allow multi-part replies for xiaoniu qa responses

plugins/ai_auto_response/main.py

@@ -298,13 +298,15 @@ class AIAutoResponsePlugin(MessagePluginInterface):
             )
             return False, "empty_response"
 
-        response = self._finalize_reply(response, reply_mode)
+        reply_chunks = self._finalize_reply(response, reply_mode)
 
-        await bot.send_text_message(room_id, response, sender)
+        for chunk in reply_chunks:
+            await bot.send_text_message(room_id, chunk, sender)
         self.last_reply_at[room_id] = time.time()
         self.flow_manager.note_bot_reply(room_id)
         self.memory_store.note_bot_reply(room_id, sender, trigger.topic)
-        self._upsert_interaction_memory(room_id, sender, sender_name, content, response, trigger.trigger_type, trigger.topic)
+        final_response_text = "\n".join(reply_chunks)
+        self._upsert_interaction_memory(room_id, sender, sender_name, content, final_response_text, trigger.trigger_type, trigger.topic)
         self._log_event(
             "sent",
             room_id=room_id,
@@ -312,8 +314,9 @@ class AIAutoResponsePlugin(MessagePluginInterface):
             sender_name=sender_name,
             trigger_type=trigger.trigger_type,
             reply_mode=reply_mode,
-            response_preview=self._preview(response),
-            response_len=len(response),
+            response_preview=self._preview(final_response_text),
+            response_len=len(final_response_text),
+            chunk_count=len(reply_chunks),
         )
         return False, "replied"
 
@@ -487,31 +490,29 @@ class AIAutoResponsePlugin(MessagePluginInterface):
         response = re.sub(r"\n{3,}", "\n\n", response)
         return response[:500].strip()
 
-    def _finalize_reply(self, response: str, reply_mode: str) -> str:
+    def _finalize_reply(self, response: str, reply_mode: str) -> List[str]:
         text = (response or "").strip()
         if not text:
-            return ""
+            return []
         text = re.sub(r"\s+", " ", text)
         text = text.replace("\n", " ").strip()
 
         if reply_mode == "social_short":
-            text = self._take_first_sentence(text, 12)
+            return [self._take_first_sentence(text, 12).strip()]
         elif reply_mode == "qa_fast":
-            text = self._take_first_sentence(text, 28)
+            return self._split_reply_chunks(text, sentence_limit=2, char_limit=28, chunk_limit=2)
         elif reply_mode == "qa_with_context":
-            text = self._take_first_sentence(text, 36)
-        else:
-            text = self._take_first_sentence(text, 24)
-        return text.strip()
+            return self._split_reply_chunks(text, sentence_limit=2, char_limit=36, chunk_limit=2)
+        return [self._take_first_sentence(text, 24).strip()]
 
     @staticmethod
     def _build_length_rule(reply_mode: str) -> str:
         if reply_mode == "social_short":
             return "默认只回一句短话，最好控制在2到8个字，除非非常不自然。"
         if reply_mode == "qa_fast":
-            return "尽量只回1句话，总长度优先控制在28字内，先给结论，不要主动补解释。"
+            return "优先1句话；如果确实需要，可以拆成2条很短的话发出，总长度每条优先控制在28字内，先给结论，不要主动补解释。"
         if reply_mode == "qa_with_context":
-            return "优先控制在1句话，必要时最多2句，总长度优先控制在36字内，只给第一层答案。"
+            return "优先控制在1句话；必要时可以拆成2条短消息发出，每条优先控制在36字内，只给第一层答案。"
         return "尽量短，像群友临时接一句，不要长篇大论。"
 
     @staticmethod
@@ -523,6 +524,24 @@ class AIAutoResponsePlugin(MessagePluginInterface):
         clipped = first[:limit].rstrip("，,、；;：:")
         return clipped
 
+    @staticmethod
+    def _split_reply_chunks(text: str, sentence_limit: int, char_limit: int, chunk_limit: int) -> List[str]:
+        parts = [item.strip() for item in re.split(r"(?<=[。！？!?；;])", text) if item.strip()]
+        if not parts:
+            short = text.strip()
+            return [short[:char_limit].rstrip("，,、；;：:").strip()] if short else []
+
+        chunks: List[str] = []
+        for part in parts[:sentence_limit]:
+            current = part
+            if len(current) > char_limit:
+                current = current[:char_limit].rstrip("，,、；;：:")
+            if current:
+                chunks.append(current.strip())
+            if len(chunks) >= chunk_limit:
+                break
+        return chunks[:chunk_limit] or [text[:char_limit].strip()]
+
 
     def _sync_member_memory(self, room_id: str, sender: str, sender_name: str, member_context: Dict) -> None:
         if not member_context:
@@ -663,6 +682,7 @@ class AIAutoResponsePlugin(MessagePluginInterface):
                 f"[XIAONIU] SENT room={room} user={sender_name}/{sender} "
                 f"trigger={data.get('trigger_type', 'none')} "
                 f"mode={data.get('reply_mode', '')} "
+                f"chunks={data.get('chunk_count', 1)} "
                 f"len={data.get('response_len', 0)} "
                 f"reply={data.get('response_preview', '')}"
             ).strip()