'commit'
This commit is contained in:
HuangHai
@@ -222,7 +222,8 @@ class ChatMonitorBot:
|
||||
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
|
||||
# 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,
|
||||
@@ -232,46 +233,41 @@ class ChatMonitorBot:
|
||||
)
|
||||
|
||||
if not dialogue_log:
|
||||
logger.info("😴 未识别到有效消息")
|
||||
# logger.info("未检测到有效对话内容")
|
||||
await asyncio.sleep(self.check_interval)
|
||||
continue
|
||||
|
||||
logger.info(f"📊 当前识别到 {len(dialogue_log)} 条消息，最后一条: {dialogue_log[-1]}")
|
||||
|
||||
# 更新当前对话日志（可用于上下文参考）
|
||||
# 更新当前对话日志
|
||||
self.dialogue_log = dialogue_log
|
||||
self.input_pos = input_pos
|
||||
|
||||
# D. 只关注最后一条消息
|
||||
# D. 提取最新消息并检查是否需要回复
|
||||
last_msg = dialogue_log[-1]
|
||||
# 计算稳定哈希（忽略坐标变化）
|
||||
current_msg_hash = self.get_stable_message_hash(last_msg)
|
||||
|
||||
# E. 判断是否需要回复 (对方发送且非重复消息)
|
||||
sender = last_msg.get('sender', '')
|
||||
|
||||
# Check if hash is already processed (in-memory only)
|
||||
# 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
|
||||
|
||||
# Log only if it changed from last *in-memory* check to avoid spam
|
||||
if is_processed and current_msg_hash != self.last_processed_msg_hash:
|
||||
# logger.info(f"🚫 [监控] 消息哈希已存在于历史记录中，跳过回复 (Hash: {current_msg_hash})")
|
||||
self.last_processed_msg_hash = current_msg_hash
|
||||
|
||||
if not is_processed and current_msg_hash != self.last_processed_msg_hash:
|
||||
if sender != "我":
|
||||
event_shot = WxUtil.get_next_debug_path("event_new_msg")
|
||||
self.device.screenshot(event_shot)
|
||||
logger.info(f"💡 [监控] 发现新消息: {last_msg}，保存现场截图: {event_shot}")
|
||||
logger.info(f"💡 发现新消息 [{last_msg.get('type')}]: {last_msg.get('content')}")
|
||||
|
||||
# 获取上下文文本 (格式化为 Sender: 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[:-1]])
|
||||
last_content = last_msg.get('content') or ""
|
||||
|
||||
# 兜底逻辑：如果最后一条是语音且内容为空（可能因无红点未被 UNREAD 策略处理），尝试强制转换
|
||||
# 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("⚠️ [监控] 最后一条语音消息未获取到内容（可能已读无红点），尝试强制转换...")
|
||||
# 强制使用 LAST 策略重试，并且不还原状态（保持文字展开），防止下次循环因无内容再次触发重试
|
||||
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,
|
||||
@@ -280,66 +276,62 @@ class ChatMonitorBot:
|
||||
restore_processed_voice=False
|
||||
)
|
||||
if dialogue_log_retry:
|
||||
# 更新引用
|
||||
self.dialogue_log = dialogue_log_retry
|
||||
dialogue_log = dialogue_log_retry
|
||||
last_msg = dialogue_log[-1]
|
||||
last_msg = dialogue_log_retry[-1]
|
||||
last_content = last_msg.get('content') or ""
|
||||
logger.info(f"🔄 [重试] 强制转换后内容: {last_content}")
|
||||
|
||||
# 重新构建哈希
|
||||
current_msg_hash = self.get_stable_message_hash(last_msg)
|
||||
# 再次检查是否已处理 (因为内容变了，哈希变了)
|
||||
if current_msg_hash in self.processed_hashes:
|
||||
logger.info(f"🚫 [重试] 转换后发现该消息已处理，跳过。")
|
||||
self.last_processed_msg_hash = current_msg_hash
|
||||
# 跳过本次循环的剩余部分
|
||||
continue
|
||||
|
||||
# 生成回复
|
||||
# 语音消息若重试后仍无内容，暂不回复
|
||||
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"🤖 [监控] LLM 建议回复: {reply}")
|
||||
logger.info(f"LLM 建议回复: {reply}")
|
||||
if self.input_pos:
|
||||
logger.info(f"⚡ [监控] 执行自动回复...")
|
||||
# input_pos 是 ((x,y), box) 格式，取第一个元素坐标点
|
||||
target_pos = self.input_pos[0] if isinstance(self.input_pos, (list, tuple)) and len(self.input_pos) == 2 and isinstance(self.input_pos[0], (list, tuple)) else self.input_pos
|
||||
# 简单兼容处理：如果 input_pos[0] 是 tuple/list 且 input_pos[1] 是 None/box，则取 input_pos[0]
|
||||
if isinstance(self.input_pos, (list, tuple)) and len(self.input_pos) == 2 and isinstance(self.input_pos[0], (list, tuple)):
|
||||
target_pos = self.input_pos[0]
|
||||
|
||||
# [User Request] 语音转文字未关闭 -> 顺序不对
|
||||
# 在回复发送前，尝试关闭语音转文字（如果最后一条是语音）
|
||||
# 这样截图看起来更干净，且符合用户期望的顺序
|
||||
# 如果触发回复的消息是语音，先点击语音条以清理状态
|
||||
if last_msg.get('type') == 'voice':
|
||||
logger.info("🧹 [监控] 尝试关闭语音转文字显示...")
|
||||
logger.info("回复前点击语音消息中心以关闭转文字遮罩")
|
||||
try:
|
||||
# 点击语音消息中心，关闭文字
|
||||
cx, cy = last_msg['center']
|
||||
cx, cy = last_msg.get('center', (0, 0))
|
||||
WxUtil.safe_device_click(self.device, cx, cy)
|
||||
# 稍微等待界面刷新
|
||||
await asyncio.sleep(1)
|
||||
await asyncio.sleep(1.5)
|
||||
self.device.screenshot(self.screenshot_path)
|
||||
except Exception as e:
|
||||
logger.warning(f"关闭语音转文字失败: {e}")
|
||||
logger.warning(f"清理语音状态失败: {e}")
|
||||
|
||||
perform_input_action(self.device, target_pos, reply)
|
||||
# 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
|
||||
|
||||
# 发送后截图留存
|
||||
reply_sent_shot = WxUtil.get_next_debug_path("event_reply_sent")
|
||||
self.device.screenshot(reply_sent_shot)
|
||||
logger.info(f"✅ [监控] 回复已发送，保存发送后截图: {reply_sent_shot}")
|
||||
# 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("❌ 未找到输入框位置，无法发送回复")
|
||||
logger.error("回复动作执行失败")
|
||||
else:
|
||||
logger.info("⚪ [监控] LLM 认为无需回复")
|
||||
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
|
||||
else:
|
||||
# 是我发的消息，更新哈希，不再处理
|
||||
self.last_processed_msg_hash = current_msg_hash
|
||||
|
||||
await asyncio.sleep(self.check_interval)
|
||||
|
||||
@@ -194,8 +194,8 @@ def setup_script_environment():
|
||||
# 重置调试计数器
|
||||
global _debug_counter
|
||||
_debug_counter = 0
|
||||
# 排除当前正在使用的日志文件
|
||||
clear_directory(LOG_DIR, exclude_files=["T2_ChatMonitor.log", "WxUtil.log"])
|
||||
# 清理所有日志和图片，确保新一轮运行有干净的环境
|
||||
clear_directory(LOG_DIR)
|
||||
clear_directory(OUTPUT_DIR)
|
||||
|
||||
def connect_device():
|
||||
|
||||
Reference in New Issue
Block a user