from .observation import Observation, ChattingObservation import asyncio from src.plugins.moods.moods import MoodManager from src.plugins.models.utils_model import LLMRequest from src.config.config import global_config import time from typing import Optional from datetime import datetime import traceback from src.plugins.chat.message import UserInfo from src.plugins.chat.utils import parse_text_timestamps # from src.plugins.schedule.schedule_generator import bot_schedule # from src.plugins.memory_system.Hippocampus import HippocampusManager from src.common.logger import get_module_logger, LogConfig, SUB_HEARTFLOW_STYLE_CONFIG # noqa: E402 # from src.plugins.chat.utils import get_embedding # from src.common.database import db # from typing import Union from src.individuality.individuality import Individuality import random from src.plugins.chat.chat_stream import ChatStream from src.plugins.person_info.relationship_manager import relationship_manager from src.plugins.chat.utils import get_recent_group_speaker from ..plugins.utils.prompt_builder import Prompt, global_prompt_manager subheartflow_config = LogConfig( # 使用海马体专用样式 console_format=SUB_HEARTFLOW_STYLE_CONFIG["console_format"], file_format=SUB_HEARTFLOW_STYLE_CONFIG["file_format"], ) logger = get_module_logger("subheartflow", config=subheartflow_config) def init_prompt(): prompt = "" # prompt += f"麦麦的总体想法是：{self.main_heartflow_info}\n\n" prompt += "{extra_info}\n" # prompt += "{prompt_schedule}\n" # prompt += "{relation_prompt_all}\n" prompt += "{prompt_personality}\n" prompt += "刚刚你的想法是{current_thinking_info}。可以适当转换话题\n" prompt += "-----------------------------------\n" prompt += "现在是{time_now}，你正在上网，和qq群里的网友们聊天，群里正在聊的话题是：\n{chat_observe_info}\n" prompt += "你现在{mood_info}\n" # prompt += "你注意到{sender_name}刚刚说：{message_txt}\n" prompt += "现在你接下去继续思考，产生新的想法，不要分点输出，输出连贯的内心独白" prompt += "思考时可以想想如何对群聊内容进行回复。回复的要求是：平淡一些，简短一些，说中文，尽量不要说你说过的话。如果你要回复，最好只回复一个人的一个话题\n" prompt += "请注意不要输出多余内容(包括前后缀，冒号和引号，括号， 表情，等)，不要带有括号和动作描写" prompt += "记得结合上述的消息，生成内心想法，文字不要浮夸，注意{bot_name}指的就是你。" Prompt(prompt, "sub_heartflow_prompt_before") prompt = "" # prompt += f"你现在正在做的事情是：{schedule_info}\n" prompt += "{extra_info}\n" prompt += "{prompt_personality}\n" prompt += "现在是{time_now}，你正在上网，和qq群里的网友们聊天，群里正在聊的话题是：\n{chat_observe_info}\n" prompt += "刚刚你的想法是{current_thinking_info}。" prompt += "你现在看到了网友们发的新消息:{message_new_info}\n" prompt += "你刚刚回复了群友们:{reply_info}" prompt += "你现在{mood_info}" prompt += "现在你接下去继续思考，产生新的想法，记得保留你刚刚的想法，不要分点输出，输出连贯的内心独白" prompt += "不要太长，但是记得结合上述的消息，要记得你的人设，关注聊天和新内容，关注你回复的内容，不要思考太多:" Prompt(prompt, "sub_heartflow_prompt_after") # prompt += f"你现在正在做的事情是：{schedule_info}\n" prompt += "{extra_info}\n" prompt += "{prompt_personality}\n" prompt += "现在是{time_now}，你正在上网，和qq群里的网友们聊天，群里正在聊的话题是：\n{chat_observe_info}\n" prompt += "刚刚你的想法是{current_thinking_info}。" prompt += "你现在看到了网友们发的新消息:{message_new_info}\n" # prompt += "你刚刚回复了群友们:{reply_info}" prompt += "你现在{mood_info}" prompt += "现在你接下去继续思考，产生新的想法，记得保留你刚刚的想法，不要分点输出，输出连贯的内心独白" prompt += "不要思考太多，不要输出多余内容(包括前后缀，冒号和引号，括号， 表情，等)，不要带有括号和动作描写" prompt += "记得结合上述的消息，生成内心想法，文字不要浮夸，注意{bot_name}指的就是你。" Prompt(prompt, "sub_heartflow_prompt_after_observe") class CurrentState: def __init__(self): self.willing = 0 self.current_state_info = "" self.mood_manager = MoodManager() self.mood = self.mood_manager.get_prompt() def update_current_state_info(self): self.current_state_info = self.mood_manager.get_current_mood() class SubHeartflow: def __init__(self, subheartflow_id): self.subheartflow_id = subheartflow_id self.current_mind = "" self.past_mind = [] self.current_state: CurrentState = CurrentState() self.llm_model = LLMRequest( model=global_config.llm_sub_heartflow, temperature=global_config.llm_sub_heartflow["temp"], max_tokens=600, request_type="sub_heart_flow", ) self.main_heartflow_info = "" self.last_reply_time = time.time() self.last_active_time = time.time() # 添加最后激活时间 if not self.current_mind: self.current_mind = "你什么也没想" self.is_active = False self.observations: list[ChattingObservation] = [] self.running_knowledges = [] self._thinking_lock = asyncio.Lock() # 添加思考锁，防止并发思考 self.bot_name = global_config.BOT_NICKNAME def add_observation(self, observation: Observation): """添加一个新的observation对象到列表中，如果已存在相同id的observation则不添加""" # 查找是否存在相同id的observation for existing_obs in self.observations: if existing_obs.observe_id == observation.observe_id: # 如果找到相同id的observation，直接返回 return # 如果没有找到相同id的observation，则添加新的 self.observations.append(observation) def remove_observation(self, observation: Observation): """从列表中移除一个observation对象""" if observation in self.observations: self.observations.remove(observation) def get_all_observations(self) -> list[Observation]: """获取所有observation对象""" return self.observations def clear_observations(self): """清空所有observation对象""" self.observations.clear() def _get_primary_observation(self) -> Optional[ChattingObservation]: """获取主要的（通常是第一个）ChattingObservation实例""" if self.observations and isinstance(self.observations[0], ChattingObservation): return self.observations[0] logger.warning(f"SubHeartflow {self.subheartflow_id} 没有找到有效的 ChattingObservation") return None async def subheartflow_start_working(self): while True: current_time = time.time() # --- 调整后台任务逻辑 --- # # 这个后台循环现在主要负责检查是否需要自我销毁 # 不再主动进行思考或状态更新，这些由 HeartFC_Chat 驱动 # 检查是否需要冻结（这个逻辑可能需要重新审视，因为激活状态现在由外部驱动） # if current_time - self.last_reply_time > global_config.sub_heart_flow_freeze_time: # self.is_active = False # else: # self.is_active = True # self.last_active_time = current_time # 由外部调用（如 thinking）更新 # 检查是否超过指定时间没有激活 (例如，没有被调用进行思考) if current_time - self.last_active_time > global_config.sub_heart_flow_stop_time: # 例如 5 分钟 logger.info(f"子心流 {self.subheartflow_id} 超过 {global_config.sub_heart_flow_stop_time} 秒没有激活，正在销毁..." f" (Last active: {datetime.fromtimestamp(self.last_active_time).strftime('%Y-%m-%d %H:%M:%S')})") # 在这里添加实际的销毁逻辑，例如从主 Heartflow 管理器中移除自身 # heartflow.remove_subheartflow(self.subheartflow_id) # 假设有这样的方法 break # 退出循环以停止任务 # 不再需要内部驱动的状态更新和思考 # self.current_state.update_current_state_info() # await self.do_a_thinking() # await self.judge_willing() await asyncio.sleep(global_config.sub_heart_flow_update_interval) # 定期检查销毁条件 async def ensure_observed(self): """确保在思考前执行了观察""" observation = self._get_primary_observation() if observation: try: await observation.observe() logger.trace(f"[{self.subheartflow_id}] Observation updated before thinking.") except Exception as e: logger.error(f"[{self.subheartflow_id}] Error during pre-thinking observation: {e}") logger.error(traceback.format_exc()) async def do_observe(self): # 现在推荐使用 ensure_observed()，但保留此方法以兼容旧用法（或特定场景） observation = self._get_primary_observation() if observation: await observation.observe() else: logger.error(f"[{self.subheartflow_id}] do_observe called but no valid observation found.") async def do_thinking_before_reply( self, chat_stream: ChatStream, extra_info: str, obs_id: list[str] = None # 修改 obs_id 类型为 list[str] ): async with self._thinking_lock: # 获取思考锁 # --- 在思考前确保观察已执行 --- # await self.ensure_observed() self.last_active_time = time.time() # 更新最后激活时间戳 current_thinking_info = self.current_mind mood_info = self.current_state.mood observation = self._get_primary_observation() if not observation: logger.error(f"[{self.subheartflow_id}] Cannot perform thinking without observation.") return "", [] # 返回空结果 # --- 获取观察信息 --- # chat_observe_info = "" if obs_id: try: chat_observe_info = observation.get_observe_info(obs_id) logger.debug(f"[{self.subheartflow_id}] Using specific observation IDs: {obs_id}") except Exception as e: logger.error(f"[{self.subheartflow_id}] Error getting observe info with IDs {obs_id}: {e}. Falling back.") chat_observe_info = observation.get_observe_info() # 出错时回退到默认观察 else: chat_observe_info = observation.get_observe_info() logger.debug(f"[{self.subheartflow_id}] Using default observation info.") # --- 构建 Prompt (基本逻辑不变) --- # extra_info_prompt = "" if extra_info: for tool_name, tool_data in extra_info.items(): extra_info_prompt += f"{tool_name} 相关信息:\n" for item in tool_data: extra_info_prompt += f"- {item['name']}: {item['content']}\n" else: extra_info_prompt = "无工具信息。\n" # 提供默认值 individuality = Individuality.get_instance() prompt_personality = f"你的名字是{self.bot_name},你" prompt_personality += individuality.personality.personality_core personality_sides = individuality.personality.personality_sides if personality_sides: random.shuffle(personality_sides); prompt_personality += f",{personality_sides[0]}" identity_detail = individuality.identity.identity_detail if identity_detail: random.shuffle(identity_detail); prompt_personality += f",{identity_detail[0]}" # who_chat_in_group = [ # (chat_stream.platform, sender_info.user_id, sender_info.user_nickname) # 先添加当前发送者 # ] # # 获取最近发言者，排除当前发送者，避免重复 # recent_speakers = get_recent_group_speaker( # chat_stream.stream_id, # (chat_stream.platform, sender_info.user_id), # limit=global_config.MAX_CONTEXT_SIZE -1 # 减去当前发送者 # ) # who_chat_in_group.extend(recent_speakers) # relation_prompt = "" # unique_speakers = set() # 确保人物信息不重复 # for person_tuple in who_chat_in_group: # person_key = (person_tuple[0], person_tuple[1]) # 使用 platform+id 作为唯一标识 # if person_key not in unique_speakers: # relation_prompt += await relationship_manager.build_relationship_info(person_tuple) # unique_speakers.add(person_key) # relation_prompt_all = (await global_prompt_manager.get_prompt_async("relationship_prompt")).format( # relation_prompt, sender_info.user_nickname # ) # sender_name_sign = ( # f"<{chat_stream.platform}:{sender_info.user_id}:{sender_info.user_nickname}:{sender_info.user_cardname or 'NoCard'}>" # ) time_now = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime()) prompt = (await global_prompt_manager.get_prompt_async("sub_heartflow_prompt_before")).format( extra_info=extra_info_prompt, # relation_prompt_all=relation_prompt_all, prompt_personality=prompt_personality, current_thinking_info=current_thinking_info, time_now=time_now, chat_observe_info=chat_observe_info, mood_info=mood_info, # sender_name=sender_name_sign, # message_txt=message_txt, bot_name=self.bot_name, ) prompt = await relationship_manager.convert_all_person_sign_to_person_name(prompt) prompt = parse_text_timestamps(prompt, mode="lite") logger.debug(f"[{self.subheartflow_id}] Thinking Prompt:\n{prompt}") try: response, reasoning_content = await self.llm_model.generate_response_async(prompt) if not response: # 如果 LLM 返回空，给一个默认想法 response = "(不知道该想些什么...)" logger.warning(f"[{self.subheartflow_id}] LLM returned empty response for thinking.") except Exception as e: logger.error(f"[{self.subheartflow_id}] 内心独白获取失败: {e}") response = "(思考时发生错误...)" # 错误时的默认想法 self.update_current_mind(response) # self.current_mind 已经在 update_current_mind 中更新 logger.info(f"[{self.subheartflow_id}] 思考前脑内状态：{self.current_mind}") return self.current_mind, self.past_mind async def do_thinking_after_observe( self, message_txt: str, sender_info: UserInfo, chat_stream: ChatStream, extra_info: str, obs_id: int = None ): current_thinking_info = self.current_mind mood_info = self.current_state.mood # mood_info = "你很生气，很愤怒" observation = self.observations[0] if obs_id: print(f"11111111111有id,开始获取观察信息{obs_id}") chat_observe_info = observation.get_observe_info(obs_id) else: chat_observe_info = observation.get_observe_info() extra_info_prompt = "" for tool_name, tool_data in extra_info.items(): extra_info_prompt += f"{tool_name} 相关信息:\n" for item in tool_data: extra_info_prompt += f"- {item['name']}: {item['content']}\n" # 开始构建prompt prompt_personality = f"你的名字是{self.bot_name},你" # person individuality = Individuality.get_instance() personality_core = individuality.personality.personality_core prompt_personality += personality_core personality_sides = individuality.personality.personality_sides random.shuffle(personality_sides) prompt_personality += f",{personality_sides[0]}" identity_detail = individuality.identity.identity_detail random.shuffle(identity_detail) prompt_personality += f",{identity_detail[0]}" # 关系 who_chat_in_group = [ (chat_stream.user_info.platform, chat_stream.user_info.user_id, chat_stream.user_info.user_nickname) ] who_chat_in_group += get_recent_group_speaker( chat_stream.stream_id, (chat_stream.user_info.platform, chat_stream.user_info.user_id), limit=global_config.MAX_CONTEXT_SIZE, ) relation_prompt = "" for person in who_chat_in_group: relation_prompt += await relationship_manager.build_relationship_info(person) # relation_prompt_all = ( # f"{relation_prompt}关系等级越大，关系越好，请分析聊天记录，" # f"根据你和说话者{sender_name}的关系和态度进行回复，明确你的立场和情感。" # ) relation_prompt_all = (await global_prompt_manager.get_prompt_async("relationship_prompt")).format( relation_prompt, sender_info.user_nickname ) sender_name_sign = ( f"<{chat_stream.platform}:{sender_info.user_id}:{sender_info.user_nickname}:{sender_info.user_cardname}>" ) time_now = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime()) prompt = (await global_prompt_manager.get_prompt_async("sub_heartflow_prompt_after_observe")).format( extra_info_prompt, # prompt_schedule, relation_prompt_all, prompt_personality, current_thinking_info, time_now, chat_observe_info, mood_info, sender_name_sign, message_txt, self.bot_name, ) prompt = await relationship_manager.convert_all_person_sign_to_person_name(prompt) prompt = parse_text_timestamps(prompt, mode="lite") try: response, reasoning_content = await self.llm_model.generate_response_async(prompt) except Exception as e: logger.error(f"回复前内心独白获取失败: {e}") response = "" self.update_current_mind(response) self.current_mind = response logger.info(f"prompt:\n{prompt}\n") logger.info(f"麦麦的思考前脑内状态：{self.current_mind}") return self.current_mind, self.past_mind # async def do_thinking_after_reply(self, reply_content, chat_talking_prompt, extra_info): # # print("麦麦回复之后脑袋转起来了") # # 开始构建prompt # prompt_personality = f"你的名字是{self.bot_name},你" # # person # individuality = Individuality.get_instance() # personality_core = individuality.personality.personality_core # prompt_personality += personality_core # extra_info_prompt = "" # for tool_name, tool_data in extra_info.items(): # extra_info_prompt += f"{tool_name} 相关信息:\n" # for item in tool_data: # extra_info_prompt += f"- {item['name']}: {item['content']}\n" # personality_sides = individuality.personality.personality_sides # random.shuffle(personality_sides) # prompt_personality += f",{personality_sides[0]}" # identity_detail = individuality.identity.identity_detail # random.shuffle(identity_detail) # prompt_personality += f",{identity_detail[0]}" # current_thinking_info = self.current_mind # mood_info = self.current_state.mood # observation = self.observations[0] # chat_observe_info = observation.observe_info # message_new_info = chat_talking_prompt # reply_info = reply_content # time_now = time.strftime("%Y-%m-%d %H:%M:%S", time.localtime()) # prompt = (await global_prompt_manager.get_prompt_async("sub_heartflow_prompt_after")).format( # extra_info_prompt, # prompt_personality, # time_now, # chat_observe_info, # current_thinking_info, # message_new_info, # reply_info, # mood_info, # ) # prompt = await relationship_manager.convert_all_person_sign_to_person_name(prompt) # prompt = parse_text_timestamps(prompt, mode="lite") # try: # response, reasoning_content = await self.llm_model.generate_response_async(prompt) # except Exception as e: # logger.error(f"回复后内心独白获取失败: {e}") # response = "" # self.update_current_mind(response) # self.current_mind = response # logger.info(f"麦麦回复后的脑内状态：{self.current_mind}") # self.last_reply_time = time.time() def update_current_mind(self, response): self.past_mind.append(self.current_mind) self.current_mind = response async def check_reply_trigger(self) -> bool: """根据观察到的信息和内部状态，判断是否应该触发一次回复。 TODO: 实现具体的判断逻辑。 例如：检查 self.observations[0].now_message_info 是否包含提及、问题， 或者 self.current_mind 中是否包含强烈的回复意图等。 """ # Placeholder: 目前始终返回 False，需要后续实现 logger.trace(f"[{self.subheartflow_id}] check_reply_trigger called. (Logic Pending)") # --- 实现触发逻辑 --- # # 示例：如果观察到的最新消息包含自己的名字，则有一定概率触发 # observation = self._get_primary_observation() # if observation and self.bot_name in observation.now_message_info[-100:]: # 检查最后100个字符 # if random.random() < 0.3: # 30% 概率触发 # logger.info(f"[{self.subheartflow_id}] Triggering reply based on mention.") # return True # ------------------ # return False # 默认不触发 init_prompt() # subheartflow = SubHeartflow()