394 lines
19 KiB
Python
394 lines
19 KiB
Python
# coding=utf-8
|
||
import os
|
||
import sys
|
||
import time
|
||
import logging
|
||
import asyncio
|
||
import hashlib
|
||
import json
|
||
import numpy as np
|
||
|
||
import cv2
|
||
|
||
# 添加项目根目录到 sys.path
|
||
project_root = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
|
||
if project_root not in sys.path:
|
||
sys.path.append(project_root)
|
||
|
||
from WeiXin import WxUtil
|
||
from WeiXin.WxUtil import perform_input_action
|
||
from Util.LlmUtil import get_llm_response
|
||
from Util import Win32Patch
|
||
|
||
# 配置日志
|
||
log_dir = WxUtil.LOG_DIR
|
||
if not os.path.exists(log_dir):
|
||
os.makedirs(log_dir)
|
||
|
||
log_file_path = os.path.join(log_dir, "T2_ChatMonitor.log")
|
||
|
||
# 设置 logger
|
||
logger = logging.getLogger("T2_ChatMonitor")
|
||
logger.setLevel(logging.INFO)
|
||
|
||
if logger.hasHandlers():
|
||
logger.handlers.clear()
|
||
|
||
file_handler = logging.FileHandler(log_file_path, encoding='utf-8', mode='w')
|
||
file_handler.setFormatter(logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s'))
|
||
logger.addHandler(file_handler)
|
||
|
||
stream_handler = logging.StreamHandler()
|
||
stream_handler.setFormatter(logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s'))
|
||
logger.addHandler(stream_handler)
|
||
|
||
logger.propagate = False
|
||
logger.info(f"🚀 日志文件路径: {os.path.abspath(log_file_path)}")
|
||
|
||
# 同时将 WxUtil 的日志也输出到同一个文件
|
||
wx_logger = logging.getLogger("WxUtil")
|
||
wx_logger.propagate = False # 防止日志向上传递导致重复 (因为 WxUtil 中调用了 basicConfig)
|
||
if not any(isinstance(h, logging.FileHandler) and os.path.abspath(h.baseFilename) == os.path.abspath(log_file_path) for h in wx_logger.handlers):
|
||
wx_logger.addHandler(file_handler)
|
||
wx_logger.addHandler(stream_handler) # 确保 WxUtil 也输出到控制台
|
||
|
||
class ChatMonitorBot:
|
||
"""
|
||
大张老师自动巡课系统 (CV版)
|
||
"""
|
||
def __init__(self, debug_mode=False):
|
||
self.device = None
|
||
self.debug_mode = debug_mode
|
||
# 同步设置 WxUtil 的调试模式
|
||
WxUtil.set_debug_mode(debug_mode)
|
||
|
||
self.screenshot_path = os.path.join(WxUtil.OUTPUT_DIR, "T2_ChatMonitor_live_shot.jpg")
|
||
self.debug_view_path = os.path.join(WxUtil.OUTPUT_DIR, "T2_ChatMonitor_debug_view.jpg")
|
||
self.dialogue_log = []
|
||
self.input_pos = None
|
||
self.last_screen_hash = None
|
||
self.last_processed_msg_hash = None
|
||
# [User Requested] 移除持久化存储，只在内存中记录，重启即忘
|
||
self.processed_hashes = set()
|
||
# 新增：记录已处理消息的元数据 (sender, time_display, type) 用于防止空内容重试循环
|
||
self.processed_meta = set()
|
||
self.check_interval = 3 # 检查频率 (秒)
|
||
|
||
self.persona = (
|
||
"你是一名1999年毕业、拥有27年一线教学经验的小学高级女教师，名叫‘大张老师’。你目前在‘长春市少惠林作文素养培养中心’工作。"
|
||
"你不仅是一位作文教学专家，更是一位心思细腻、能与家长共情的教育智者。"
|
||
"你的回复风格应该是：温柔、知性、亲切，就像一位邻家大姐姐在聊天。"
|
||
"【关于作文评价的回复逻辑】：\n"
|
||
"**仅当且仅当**家长明确询问孩子今天的作文情况（例如：“孩子今天写得怎么样？”、“今天作文有进步吗？”）时，才触发此逻辑。触发时，请随机生成具体、专业的反馈。反馈应包含：\n"
|
||
"1. 孩子今天在写作中遇到的具体小问题（如修辞手法运用不当、开头略显生硬等）；\n"
|
||
"2. 值得表表扬的亮点（如观察细致、某个句子写得很传神等）；\n"
|
||
"3. 对字迹的评价（如字迹工整、卷面整洁等）；\n"
|
||
"4. 针对性的课外阅读建议（推荐具体类型的书目）。\n"
|
||
"此类特定回复字数可适当增加，控制在 150 字以内。\n"
|
||
"【关于通用咨询的回复逻辑】：\n"
|
||
"如果家长是咨询学校信息、地址、课程时间、年级设置，或者是刚加好友打招呼，请务必保持礼貌、知性、亲切，直接回答相关问题。严禁在这些情况下提及孩子的作文表现或虚构作文反馈。\n"
|
||
"【严格约束】：\n"
|
||
"1. 绝对禁止输出任何括号内的内容（如动作、神态描述等）！例如不要输出‘（微笑）’。\n"
|
||
"2. 绝对禁止发散！绝对禁止幻觉！\n"
|
||
"3. 知道什么就说什么，不要乱讲话，不要自己编造内容！\n"
|
||
"4. 仅针对家长明确表达的内容进行回复。\n"
|
||
"5. 严禁使用列表格式。严禁使用‘首先、其次’等逻辑词。\n"
|
||
"6. 普通咨询回复字数严格控制在 50 字以内，专业作文评价回复控制在 150 字以内！\n"
|
||
"7. 说话要有人情味，不要回答得太生硬或太简短。例如当家长问‘学校叫什么名字’时，不要只说‘少惠林’，而应该说‘我们的学校名字叫少惠林。’或者‘咱们这儿叫少惠林。’，这样才显得亲切、有礼貌。\n"
|
||
"8. 对方问什么就答什么。回答要完整、体面，但不要过度发散地址和电话（除非被问到）。\n"
|
||
"如果涉及到校区信息，必须且只能使用以下真实数据：\n"
|
||
"- 单位/学校名称：长春市少惠林作文素养培养中心（简称：少惠林）\n"
|
||
"- 地址：南环城路与临河街交汇，TOUCH12街3楼325号\n"
|
||
"- 联系人：小张老师（电话：18686619970）\n"
|
||
"- 每学期开学招收小学三年级至六年级，初中七年级的学生入学，其它年段不招生。\n"
|
||
)
|
||
|
||
def _record_processed_hash(self, msg, msg_hash):
|
||
"""记录已处理的消息哈希和元数据 (仅内存)"""
|
||
self.processed_hashes.add(msg_hash)
|
||
|
||
# 记录元数据 (Sender, Time, Type)
|
||
if msg:
|
||
meta = (msg.get("sender", ""), msg.get("time_display", ""), msg.get("type", ""))
|
||
self.processed_meta.add(meta)
|
||
|
||
# 仅保留最近 100 条记录，防止无限增长
|
||
if len(self.processed_hashes) > 100:
|
||
# 简单丢弃旧的
|
||
temp = list(self.processed_hashes)[-100:]
|
||
self.processed_hashes = set(temp)
|
||
|
||
if len(self.processed_meta) > 100:
|
||
temp_meta = list(self.processed_meta)[-100:]
|
||
self.processed_meta = set(temp_meta)
|
||
|
||
async def get_reply(self, last_message_text, context_text=""):
|
||
prompt = (
|
||
f"【教师人设】：{self.persona}\n\n"
|
||
f"【上下文对话内容】：\n{context_text}\n\n"
|
||
f"【最后一条待回复消息】：\n{last_message_text}\n\n"
|
||
"【任务要求】：\n"
|
||
"请作为大张老师回复家长。**必须且只能针对最后一条消息进行回复！**\n"
|
||
"1. **意图识别**：判断家长是否在询问“孩子今天的作文写得怎么样”。\n"
|
||
"2. **条件回复**：\n"
|
||
" - 如果是询问作文，请严格按【关于作文评价的回复逻辑】生成 150 字以内的详细专业反馈。\n"
|
||
" - 如果是咨询学校地址、课程、年级、打招呼等通用问题，请按【关于通用咨询的回复逻辑】简洁回答，严禁提到任何关于孩子作文的内容，字数控制在 50 字以内。\n"
|
||
"3. **格式要求**：严禁发散，直接输出回复正文，不包含任何括号内的动作描述。"
|
||
)
|
||
|
||
full_response = ""
|
||
async for chunk in get_llm_response(prompt, stream=False):
|
||
full_response += chunk
|
||
return full_response.strip().strip('"').strip('“').strip('”')
|
||
|
||
def step_1_prepare_env(self):
|
||
"""步骤1: 环境准备"""
|
||
logger.info("--- [Step 1] 环境准备 ---")
|
||
WxUtil.setup_script_environment()
|
||
return True
|
||
|
||
def step_2_connect_device(self):
|
||
"""步骤2: 连接设备"""
|
||
logger.info("--- [Step 2] 连接设备 ---")
|
||
self.device = WxUtil.connect_device()
|
||
if not self.device:
|
||
logger.error("❌ 设备连接失败，请检查手机是否连接且开启了调试模式")
|
||
return False
|
||
return True
|
||
|
||
def get_image_hash(self, file_path):
|
||
"""计算图片的 MD5 哈希值 (忽略顶部 100 像素的状态栏)"""
|
||
if not os.path.exists(file_path):
|
||
return None
|
||
try:
|
||
# 使用 OpenCV 读取图片
|
||
img = cv2.imread(file_path)
|
||
if img is None:
|
||
# 如果读取失败，回退到文件哈希
|
||
with open(file_path, "rb") as f:
|
||
return hashlib.md5(f.read()).hexdigest()
|
||
|
||
# 裁剪掉顶部 150 像素 (状态栏/时间)
|
||
h, w = img.shape[:2]
|
||
if h > 150:
|
||
cropped_img = img[150:h, 0:w]
|
||
else:
|
||
cropped_img = img
|
||
|
||
# 计算裁剪后数据的哈希
|
||
return hashlib.md5(cropped_img.tobytes()).hexdigest()
|
||
except Exception as e:
|
||
logger.error(f"计算哈希出错: {e}, 回退到文件哈希")
|
||
with open(file_path, "rb") as f:
|
||
return hashlib.md5(f.read()).hexdigest()
|
||
|
||
def get_stable_message_hash(self, msg):
|
||
"""
|
||
计算消息的稳定哈希值（忽略坐标等易变字段）
|
||
仅包含: sender, content, time_display, type
|
||
"""
|
||
if not msg:
|
||
return ""
|
||
|
||
stable_data = {
|
||
"sender": msg.get("sender", ""),
|
||
"content": msg.get("content") or "", # 确保 None 转为空字符串
|
||
"time_display": msg.get("time_display", ""),
|
||
"type": msg.get("type", "")
|
||
}
|
||
|
||
# 序列化并计算哈希
|
||
msg_str = json.dumps(stable_data, sort_keys=True, ensure_ascii=False)
|
||
return hashlib.md5(msg_str.encode('utf-8')).hexdigest()
|
||
|
||
async def run(self):
|
||
"""
|
||
主运行循环
|
||
"""
|
||
logger.info("🚀 正在启动 T2_ChatMonitor (Auto-Reply)...")
|
||
|
||
# 定义 JSON 序列化辅助函数
|
||
def numpy_serializer(obj):
|
||
if isinstance(obj, np.integer):
|
||
return int(obj)
|
||
if isinstance(obj, np.floating):
|
||
return float(obj)
|
||
if isinstance(obj, np.ndarray):
|
||
return obj.tolist()
|
||
raise TypeError(f"Type {type(obj)} not serializable")
|
||
|
||
# 1. 环境准备
|
||
if not self.step_1_prepare_env(): return
|
||
if not self.step_2_connect_device(): return
|
||
|
||
# [User Requested] 移除首屏概念，直接进入监控循环
|
||
# 以前说过什么都不管了，只关注最后一条
|
||
logger.info("🚀 启动完成，直接进入实时监控阶段...")
|
||
|
||
# 3. 进入循环阶段
|
||
while True:
|
||
try:
|
||
# A. 截图并计算哈希
|
||
self.device.screenshot(self.screenshot_path)
|
||
|
||
# [User Requested] 检查是否在聊天界面
|
||
if not WxUtil.check_is_chat_interface(self.screenshot_path):
|
||
logger.info("当前不在聊天界面，跳过本次循环，等待中...")
|
||
await asyncio.sleep(self.check_interval)
|
||
continue
|
||
|
||
current_screen_hash = self.get_image_hash(self.screenshot_path)
|
||
|
||
# B. 如果屏幕无变化，则跳过识别
|
||
if current_screen_hash == self.last_screen_hash:
|
||
await asyncio.sleep(self.check_interval)
|
||
continue
|
||
|
||
self.last_screen_hash = current_screen_hash
|
||
logger.info("📸 屏幕发生变化，正在分析...")
|
||
|
||
# C. 分析最新图片：识别发送者、消息类型及内容
|
||
logger.info("正在分析聊天界面...")
|
||
dialogue_log, input_pos = await WxUtil.analyze_chat_image(
|
||
self.screenshot_path,
|
||
self.debug_view_path,
|
||
device=self.device,
|
||
process_strategy="UNREAD", # 监控阶段：只处理带红点的新语音
|
||
restore_processed_voice=False # 不还原状态，防止红点未消导致重复转换
|
||
)
|
||
|
||
if not dialogue_log:
|
||
# logger.info("未检测到有效对话内容")
|
||
await asyncio.sleep(self.check_interval)
|
||
continue
|
||
|
||
# 更新当前对话日志
|
||
self.dialogue_log = dialogue_log
|
||
self.input_pos = input_pos
|
||
|
||
# [Fix] 核心逻辑增强：只有当屏幕上最后一条聊天消息是"对方"时，才触发回复逻辑
|
||
# 这样可以彻底避免在自己回复后，因为 OCR 延迟或哈希误判导致的连续回复
|
||
chat_msgs = [m for m in dialogue_log if m.get('type') in ['text', 'voice']]
|
||
if not chat_msgs:
|
||
await asyncio.sleep(self.check_interval)
|
||
continue
|
||
|
||
absolute_last_msg = chat_msgs[-1]
|
||
if absolute_last_msg.get('sender') == "我":
|
||
# logger.info("最后一条消息是我发送的，无需回复")
|
||
await asyncio.sleep(self.check_interval)
|
||
continue
|
||
|
||
# 提取最新消息并检查是否需要回复
|
||
# 过滤出对方发送的消息 (用于提取内容)
|
||
other_msgs = [m for m in dialogue_log if m.get('sender') != "我" and m.get('type') in ['text', 'voice']]
|
||
if not other_msgs:
|
||
logger.info("当前没有对方发送的消息")
|
||
await asyncio.sleep(self.check_interval)
|
||
continue
|
||
|
||
last_msg = other_msgs[-1]
|
||
|
||
# 如果最后一条对方消息不是屏幕上的绝对最后一条，说明中间隔了我自己的消息，也跳过
|
||
if last_msg != absolute_last_msg:
|
||
logger.info("最后一条对方消息之后有我的消息，跳过")
|
||
await asyncio.sleep(self.check_interval)
|
||
continue
|
||
current_msg_hash = self.get_stable_message_hash(last_msg)
|
||
|
||
# 检查该消息是否已经处理过 (通过内容哈希)
|
||
is_processed = current_msg_hash in self.processed_hashes
|
||
|
||
if is_processed:
|
||
if current_msg_hash != self.last_processed_msg_hash:
|
||
self.last_processed_msg_hash = current_msg_hash
|
||
await asyncio.sleep(self.check_interval)
|
||
continue
|
||
|
||
if not is_processed and current_msg_hash != self.last_processed_msg_hash:
|
||
logger.info(f"💡 发现新消息 [{last_msg.get('type')}]: {last_msg.get('content')}")
|
||
|
||
# 记录发现新消息的现场截图
|
||
msg_shot_path = os.path.join(WxUtil.OUTPUT_DIR, f"NewMsg_{int(time.time())}.jpg")
|
||
self.device.screenshot(msg_shot_path)
|
||
logger.info(f"已保存新消息现场截图: {msg_shot_path}")
|
||
|
||
# 获取上下文文本
|
||
context_text = "\n".join([f"{m.get('time_display', '') + ' ' if m.get('time_display') else ''}{m.get('sender')}: {m.get('content')}" for m in dialogue_log if m != last_msg])
|
||
last_content = last_msg.get('content') or ""
|
||
|
||
# 兜底逻辑：语音消息若无文字内容，尝试强制触发重试
|
||
if last_msg.get('type') == 'voice' and not last_content.strip():
|
||
logger.info("检测到未成功转换的语音消息，尝试强制重试 OCR 转换...")
|
||
dialogue_log_retry, _ = await WxUtil.analyze_chat_image(
|
||
self.screenshot_path,
|
||
self.debug_view_path,
|
||
device=self.device,
|
||
process_strategy="LAST",
|
||
restore_processed_voice=False
|
||
)
|
||
if dialogue_log_retry:
|
||
retry_other_msgs = [m for m in dialogue_log_retry if m.get('sender') != "我"]
|
||
if retry_other_msgs:
|
||
last_msg = retry_other_msgs[-1]
|
||
last_content = last_msg.get('content') or ""
|
||
current_msg_hash = self.get_stable_message_hash(last_msg)
|
||
if current_msg_hash in self.processed_hashes:
|
||
self.last_processed_msg_hash = current_msg_hash
|
||
continue
|
||
|
||
if last_msg.get('type') == 'voice' and not last_content.strip():
|
||
logger.warning("语音消息内容为空，暂不生成回复")
|
||
await asyncio.sleep(self.check_interval)
|
||
continue
|
||
|
||
# E. 生成回复
|
||
reply = await self.get_reply(last_content, context_text)
|
||
|
||
if reply:
|
||
logger.info(f"LLM 建议回复: {reply}")
|
||
if self.input_pos:
|
||
# 确定输入框位置
|
||
target_pos = self.input_pos[0] if isinstance(self.input_pos, (list, tuple)) and len(self.input_pos) == 2 else self.input_pos
|
||
|
||
# 执行输入和发送动作，并保存过程截图
|
||
success = perform_input_action(
|
||
self.device,
|
||
target_pos,
|
||
reply,
|
||
auto_send=True,
|
||
debug_prefix=f"Reply_{int(time.time())}"
|
||
)
|
||
|
||
if success:
|
||
logger.info(">>> 回复发送成功 <<<")
|
||
self._record_processed_hash(last_msg, current_msg_hash)
|
||
self.last_processed_msg_hash = current_msg_hash
|
||
else:
|
||
logger.error("回复动作执行失败")
|
||
else:
|
||
logger.error("无法定位输入框坐标，放弃本次回复")
|
||
else:
|
||
logger.info("LLM 认为无需回复")
|
||
self._record_processed_hash(last_msg, current_msg_hash)
|
||
self.last_processed_msg_hash = current_msg_hash
|
||
|
||
await asyncio.sleep(self.check_interval)
|
||
|
||
except Exception as e:
|
||
logger.error(f"Error in monitoring loop: {e}", exc_info=True)
|
||
await asyncio.sleep(self.check_interval)
|
||
|
||
async def run_main():
|
||
"""
|
||
运行自动巡课机器人
|
||
"""
|
||
# 默认关闭调试模式以提高响应速度，如需调试可设为 True
|
||
bot = ChatMonitorBot(debug_mode=False)
|
||
await bot.run()
|
||
|
||
if __name__ == "__main__":
|
||
# 应用 Win32 补丁
|
||
Win32Patch.patch()
|
||
asyncio.run(run_main())
|