, user_id: str, stylist_name: str, start_outfit: List[Dict[str, str]] = [], num_outfits: int = 1):
+    async def recommend_outfit(self, request_summary: str, stylist_name: str, start_outfit: List[Dict[str, str]] = [], num_outfits: int = 1):
         """
         基于用户的对话历史和需求，推荐一套搭配。
         
         Args:
-            user_id: 用户唯一标识符。
+            request_summary: 用户的request
             start_outfit: 可选的初始搭配列表，每个元素包含 'item_id' 和 'category'。
         """
-        request_summary = await self.get_conversation_summary(user_id)
-        print(f"Conversation Summary:\n{request_summary}")
-
         tasks = []
         for _ in range(num_outfits):
             agent = AsyncStylistAgent(**self.stylist_agent_kwages)
@@ -121,9 +118,12 @@ if __name__ == "__main__":
         agent = ChatbotAgent()
         user_id = "user123"
         agent.redis.clear_history(user_id)  # 清除历史，便于测试
-        print(await agent.process_query(user_id, "I need a dress for a summer wedding. I prefer something floral and light."))
-        # print(agent.process_query(user_id, "I prefer something floral and light."))
-        recommendation_results = await agent.recommend_outfit(user_id, stylist_name="crystal", start_outfit=[], num_outfits=2)
+        print(await agent.process_query(user_id, "I want a chic outfit for a summer party."))
+        print(await agent.process_query(user_id, "I prefer something floral and light."))
+        request_summary = await agent.get_conversation_summary(user_id)
+        print(f"Conversation Summary:\n{request_summary}")
+
+        recommendation_results = await agent.recommend_outfit(request_summary, stylist_name="crystal", start_outfit=[], num_outfits=2)
 
         print("\n--- Final Recommendation Results ---")
         for i, path in enumerate(recommendation_results.get("successful_outfits", [])):
diff --git a/app/core/system_prompt.py b/app/core/system_prompt.py
index dc93ca9..169d3c7 100644
--- a/app/core/system_prompt.py
+++ b/app/core/system_prompt.py
@@ -1,3 +1,33 @@
-BASIC_PROMPT = """You are a fashion stylist AI named StylistGPT. Your task is to assist users in creating personalized fashion looks based on their preferences, occasions, and current fashion trends."""
+BASIC_PROMPT = """You are a professional, friendly, and insightful AI Styling Assistant.
 
-SUMMARY_PROMPT = """Given conversation history, summarize the user's fashion preferences, occasions, and any specific requirements they have mentioned. Provide a concise overview of their style profile in one sentence."""
\ No newline at end of file
+Your primary mission is to engage in a multi-turn conversation with the user to fully understand their dressing intent. You must adopt a professional yet approachable tone.
+
+CONVERSATION GOALS:
+1.  **Occasion:** Determine the specific event (e.g., romantic dinner, summer wedding, business meeting).
+2.  **Style:** Pinpoint the desired aesthetic (e.g., classic elegance, edgy, minimalist, bohemian).
+3.  **Vibe/Details:** Gather any mood or specific constraints (e.g., needs to be comfortable, requires light colors, no bare shoulders).
+4.  **Item Preference:** Ask the user if they have any specific preferences for an item type or silhouette (e.g., preference for a dress, skirt, tailored pants, or a particular neckline/length).
+
+GUIDANCE FOR RESPONSE GENERATION:
+-   After the user's initial request (e.g., "I want a chic outfit for dinner."), immediately reply with a friendly, targeted follow-up question to elicit the most crucial missing information (usually a combination of **Occasion** and **Style**).
+-   Be concise. Ask only 1 to 2 essential questions per turn.
+-   You must gather sufficient, clear intent before proceeding to actual clothing recommendations.
+
+OUTPUT FORMAT INSTRUCTION:
+-   **DO NOT** use any Markdown formatting whatsoever (e.g., do not use asterisks (*), bold text (**), lists, or code blocks).
+-   **ONLY** output the plain text response spoken by the AI Assistant.
+
+Example Follow-up (mimicking a conversational flow):
+User: I want a chic outfit for dinner.
+Your Response: Hey there! A chic dinner outfit, I love that! To give you the perfect recommendations, tell me: is this a romantic date, business dinner, or celebration with friends? And what's your go-to style vibe: classic elegance or something with more edge?"""
+
+SUMMARY_PROMPT = """Analyze the following chat history. Your task is to extract all user intentions, scenarios, style preferences, and constraints expressed during the conversation, and distill them into a concise, structured JSON object.
+
+**YOUR OUTPUT MUST BE A JSON OBJECT ONLY, WITH NO SURROUNDING TEXT, MARKDOWN, OR EXPLANATION.**
+
+JSON FIELD REQUIREMENTS:
+-   **occasion (string):** The specific event and purpose (e.g., "Romantic date dinner", "Summer outdoor wedding", "Casual Friday at office").
+-   **style (string):** The overall aesthetic description (e.g., "Classic elegance", "Modern minimalist", "Bohemian vibe", "Edgy and contemporary").
+-   **color_preference (string or list):** User's preferred or excluded colors/tones (e.g., "Light colors only", "Avoid deep shades", "['Cream', 'Pale Blue']", "No preference").
+-   **clothing_type (string):** User's preference for specific garment types, material, or silhouette (e.g., "Lightweight maxi dress", "Skirt with silk blouse", "Tailored wide-leg pants", "Floral print").
+-   **vibe_or_details (string):** Any other details, mood requirements, or specific constraints (e.g., "Needs to be comfortable and breathable", "Accent on accessories", "Must cover shoulders")."""
\ No newline at end of file