prompt: detailed instructions for extracting unsubscribe links from promotional emails

main.py

@@ -8,7 +8,7 @@ import time
 from concurrent.futures import ThreadPoolExecutor
 from src.config import load_config
 from src.summarizer import poll_accounts, ai_process, tg_send_and_mark
-from src.tg_bot import poll_telegram
+from src.tg_bot import poll_telegram, send_thinking
 
 # Thread-safe logging: all log records go through a single QueueListener thread
 _log_queue = queue_module.Queue(-1)
@@ -28,6 +28,7 @@ _log_listener.start()
 logger = logging.getLogger("main")
 
 _running = True
+_shutdown_event = threading.Event()
 _pending_lock = threading.Lock()
 _pending_uids: set[bytes] = set()
 _processing_uids: set[bytes] = set()  # 正在处理（含等待TG发送）的邮件UID
@@ -42,6 +43,7 @@ def _signal_handler(signum, frame):
     global _running
     logger.info("收到退出信号，正在停止...")
     _running = False
+    _shutdown_event.set()
 
 
 def _email_poller(cfg):
@@ -55,15 +57,17 @@ def _email_poller(cfg):
                     if mail.uid not in _processing_uids:
                         _processing_uids.add(mail.uid)
                         _pending_uids.add(mail.uid)
-                        _email_queue.put((acct_idx, mail))
+                        # 发送"获取邮件中"提示
+                        try:
+                            tg_msg_id = send_thinking(cfg.telegram, cfg.telegram.chat_id, "📥 获取邮件中...")
+                        except Exception:
+                            tg_msg_id = 0
+                        _email_queue.put((acct_idx, mail, tg_msg_id))
         except Exception as e:
             logger.error(f"邮件轮询线程异常: {e}", exc_info=True)
             time.sleep(5)
         if _running:
-            for _ in range(cfg.polling.interval_seconds):
-                if not _running:
-                    return
-                time.sleep(1)
+            _shutdown_event.wait(cfg.polling.interval_seconds)
 
 
 def _ai_processor(cfg):
@@ -74,10 +78,10 @@ def _ai_processor(cfg):
         while _running:
             while len(pending_futures) < max_workers:
                 try:
-                    acct_idx, mail = _email_queue.get_nowait()
+                    acct_idx, mail, tg_msg_id = _email_queue.get_nowait()
                 except queue_module.Empty:
                     break
-                future = executor.submit(ai_process, cfg, acct_idx, mail)
+                future = executor.submit(ai_process, cfg, acct_idx, mail, tg_msg_id)
                 pending_futures[future] = (acct_idx, mail)
 
             completed = [f for f in pending_futures if f.done()]
@@ -124,7 +128,7 @@ def _tg_worker(cfg):
                     pass
 
         if _running:
-            time.sleep(1)
+            _shutdown_event.wait(1)
 
 
 def main():
@@ -153,9 +157,10 @@ def main():
             if tick % 30 == 0:
                 logger.info("队列状态: email_queue=%d tg_queue=%d pending_uids=%d",
                             _email_queue.qsize(), _tg_queue.qsize(), len(_pending_uids))
-            time.sleep(1)
+            _shutdown_event.wait(1)
     except KeyboardInterrupt:
         _running = False
+        _shutdown_event.set()
 
     for t in threads:
         t.join(timeout=5)

src/ai_client.py

@@ -34,6 +34,7 @@ SYSTEM_PROMPT = """你是一个邮件摘要助手。请分析邮件内容并以
 - verification_code: 仅当 category 为 "notification" 且邮件包含验证码时填写，提取完整的验证码字符串（纯数字、字母数字混合等均可，如「123456」「abc-789」「701383」）。非验证码邮件留空字符串
 - app_name: 仅当 category 为 "notification" 且邮件包含验证码时填写，填写发送验证码的应用/平台名称（如「GitHub」「微信」「Kraken」）。非验证码邮件留空字符串
 - links: 仅提取以下两类链接：1）邮件明确要求用户点击才能完成某操作的链接（如确认邮箱、重置密码、查看账单、审批请求、查看通知详情等）；2）退订/取消订阅链接。不要提取推广性质的链接（如下载App、注册领优惠、查看更多商品等营销链接）、追踪像素、邮件签名中的社交链接。每条包含简短描述文字和完整 URL。最多 3 条。没有符合条件的链接时为空数组
+- 退订链接提取规则：推广邮件必须提取退订链接。退订链接通常在邮件最底部，包含以下关键词之一：unsubscribe、退订、取消订阅、opt out、manage preferences、email preferences。即使链接被包裹在 HTML 标签中，也要提取 href 属性里的完整 URL。如果邮件底部同时有多个退订相关链接，只提取最主要的那个
 - summary 要简洁，把需要知道的信息浓缩成一句话，不需要分条列出
 - 当 category 为 "promotion" 时，summary 直接写成一句话：「[发送方名称] 在推广 [推广的产品/服务/活动]」，例如「腾讯云在推广双十一云服务器折扣」。此时 can_reply 为 false，reply_suggestions 为空数组
 - can_reply 为 false 表示无需回复或不适合建议回复（此时忽略 reply_suggestions）

src/email_client.py

@@ -113,9 +113,8 @@ def fetch_unseen_emails(account: EmailAccount, skip_uids: set[bytes] | None = No
     conn = _login_and_prepare(account, select_inbox=False)
     if skip_uids is None:
         skip_uids = set()
-    logger.info("fetch_unseen_emails: %s, skip_uids大小=%d", account.username, len(skip_uids))
 
-    # 列出所有文件夹
+    # 列出所有文件夹，跳过 NoSelect 文件夹
     _, folder_data = conn.list()
     folders = []
     if folder_data:
@@ -123,6 +122,9 @@ def fetch_unseen_emails(account: EmailAccount, skip_uids: set[bytes] | None = No
             if item is None:
                 continue
             line = item.decode("utf-8", errors="replace")
+            # 跳过 \NoSelect 文件夹（父文件夹，不能直接 SELECT）
+            if "\\NoSelect" in line:
+                continue
             m = re.search(r'"([^"]*)"\s*$', line)
             if m:
                 folders.append(m.group(1))
@@ -136,7 +138,7 @@ def fetch_unseen_emails(account: EmailAccount, skip_uids: set[bytes] | None = No
         try:
             status, _ = conn.select(folder, readonly=True)
             if status != "OK":
-                logger.warning("无法选择文件夹: %s", folder)
+                logger.debug("无法选择文件夹: %s", folder)
                 continue
             _, data = conn.uid("SEARCH", None, "UNSEEN")
             uids = data[0].split() if data[0] else []
@@ -148,11 +150,7 @@ def fetch_unseen_emails(account: EmailAccount, skip_uids: set[bytes] | None = No
             skipped += len(uids) - len(new_uids)
             if not new_uids:
                 continue
-            if skipped > 0 or len(uids) > 0:
-                logger.info("  %s: %d 封未读 (%d 封跳过) skip_uids=%d",
-                            folder, len(new_uids), len(uids) - len(new_uids), len(skip_uids))
-                if skip_uids and len(skip_uids) <= 5:
-                    logger.info("    skip_uids 内容: %s", [u for u in skip_uids])
+            logger.info("  %s: %d 封未读 (%d 封跳过)", folder, len(new_uids), len(uids) - len(new_uids))
 
             for uid in new_uids:
                 try:
@@ -206,7 +204,7 @@ def mark_as_seen(account: EmailAccount, uids_with_folder: list[tuple[bytes, str]
         by_folder[folder].append(uid)
     for folder, uids in by_folder.items():
         try:
-            conn.select(folder, readonly=True)
+            conn.select(folder, readonly=False)
             for uid in uids:
                 conn.uid("STORE", uid, "+FLAGS", "\\Seen")
             logger.info("  %s: 标记 %d 封", folder, len(uids))

src/summarizer.py

@@ -21,17 +21,22 @@ def poll_accounts(cfg: Config, skip_uids: set[bytes] | None = None) -> Generator
             logger.error(f"轮询 {acct.username} 失败: {e}", exc_info=True)
 
 
-def ai_process(cfg: Config, acct_idx: int, mail: Email) -> Optional[dict]:
+def ai_process(cfg: Config, acct_idx: int, mail: Email, tg_msg_id: int = 0) -> Optional[dict]:
     logger.info(f"  正在摘要: {mail.subject}")
-    # 先发思考中消息
-    thinking_msg_id = send_thinking(cfg.telegram, cfg.telegram.chat_id)
-    logger.info(f"  思考中消息已发送: msg_id={thinking_msg_id}")
+    # 阶段1：显示等待处理
+    if tg_msg_id:
+        edit_message(cfg.telegram, cfg.telegram.chat_id, tg_msg_id, "⏳ 等待处理…")
+    else:
+        tg_msg_id = send_thinking(cfg.telegram, cfg.telegram.chat_id, "⏳ 等待处理…")
 
+    # 阶段2：AI 处理，更新为思考中
+    edit_message(cfg.telegram, cfg.telegram.chat_id, tg_msg_id, "🤖 AI思考中…")
     summary = summarize_email(cfg.ai, mail.recipient, mail.subject, mail.sender, mail.body, mail.account_email)
     summary["recipient"] = mail.recipient
     if "@" not in summary.get("sender", ""):
         summary["sender"] = mail.sender
     text = format_summary(summary, mail.account_email)
+    # 阶段3：编辑为最终结果
     return {
         "text": text,
         "data": summary,
@@ -43,7 +48,7 @@ def ai_process(cfg: Config, acct_idx: int, mail: Email) -> Optional[dict]:
         "account_email": mail.account_email,
         "uid": mail.uid,
         "folder": mail.folder,
-        "thinking_msg_id": thinking_msg_id,
+        "thinking_msg_id": tg_msg_id if tg_msg_id else 0,
     }
 
 

src/tg_bot.py

@@ -57,10 +57,10 @@ def send_text(tg_cfg: TelegramConfig, chat_id: str, text: str,
     return result["result"]["message_id"]
 
 
-def send_thinking(tg_cfg: TelegramConfig, chat_id: str) -> int:
+def send_thinking(tg_cfg: TelegramConfig, chat_id: str, text: str = "💭 思考中…") -> int:
     result = _tg_req(tg_cfg, "sendMessage", {
         "chat_id": chat_id,
-        "text": "💭 思考中...",
+        "text": text,
     })
     return result["result"]["message_id"]
 
@@ -345,10 +345,10 @@ def _handle_callback(tg_cfg: TelegramConfig, cfg: Config, cb: dict):
         if not conv or conv.get("summary_msg_id") != msg_id:
             return
         logger.info("  换一批 (已有 %d 条历史)", len(conv.get("all_suggestions", [])))
-        # 先显示思考中
+        # AI 处理中
         _tg_req(tg_cfg, "editMessageText", {
             "chat_id": chat_id, "message_id": msg_id,
-            "text": "💭 思考中...", "parse_mode": None,
+            "text": "🤖 AI思考中...", "parse_mode": None,
         })
         try:
             new_suggestions = generate_more_replies(
@@ -390,10 +390,9 @@ def _handle_callback(tg_cfg: TelegramConfig, cfg: Config, cb: dict):
             logger.info("  取消, 删除提示消息")
 
     elif action == CALLBACK_PROMO_DETAIL and ctx:
-        # 先显示思考中
         _tg_req(tg_cfg, "editMessageText", {
             "chat_id": chat_id, "message_id": msg_id,
-            "text": "💭 思考中...", "parse_mode": None,
+            "text": "🤖 AI思考中...", "parse_mode": None,
         })
         try:
             details = expand_promo_detail(