团队协作（Team Collaboration）

"""
Prompt 版本管理 + Agent 行为评估的最小协作示例
无外部依赖，纯标准库实现
"""
from datetime import datetime
from typing import Dict, List, Optional

# ============ Prompt 版本管理 ============

class PromptVersion:
    """单个 Prompt 版本的元数据"""
    def __init__(self, prompt_id: str, content: str, version: str,
                 author: str, change_log: str):
        self.prompt_id = prompt_id
        self.content = content
        self.version = version
        self.author = author
        self.change_log = change_log
        self.created_at = datetime.now().isoformat()

class PromptManager:
    """Prompt 版本管理器：存储、更新、回溯"""
    def __init__(self):
        self._store: Dict[str, List[PromptVersion]] = {}

    def create(self, prompt_id: str, content: str,
               author: str, change_log: str = "初始版本") -> str:
        """创建新 Prompt，返回版本号"""
        ver = PromptVersion(prompt_id, content, "v1.0", author, change_log)
        self._store[prompt_id] = [ver]
        return ver.version

    def update(self, prompt_id: str, content: str,
               author: str, change_log: str) -> str:
        """更新 Prompt（追加新版本），返回新版本号"""
        versions = self._store.get(prompt_id)
        if not versions:
            raise KeyError(f"Prompt '{prompt_id}' 不存在")
        last = versions[-1]
        # 小版本号递增：v1.0 → v1.1 → v1.2 ...
        major, minor = last.version.split(".")
        new_ver_str = f"{major}.{int(minor) + 1}"
        new_ver = PromptVersion(prompt_id, content, new_ver_str,
                                author, change_log)
        versions.append(new_ver)
        return new_ver_str

    def get_latest(self, prompt_id: str) -> Optional[PromptVersion]:
        """获取最新版本"""
        versions = self._store.get(prompt_id)
        return versions[-1] if versions else None

    def rollback(self, prompt_id: str) -> str:
        """回滚到上一个版本，返回回滚后的版本号"""
        versions = self._store.get(prompt_id)
        if not versions or len(versions) < 2:
            raise ValueError("没有可回滚的历史版本")
        versions.pop()
        return versions[-1].version

# ============ Agent 行为评估 ============

class AgentEvaluator:
    """Agent 输出质量评估器"""

    @staticmethod
    def evaluate(response: str, expected: str,
                 latency_ms: float, user_score: float) -> Dict:
        """
        评估单次 Agent 响应，返回多维度指标。

        参数:
            response: Agent 实际输出
            expected: 期望输出（标准答案）
            latency_ms: 响应延迟（毫秒）
            user_score: 用户满意度评分（1-5）
        """
        resp_words = set(response.lower().split())
        exp_words = set(expected.lower().split())
        common = resp_words & exp_words

        precision = len(common) / len(resp_words) if resp_words else 0.0
        recall = len(common) / len(exp_words) if exp_words else 0.0
        success = 1.0 if (precision > 0.5 and recall > 0.5) else 0.0

        return {
            "success_rate": success,
            "precision": round(precision, 3),
            "recall": round(recall, 3),
            "latency_ms": latency_ms,
            "user_satisfaction": user_score,
        }

    @staticmethod
    def composite_score(metrics: Dict,
                        weights: Optional[Dict] = None) -> float:
        """加权综合得分（0-1）"""
        w = weights or {
            "success_rate": 0.35, "precision": 0.20,
            "recall": 0.15, "user_satisfaction": 0.15,
            "latency": 0.10, "cost": 0.05,
        }
        latency_score = max(0, 1 - metrics["latency_ms"] / 5000)
        sat_score = metrics["user_satisfaction"] / 5.0
        score = (
            metrics["success_rate"] * w["success_rate"]
            + metrics["precision"] * w["precision"]
            + metrics["recall"] * w["recall"]
            + sat_score * w["user_satisfaction"]
            + latency_score * w["latency"]
            + 0.95 * w["cost"]  # 假设单次成本很低
        )
        return round(score, 3)

# ============ 协作流程演示 ============

if __name__ == "__main__":
    pm = PromptManager()
    ev = AgentEvaluator()

    # 第一步：团队成员 A 创建初始 Prompt
    pm.create("cs-bot", "你是客服助手。回答订单、退货问题。", "张三")
    print(f"当前版本: {pm.get_latest('cs-bot').version}")

    # 第二步：评估 v1.0 效果
    m1 = ev.evaluate(
        response="订单已发货，预计3天到。",
        expected="订单号123已发货，预计3个工作日送达。",
        latency_ms=320, user_score=3.5
    )
    print(f"v1.0 综合得分: {ev.composite_score(m1)}")

    # 第三步：团队评审后，成员 B 优化 Prompt
    pm.update("cs-bot",
              "你是电商客服助手。回答订单、退货、退款问题。"
              "每个回答包含具体步骤。",
              "李四", "添加步骤要求，扩展退款场景")
    print(f"更新后版本: {pm.get_latest('cs-bot').version}")

    # 第四步：评估 v1.1 效果
    m2 = ev.evaluate(
        response="订单号123已发货，预计3个工作日送达。如需退货请按以下步骤操作。",
        expected="订单号123已发货，预计3个工作日送达。",
        latency_ms=280, user_score=4.5
    )
    print(f"v1.1 综合得分: {ev.composite_score(m2)}")

    # 如果新版本出了问题，一键回滚
    # rolled = pm.rollback("cs-bot")
    # print(f"回滚到: {rolled}")