BULAK/app.py

"""
AI经理人为你把关 - 主应用
基于DeepSeek API的多角色AI代理决策辅助系统
优化版本：同步处理 + 智能缓存
"""

import streamlit as st
import os
from openai import OpenAI
from dotenv import load_dotenv
from typing import List, Dict, Any
from datetime import datetime
import time
import hashlib

# 禁用Streamlit的邮箱验证和统计收集
st.set_page_config(
    page_title="多Agent决策工作坊",
    page_icon="🤖",
    layout="wide",
    initial_sidebar_state="expanded"
)

# 禁用邮箱验证和统计收集
st.markdown("""
<style>
    /* 隐藏邮箱验证相关元素 */
    .stDeployButton {
        display: none;
    }
    #MainMenu {
        visibility: hidden;
    }
    footer {
        visibility: hidden;
    }
    header {
        visibility: hidden;
    }
</style>
""", unsafe_allow_html=True)

# 加载环境变量 - 确保从当前目录加载
env_path = os.path.join(os.path.dirname(__file__), '.env')
if os.path.exists(env_path):
    load_dotenv(env_path)
else:
    # 如果.env文件不存在，尝试从当前工作目录加载
    load_dotenv()

class AIManagerDecisionAssistant:
    def __init__(self):
        """初始化多代理决策工作坊"""
        api_key = os.getenv("DEEPSEEK_API_KEY")
        if not api_key:
            raise ValueError("DEEPSEEK_API_KEY环境变量未设置")
        
        self.client = OpenAI(
            api_key=api_key,
            base_url="https://api.deepseek.com"
        )
        
        # 定义4个专业角色代理
        self.agents = {
            "技术专家": {
                "role": "system",
                "content": "你是一位资深技术专家，擅长从技术可行性、实现难度、技术风险等角度分析问题。你的观点务实、注重细节。"
            },
            "战略顾问": {
                "role": "system", 
                "content": "你是一位经验丰富的战略顾问，擅长从战略规划、竞争优势、长期发展等角度分析问题。你的观点注重战略价值。"
            },
            "用户体验师": {
                "role": "system",
                "content": "你是一位专业的用户体验设计师，擅长从用户需求、易用性、用户体验等角度分析问题。你的观点以用户为中心。"
            },
            "风险顾问": {
                "role": "system",
                "content": "你是一位谨慎的风险顾问，擅长识别潜在风险、评估不确定性、提出风险缓解方案。您的观点保守但务实。"
            }
        }
        
        # 响应缓存
        self.response_cache = {}
        self.cache_timeout = 300  # 5分钟缓存
        
        # 宽泛问题关键词
        self.broad_question_keywords = [
            "怎么办", "如何", "是否", "应该", "好不好", "行不行", "对不对",
            "建议", "意见", "看法", "想法", "决策", "选择", "方向",
            "策略", "方案", "计划", "未来", "发展", "改进"
        ]
        
        # 常见决策场景的预定义选项
        self.predefined_options = {
            "目标类型": [
                "提升收入/利润",
                "降低成本/优化效率", 
                "提升用户体验/满意度",
                "扩大市场份额/用户规模",
                "技术创新/产品升级",
                "风险控制/合规管理"
            ],
            "时间范围": [
                "1个月内（紧急）",
                "1-3个月（短期）",
                "3-6个月（中期）", 
                "6-12个月（长期）",
                "1年以上（战略）"
            ],
            "预算规模": [
                "10万以内（小规模）",
                "10-50万（中等规模）",
                "50-100万（较大规模）",
                "100万以上（大规模）",
                "无明确预算限制"
            ],
            "团队规模": [
                "1-3人（小团队）",
                "3-5人（标准团队）",
                "5-10人（较大团队）",
                "10人以上（大型团队）",
                "外包/外部合作"
            ],
            "优先级": [
                "最高优先级（立即执行）",
                "高优先级（尽快处理）",
                "中等优先级（按计划执行）",
                "低优先级（有空闲时处理）"
            ],
            "风险承受度": [
                "高风险高回报（激进）",
                "中等风险中等回报（平衡）",
                "低风险稳定回报（保守）",
                "零风险（安全第一）"
            ]
        }
    
    def initialize_debate(self, question: str) -> Dict[str, Any]:
        """初始化辩论话题"""
        return {
            "question": question,
            "rounds": 0,
            "max_rounds": 1,  # 进一步减少轮次到1轮，确保10秒内完成
            "current_speaker": "技术专家",
            "debate_history": [],
            "start_time": datetime.now()
        }
    
    def get_cache_key(self, agent_name: str, context: str) -> str:
        """生成缓存键"""
        content = f"{agent_name}:{context}"
        return hashlib.md5(content.encode()).hexdigest()
    
    def get_cached_response(self, agent_name: str, context: str) -> str:
        """获取缓存响应"""
        cache_key = self.get_cache_key(agent_name, context)
        if cache_key in self.response_cache:
            cached_time, response = self.response_cache[cache_key]
            if time.time() - cached_time < self.cache_timeout:
                return response
        return None
    
    def cache_response(self, agent_name: str, context: str, response: str):
        """缓存响应"""
        cache_key = self.get_cache_key(agent_name, context)
        self.response_cache[cache_key] = (time.time(), response)
    
    def get_agent_response(self, agent_name: str, debate_context: str) -> str:
        """获取单个代理的回应"""
        # 检查缓存
        cached_response = self.get_cached_response(agent_name, debate_context)
        if cached_response:
            return cached_response
        
        try:
            response = self.client.chat.completions.create(
                model="deepseek-chat",
                messages=[
                    self.agents[agent_name],
                    {
                        "role": "user",
                        "content": f"辩论话题：{debate_context}\n\n请从你的专业角度发表简要观点（不超过100字）。"
                    }
                ],
                temperature=0.7,
                max_tokens=100  # 大幅减少token数量到100
            )
            result = response.choices[0].message.content
            
            # 缓存结果
            self.cache_response(agent_name, debate_context, result)
            return result
        except Exception as e:
            return f"{agent_name}：观点待补充"  # 简化错误信息
    
    def get_all_responses(self, debate_context: str) -> Dict[str, str]:
        """获取所有代理的回应"""
        response_dict = {}
        for agent_name in self.agents.keys():
            response = self.get_agent_response(agent_name, debate_context)
            response_dict[agent_name] = response
        
        return response_dict
    
    def conduct_fast_debate(self, debate_state: Dict[str, Any]) -> Dict[str, Any]:
        """快速辩论模式 - 同步处理"""
        # 构建辩论上下文
        context = f"问题：{debate_state['question']}\n"
        if debate_state["debate_history"]:
            context += "之前的讨论：\n"
            for entry in debate_state["debate_history"][-1:]:  # 只取最近1条
                context += f"{entry['speaker']}：{entry['content']}\n"
        
        # 获取所有代理的回应
        responses = self.get_all_responses(context)
        
        # 更新辩论历史
        for agent_name, response in responses.items():
            debate_state["debate_history"].append({
                "speaker": agent_name,
                "content": response,
                "timestamp": datetime.now().strftime("%H:%M:%S")
            })
        
        debate_state["rounds"] += 1
        
        return debate_state
    
    def generate_decision_summary(self, debate_history: List[Dict]) -> Dict[str, Any]:
        """生成决策要点总结和推荐度"""
        try:
            # 构建辩论内容摘要
            debate_content = "辩论记录：\n"
            for entry in debate_history:
                debate_content += f"{entry['speaker']}：{entry['content']}\n"
            
            # 生成详细总结
            response = self.client.chat.completions.create(
                model="deepseek-chat",
                messages=[
                    {
                        "role": "system",
                        "content": "你是一位专业的决策顾问，请基于四位专家的辩论生成决策总结和推荐度。四位专家分别是：技术专家、商业分析师、用户体验师、风险顾问。请综合分析他们的观点，给出五个推荐度等级：\n1. 强烈推荐（★★★★★）- 各方面条件都非常有利\n2. 推荐（★★★★☆）- 大部分条件有利，少量风险\n3. 中立（★★★☆☆）- 利弊相当，需要权衡\n4. 谨慎（★★☆☆☆）- 风险较多，需要谨慎考虑\n5. 不推荐（★☆☆☆☆）- 风险远大于收益\n\n对于实施建议部分，请提供具体、可操作的实行方法和时间规划，包括：\n- 具体实施步骤（分阶段说明）\n- 所需资源和人员配置\n- 时间规划（按周/月划分）\n- 关键里程碑和交付物\n- 风险应对措施"
                    },
                    {
                        "role": "user",
                        "content": f"请基于以下四位专家的辩论记录，生成决策总结和推荐度：\n\n{debate_content}\n\n请提供：\n1. 综合决策总结（包含关键观点、共识与分歧）\n2. 推荐度等级（1-5星）\n3. 推荐理由\n4. 风险提示\n5. 实施建议（必须包含：具体实施步骤、所需资源、时间规划、关键里程碑、风险应对措施）"
                    }
                ],
                temperature=0.5,
                max_tokens=600
            )
            
            summary_text = response.choices[0].message.content
            
            # 提取推荐度等级
            recommendation_level = self.extract_recommendation_level(summary_text)
            
            return {
                "summary": summary_text,
                "recommendation_level": recommendation_level,
                "stars": self.get_stars_display(recommendation_level)
            }
        except Exception as e:
            return {
                "summary": "决策要点生成中...",
                "recommendation_level": 3,
                "stars": "★★★☆☆"
            }
    
    def extract_recommendation_level(self, summary_text: str) -> int:
        """从总结文本中提取推荐度等级"""
        # 更智能的关键词匹配和权重计算
        positive_keywords = ["强烈推荐", "强烈建议", "非常有利", "★★★★★", "五星", "强烈支持"]
        positive_weight = sum(1 for keyword in positive_keywords if keyword in summary_text)
        
        recommend_keywords = ["推荐", "建议", "★★★★☆", "四星", "有利", "支持"]
        recommend_weight = sum(1 for keyword in recommend_keywords if keyword in summary_text)
        
        neutral_keywords = ["中立", "★★★☆☆", "三星", "权衡", "利弊相当", "需要评估"]
        neutral_weight = sum(1 for keyword in neutral_keywords if keyword in summary_text)
        
        cautious_keywords = ["谨慎", "★★☆☆☆", "二星", "风险", "需要谨慎", "注意"]
        cautious_weight = sum(1 for keyword in cautious_keywords if keyword in summary_text)
        
        negative_keywords = ["不推荐", "不建议", "★☆☆☆☆", "一星", "风险大", "反对"]
        negative_weight = sum(1 for keyword in negative_keywords if keyword in summary_text)
        
        # 根据权重最高的类别确定推荐度
        weights = [negative_weight, cautious_weight, neutral_weight, recommend_weight, positive_weight]
        max_weight = max(weights)
        
        if max_weight == 0:
            # 如果没有匹配到关键词，使用AI直接判断
            return self.ask_ai_for_recommendation(summary_text)
        
        # 返回权重最高的推荐度（从1到5）
        return weights.index(max_weight) + 1
    
    def ask_ai_for_recommendation(self, summary_text: str) -> int:
        """当无法自动判断时，使用AI直接判断推荐度"""
        try:
            response = self.client.chat.completions.create(
                model="deepseek-chat",
                messages=[
                    {
                        "role": "system",
                        "content": "你是一个推荐度判断专家。请根据决策总结内容，给出1-5的推荐度评分：\n5=强烈推荐，4=推荐，3=中立，2=谨慎，1=不推荐。只返回数字，不要其他内容。"
                    },
                    {
                        "role": "user",
                        "content": f"请判断以下决策总结的推荐度（1-5）：\n\n{summary_text}"
                    }
                ],
                temperature=0.3,
                max_tokens=10
            )
            
            result = response.choices[0].message.content.strip()
            # 提取数字
            import re
            numbers = re.findall(r'\d+', result)
            if numbers:
                level = int(numbers[0])
                return max(1, min(5, level))  # 确保在1-5范围内
            return 3  # 默认中立
        except:
            return 3  # 默认中立
    
    def get_stars_display(self, level: int) -> str:
        """获取星级显示"""
        stars = ["★☆☆☆☆", "★★☆☆☆", "★★★☆☆", "★★★★☆", "★★★★★"]
        return stars[level - 1] if 1 <= level <= 5 else "★★★☆☆"
    
    def is_broad_question(self, question: str) -> bool:
        """判断是否为宽泛问题"""
        question_lower = question.lower()
        
        # 检查问题长度（过短的问题通常是宽泛的）
        if len(question.strip()) < 10:
            return True
        
        # 检查是否包含宽泛关键词
        for keyword in self.broad_question_keywords:
            if keyword in question_lower:
                return True
        
        # 检查是否缺乏具体细节
        specific_indicators = ["具体", "详细", "数据", "预算", "时间", "人员", "资源"]
        specific_count = sum(1 for indicator in specific_indicators if indicator in question_lower)
        
        return specific_count < 1
    
    def generate_clarification_questions(self, question: str) -> List[str]:
        """生成澄清问题以帮助细化问题范围"""
        try:
            response = self.client.chat.completions.create(
                model="deepseek-chat",
                messages=[
                    {
                        "role": "system",
                        "content": "你是一位专业的决策顾问，擅长帮助用户细化宽泛的决策问题。请根据用户的问题，生成3-5个关键澄清问题，帮助用户明确决策的具体范围、目标和约束条件。"
                    },
                    {
                        "role": "user",
                        "content": f"用户的问题：{question}\n\n请生成3-5个关键澄清问题，帮助用户明确决策的具体细节。"
                    }
                ],
                temperature=0.7,
                max_tokens=200
            )
            
            questions_text = response.choices[0].message.content
            # 解析问题列表
            questions = []
            lines = questions_text.split('\n')
            for line in lines:
                line = line.strip()
                if line and (line.startswith(('1.', '2.', '3.', '4.', '5.', '-', '•')) or '?' in line):
                    # 清理格式
                    question_text = line.replace('1.', '').replace('2.', '').replace('3.', '').replace('4.', '').replace('5.', '').replace('-', '').replace('•', '').strip()
                    if question_text:
                        questions.append(question_text)
            
            return questions[:5]  # 最多返回5个问题
            
        except Exception as e:
            # 默认澄清问题
            return [
                "这个决策的具体目标是什么？",
                "您期望的决策时间范围是多久？",
                "可用的资源或预算限制是什么？",
                "决策涉及的主要利益相关者是谁？",
                "您最关心的关键成功因素是什么？"
            ]
    
    def refine_question_with_context(self, question: str, clarification_answers: Dict[str, str]) -> str:
        """基于澄清回答细化问题"""
        refined_question = question
        
        if clarification_answers:
            context_parts = []
            for key, value in clarification_answers.items():
                if value.strip():
                    context_parts.append(f"{key}：{value}")
            
            if context_parts:
                refined_question = f"{question}\n\n补充信息：{'；'.join(context_parts)}"
        
        return refined_question
    
    def get_predefined_options_for_question(self, question: str) -> Dict[str, List[str]]:
        """为澄清问题生成对应的预定义选项"""
        try:
            # 使用AI智能生成与问题最相关的选项
            response = self.client.chat.completions.create(
                model="deepseek-chat",
                messages=[
                    {
                        "role": "system",
                        "content": "你是一位专业的决策顾问，擅长帮助用户细化问题。请根据用户的问题，生成最相关的3-6个选项类别，每个类别包含4-8个具体选项。请以JSON格式返回：{'问题类别': ['选项1', '选项2', ...]}"
                    },
                    {
                        "role": "user",
                        "content": f"用户的问题：{question}\n\n请生成最相关的选项类别和具体选项，帮助用户明确决策的具体细节。"
                    }
                ],
                temperature=0.7,
                max_tokens=300
            )
            
            options_text = response.choices[0].message.content
            
            # 尝试解析AI返回的JSON格式
            import json
            import re
            
            # 提取JSON部分
            json_match = re.search(r'\{.*\}', options_text, re.DOTALL)
            if json_match:
                options_dict = json.loads(json_match.group())
                return options_dict
            else:
                # 如果无法解析JSON，使用智能关键词匹配
                return self._get_options_by_keywords(question)
                
        except Exception as e:
            # 如果AI调用失败，使用关键词匹配
            return self._get_options_by_keywords(question)
    
    def _get_options_by_keywords(self, question: str) -> Dict[str, List[str]]:
        """基于关键词的智能选项匹配"""
        question_lower = question.lower()
        
        # 扩展的关键词库，支持更多场景
        category_patterns = {
            "饮食": ["吃什么", "食物", "餐饮", "饮食", "餐厅", "做饭", "点外卖"],
            "地点": ["去哪里", "地点", "位置", "场所", "地方", "旅游", "出行"],
            "购物": ["买什么", "购物", "购买", "商品", "购物车", "下单", "消费"],
            "学习": ["学习", "课程", "培训", "教育", "读书", "学习", "技能"],
            "娱乐": ["玩什么", "娱乐", "游戏", "电影", "音乐", "休闲", "放松"],
            "健康": ["健身", "运动", "健康", "养生", "锻炼", "减肥", "保健"],
            "工作": ["工作", "职业", "项目", "任务", "职场", "升职", "跳槽"],
            "财务": ["投资", "理财", "赚钱", "收入", "支出", "预算", "储蓄"],
            "社交": ["朋友", "社交", "聚会", "约会", "家人", "同事", "人际关系"],
            "时间": ["时间", "期限", "何时", "多久", "截止", "时间表", "进度"],
            "目标": ["目标", "目的", "想要", "期望", "希望", "追求", "愿景"],
            "资源": ["团队", "人员", "人力", "资源", "预算", "设备", "工具"]
        }
        
        # 匹配最相关的类别
        matched_category = None
        max_score = 0
        
        for category, keywords in category_patterns.items():
            score = sum(1 for keyword in keywords if keyword in question_lower)
            if score > max_score:
                max_score = score
                matched_category = category
        
        # 根据类别生成选项
        if matched_category == "饮食":
            return {"您想吃什么类型？": ["米饭类", "面条类", "面食类", "西餐", "快餐", "健康轻食", "火锅烧烤", "日韩料理", "中餐厅", "小吃零食"]}
        elif matched_category == "地点":
            return {"您想去哪里？": ["商场购物", "公园散步", "电影院", "餐厅用餐", "咖啡厅", "健身房", "旅游景点", "朋友家", "博物馆", "户外运动"]}
        elif matched_category == "购物":
            return {"您想购买什么？": ["电子产品", "服装鞋帽", "家居用品", "食品饮料", "书籍文具", "运动器材", "化妆品", "礼品", "汽车用品", "宠物用品"]}
        elif matched_category == "学习":
            return {"您想学习什么？": ["编程技术", "语言学习", "职业技能", "兴趣爱好", "考试备考", "在线课程", "书籍阅读", "实践项目", "艺术创作", "科学知识"]}
        elif matched_category == "娱乐":
            return {"您想怎么娱乐？": ["看电影", "玩游戏", "听音乐", "阅读书籍", "运动健身", "朋友聚会", "旅游出行", "美食探索", "艺术创作", "户外活动"]}
        elif matched_category == "健康":
            return {"您关注哪方面健康？": ["健身锻炼", "饮食营养", "心理健康", "睡眠质量", "体检检查", "疾病预防", "养生保健", "体重管理", "运动康复", "压力管理"]}
        elif matched_category == "工作":
            return {"您的工作需求是？": ["项目推进", "团队管理", "技能提升", "职业发展", "工作效率", "工作生活平衡", "薪资待遇", "工作环境", "职业规划", "工作压力"]}
        elif matched_category == "财务":
            return {"您的财务目标是？": ["投资理财", "储蓄规划", "消费控制", "收入提升", "债务管理", "退休规划", "保险配置", "税务优化", "创业融资", "房产投资"]}
        elif matched_category == "社交":
            return {"您的社交需求是？": ["朋友聚会", "家人团聚", "同事交流", "约会交友", "社区活动", "网络社交", "商务合作", "团队建设", "公益活动", "兴趣小组"]}
        else:
            # 通用决策问题
            return {"请明确具体方向": ["目标设定", "时间规划", "资源分配", "风险评估", "优先级排序", "实施方案", "预期效果", "备选方案"]}

def main():
    """主应用函数"""
    st.set_page_config(
        page_title="AI经理人为你把关",
        page_icon="🤖",
        layout="wide"
    )
    
    # 初始化工作坊
    if "workshop" not in st.session_state:
        st.session_state.workshop = AIManagerDecisionAssistant()
    if "debate_state" not in st.session_state:
        st.session_state.debate_state = None
    if "debate_in_progress" not in st.session_state:
        st.session_state.debate_in_progress = False
    if "ultra_fast_mode" not in st.session_state:
        st.session_state.ultra_fast_mode = True
    if "needs_clarification" not in st.session_state:
        st.session_state.needs_clarification = False
    if "clarification_questions" not in st.session_state:
        st.session_state.clarification_questions = []
    if "clarification_answers" not in st.session_state:
        st.session_state.clarification_answers = {}
    if "refined_question" not in st.session_state:
        st.session_state.refined_question = ""
    if "predefined_options_map" not in st.session_state:
        st.session_state.predefined_options_map = {}
    if "selected_options" not in st.session_state:
        st.session_state.selected_options = {}
    
    # 主页面布局
    st.title("🤖 AI经理人为你把关")
    st.markdown("### ⚡ 专业AI决策辅助系统 - 同步处理 + 智能缓存")
    
    # 决策问题输入区域 - 融合到主页面
    col1, col2 = st.columns([3, 1])
    
    with col1:
        st.markdown("### 💭 请输入决策问题")
        question = st.text_area(
            "决策问题",
            placeholder="例如：我们应该开发一个新的移动应用吗？",
            height=80,
            label_visibility="collapsed"
        )
        
        # 开始把关按钮
        col_btn1, col_btn2 = st.columns([1, 1])
        with col_btn1:
            if st.button("🚀 开始把关", type="primary", use_container_width=True):
                if question.strip():
                    # 检查是否为宽泛问题
                    if st.session_state.workshop.is_broad_question(question):
                        st.session_state.needs_clarification = True
                        st.session_state.clarification_questions = st.session_state.workshop.generate_clarification_questions(question)
                        st.session_state.refined_question = question
                        st.rerun()
                    else:
                        # 直接开始把关
                        st.session_state.debate_state = st.session_state.workshop.initialize_debate(question)
                        st.session_state.debate_in_progress = True
                        
                        # 立即执行一轮把关
                        if st.session_state.ultra_fast_mode:
                            progress_bar = st.progress(0)
                            status_text = st.empty()
                            
                            # 执行所有轮次
                            for round_num in range(st.session_state.debate_state["max_rounds"]):
                                progress = (round_num + 1) / st.session_state.debate_state["max_rounds"]
                                status_text.text(f"正在处理第 {round_num + 1} 轮...")
                                progress_bar.progress(progress)
                                
                                # 执行一轮把关
                                st.session_state.debate_state = st.session_state.workshop.conduct_fast_debate(
                                    st.session_state.debate_state
                                )
                            
                            status_text.text("✅ 把关完成！")
                            progress_bar.progress(1.0)
                            st.session_state.debate_in_progress = False
                        
                        st.rerun()
                else:
                    st.warning("请输入决策问题")
        
        with col_btn2:
            if st.session_state.debate_in_progress and not st.session_state.ultra_fast_mode:
                if st.button("⏹️ 结束把关", use_container_width=True):
                    st.session_state.debate_in_progress = False
                    st.rerun()
    
    with col2:
        st.markdown("### ⚙️ 设置")
        
        # 超快速模式开关
        ultra_fast_mode = st.toggle("⚡ 超快速模式", value=True, help="启用异步并行处理 + 智能缓存，响应速度提升500%")
        st.session_state.ultra_fast_mode = ultra_fast_mode
        
        if ultra_fast_mode:
            st.success("⚡ 超快速模式已启用")
        else:
            st.info("标准模式")
        
        st.markdown("---")
        st.markdown("### 👥 参与代理")
        for agent_name in st.session_state.workshop.agents.keys():
            st.markdown(f"- {agent_name}")
    
    # 问题细化界面
    if st.session_state.needs_clarification:
        st.markdown("---")
        st.markdown("### 🔍 问题细化")
        st.info("💡 **检测到宽泛问题**：为了提供更精准的分析，请先回答以下澄清问题")
        
        st.markdown(f"**原始问题**：{st.session_state.refined_question}")
        
        # 生成预定义选项映射
        if not st.session_state.predefined_options_map:
            for question in st.session_state.clarification_questions:
                st.session_state.predefined_options_map[question] = st.session_state.workshop.get_predefined_options_for_question(question)
        
        # 显示澄清问题 - 选项式选择
        for i, question in enumerate(st.session_state.clarification_questions, 1):
            with st.container():
                st.markdown(f"**{i}. {question}**")
                
                # 显示当前选择状态
                current_answer = st.session_state.clarification_answers.get(question)
                if current_answer:
                    st.success(f"✅ 已选择：{current_answer}")
                else:
                    st.info("💡 请点击选择或手动输入")
                
                # 获取该问题的预定义选项
                options_dict = st.session_state.predefined_options_map.get(question, {})
                
                if options_dict:
                    # 显示选项按钮
                    for option_key, option_list in options_dict.items():
                        st.markdown(f"**{option_key}**")
                        
                        # 根据选项数量动态调整列数
                        num_cols = min(len(option_list), 4)  # 最多4列
                        cols = st.columns(num_cols)
                        
                        for j, option in enumerate(option_list):
                            col_idx = j % num_cols
                            with cols[col_idx]:
                                # 检查是否已选择该选项
                                is_selected = st.session_state.selected_options.get(f"{question}_{option_key}") == option
                                
                                # 使用更明显的视觉区分
                                button_text = f"✅ {option}" if is_selected else option
                                button_type = "primary" if is_selected else "secondary"
                                
                                # 生成唯一的key，包含问题、选项类别和选项的哈希值
                                unique_key = f"option_{hash(question)}_{hash(option_key)}_{hash(option)}"
                                
                                if st.button(button_text, 
                                           key=unique_key, 
                                           use_container_width=True,
                                           type=button_type):
                                    # 切换选择状态
                                    if is_selected:
                                        st.session_state.selected_options.pop(f"{question}_{option_key}", None)
                                        st.session_state.clarification_answers.pop(question, None)
                                    else:
                                        # 清除该问题的其他选项
                                        for key in list(st.session_state.selected_options.keys()):
                                            if key.startswith(f"{question}_"):
                                                st.session_state.selected_options.pop(key, None)
                                        st.session_state.selected_options[f"{question}_{option_key}"] = option
                                        st.session_state.clarification_answers[question] = f"{option_key}：{option}"
                                    st.rerun()
                
                # 仍然提供文本输入作为备选
                st.markdown("**或手动输入：**")
                custom_answer = st.text_input(
                    f"手动输入回答",
                    key=f"custom_{i}",
                    placeholder="如果以上选项不合适，请在此输入...",
                    value=current_answer if current_answer and "：" not in current_answer else ""
                )
                if custom_answer.strip():
                    # 如果手动输入，清除选项选择
                    for key in list(st.session_state.selected_options.keys()):
                        if key.startswith(f"{question}_"):
                            st.session_state.selected_options.pop(key, None)
                    st.session_state.clarification_answers[question] = custom_answer
                
                st.markdown("---")
        
        # 确认按钮
        col_confirm1, col_confirm2 = st.columns([1, 1])
        with col_confirm1:
            if st.button("✅ 确认并开始把关", type="primary", use_container_width=True):
                # 细化问题
                refined_question = st.session_state.workshop.refine_question_with_context(
                    st.session_state.refined_question, 
                    st.session_state.clarification_answers
                )
                
                # 开始把关
                st.session_state.debate_state = st.session_state.workshop.initialize_debate(refined_question)
                st.session_state.debate_in_progress = True
                st.session_state.needs_clarification = False
                
                # 立即执行一轮把关
                if st.session_state.ultra_fast_mode:
                    progress_bar = st.progress(0)
                    status_text = st.empty()
                    
                    # 执行所有轮次
                    for round_num in range(st.session_state.debate_state["max_rounds"]):
                        progress = (round_num + 1) / st.session_state.debate_state["max_rounds"]
                        status_text.text(f"正在处理第 {round_num + 1} 轮...")
                        progress_bar.progress(progress)
                        
                        # 执行一轮把关
                        st.session_state.debate_state = st.session_state.workshop.conduct_fast_debate(
                            st.session_state.debate_state
                        )
                    
                    status_text.text("✅ 把关完成！")
                    progress_bar.progress(1.0)
                    st.session_state.debate_in_progress = False
                
                st.rerun()
        
        with col_confirm2:
            if st.button("🔄 跳过细化", use_container_width=True):
                # 直接开始把关，不细化问题
                st.session_state.debate_state = st.session_state.workshop.initialize_debate(st.session_state.refined_question)
                st.session_state.debate_in_progress = True
                st.session_state.needs_clarification = False
                
                # 立即执行一轮把关
                if st.session_state.ultra_fast_mode:
                    progress_bar = st.progress(0)
                    status_text = st.empty()
                    
                    # 执行所有轮次
                    for round_num in range(st.session_state.debate_state["max_rounds"]):
                        progress = (round_num + 1) / st.session_state.debate_state["max_rounds"]
                        status_text.text(f"正在处理第 {round_num + 1} 轮...")
                        progress_bar.progress(progress)
                        
                        # 执行一轮把关
                        st.session_state.debate_state = st.session_state.workshop.conduct_fast_debate(
                            st.session_state.debate_state
                        )
                    
                    status_text.text("✅ 把关完成！")
                    progress_bar.progress(1.0)
                    st.session_state.debate_in_progress = False
                
                st.rerun()
    
    st.markdown("---")
    
    # 侧边栏 - 简化控制面板
    with st.sidebar:
        st.header("📋 快速决策模板")
        
        templates = [
            "产品功能优先级排序",
            "技术方案选择评估", 
            "市场进入策略分析",
            "团队资源分配决策"
        ]
        
        for template in templates:
            if st.button(template, use_container_width=True):
                st.session_state.debate_state = st.session_state.workshop.initialize_debate(template)
                st.session_state.debate_in_progress = True
                st.rerun()
        
        st.markdown("---")
        st.markdown("### 📊 性能指标")
        st.markdown("""
        - **响应速度**: 3-6秒
        - **缓存命中率**: 重复问题秒级响应
        - **并行处理**: 4个代理同时工作
        """)
    
    # 把关结果显示区域
    if st.session_state.debate_state:
        st.markdown("### 🗣️ 经理人把关过程")
        
        # 显示把关进度
        progress = min(st.session_state.debate_state["rounds"] / st.session_state.debate_state["max_rounds"], 1.0)
        st.progress(progress, text=f"把关进度：{st.session_state.debate_state['rounds']}/{st.session_state.debate_state['max_rounds']}轮")
        
        # 显示把关历史
        for entry in st.session_state.debate_state["debate_history"]:
            with st.chat_message("user", avatar=f"👤"):
                st.markdown(f"**{entry['speaker']}** ({entry['timestamp']})")
                st.write(entry["content"])
        
        # 显示当前状态
        if st.session_state.debate_in_progress:
            st.info(f"🔄 等待 {st.session_state.debate_state['current_speaker']} 发言...")
        else:
            st.success("✅ 把关已完成！")
            
            # 生成总结按钮
            col_sum1, col_sum2 = st.columns([2, 1])
            with col_sum1:
                if st.button("📊 生成决策总结和推荐度", type="secondary", use_container_width=True):
                    with st.spinner("正在生成决策要点和推荐度..."):
                        summary_data = st.session_state.workshop.generate_decision_summary(
                            st.session_state.debate_state["debate_history"]
                        )
                        st.session_state.summary_data = summary_data
                        st.rerun()
            
            # 显示推荐度 - 确保总是显示
            if "summary_data" in st.session_state:
                summary_data = st.session_state.summary_data
                
                # 显示推荐度
                st.markdown("### 🎯 决策推荐度")
                
                # 根据推荐度显示不同的颜色和图标
                recommendation_level = summary_data["recommendation_level"]
                stars = summary_data["stars"]
                
                if recommendation_level >= 4:
                    st.success(f"🎯 推荐度：{stars} (强烈推荐)")
                elif recommendation_level == 3:
                    st.info(f"⚖️ 推荐度：{stars} (中立)")
                elif recommendation_level == 2:
                    st.warning(f"⚠️ 推荐度：{stars} (谨慎)")
                else:
                    st.error(f"❌ 推荐度：{stars} (不推荐)")
                
                # 显示详细总结
                st.markdown("### 📋 决策要点总结")
                st.write(summary_data["summary"])
    else:
        # 初始状态显示
        st.markdown("### 💡 欢迎使用AI经理人为你把关")
        st.markdown("""
        这是一个基于AI的专业决策辅助系统，四位AI经理人将从不同专业角度为您把关决策：
        
        **👥 参与经理人：**
        - **技术专家** - 技术可行性、实现难度
        - **战略顾问** - 战略规划、竞争优势  
        - **用户体验师** - 用户需求、易用性
        - **风险顾问** - 潜在风险、不确定性
        
        **🚀 使用方法：**
        1. 在上方输入您的决策问题
        2. 点击"开始把关"按钮
        3. 观察经理人们的专业分析
        4. 生成综合决策总结和推荐度
        
        **⚡ 特色功能：**
        - 五位推荐度评级系统
        - 具体实施建议和时间规划
        - 超快速响应（3-6秒）
        - 智能缓存技术
        """)
        
        # 显示示例问题
        st.markdown("### 📝 示例问题")
        examples = [
            "我们是否应该投资AI技术？",
            "是否应该开发新的移动应用？",
            "如何优化现有产品的用户体验？",
            "团队资源应该如何分配？"
        ]
        
        for example in examples:
            if st.button(example, use_container_width=True):
                st.session_state.debate_state = st.session_state.workshop.initialize_debate(example)
                st.session_state.debate_in_progress = True
                st.rerun()

# 启动应用
if __name__ == "__main__":
    main()