diff --git a/src/chat/emoji_system/emoji_manager.py b/src/chat/emoji_system/emoji_manager.py
index b1a359d3c..6a9437757 100644
--- a/src/chat/emoji_system/emoji_manager.py
+++ b/src/chat/emoji_system/emoji_manager.py
@@ -1107,28 +1107,22 @@ class EmojiManager:
                 if emoji_base64 is None:  # 再次检查读取
                     logger.error(f"[注册失败] 无法读取图片以生成描述: {filename}")
                     return False
-                task = asyncio.create_task(self.build_emoji_description(emoji_base64))
                 
-                def after_built_description(fut: asyncio.Future):
-                    if fut.cancelled():
-                        logger.error(f"[注册失败] 描述生成任务被取消: {filename}")
-                    elif fut.exception():
-                        logger.error(f"[注册失败] 描述生成任务出错 ({filename}): {fut.exception()}")
-                    else:
-                        description, emotions = fut.result()
+                # 等待描述生成完成
+                description, emotions = await self.build_emoji_description(emoji_base64)
 
-                        if not description:  # 检查描述是否成功生成或审核通过
-                            logger.warning(f"[注册失败] 未能生成有效描述或审核未通过: {filename}")
-                            # 删除未能生成描述的文件
-                            try:
-                                os.remove(file_full_path)
-                                logger.info(f"[清理] 删除描述生成失败的文件: {filename}")
-                            except Exception as e:
-                                logger.error(f"[错误] 删除描述生成失败文件时出错: {e!s}")
-                            return False
-                        new_emoji.description = description
-                        new_emoji.emotion = emotions
-                task.add_done_callback(after_built_description)
+                if not description:  # 检查描述是否成功生成或审核通过
+                    logger.warning(f"[注册失败] 未能生成有效描述或审核未通过: {filename}")
+                    # 删除未能生成描述的文件
+                    try:
+                        os.remove(file_full_path)
+                        logger.info(f"[清理] 删除描述生成失败的文件: {filename}")
+                    except Exception as e:
+                        logger.error(f"[错误] 删除描述生成失败文件时出错: {e!s}")
+                    return False
+                
+                new_emoji.description = description
+                new_emoji.emotion = emotions
             except Exception as build_desc_error:
                 logger.error(f"[注册失败] 生成描述/情感时出错 ({filename}): {build_desc_error}")
                 # 同样考虑删除文件
diff --git a/src/chat/message_receive/message_handler.py b/src/chat/message_receive/message_handler.py
index a7238c2fd..29d78e2b3 100644
--- a/src/chat/message_receive/message_handler.py
+++ b/src/chat/message_receive/message_handler.py
@@ -134,7 +134,7 @@ class MessageHandler:
             predicate=_is_adapter_response,
             handler=self._handle_adapter_response_route,
             name="adapter_response_handler",
-            message_type="adapter_response",
+            priority=100
         )
 
         # 注册 notice 消息处理器（处理通知消息，如戳一戳、禁言等）
@@ -152,7 +152,7 @@ class MessageHandler:
             predicate=_is_notice_message,
             handler=self._handle_notice_message,
             name="notice_message_handler",
-            message_type="notice",
+            priority=90
         )
 
         # 注册默认消息处理器（处理所有其他消息）
@@ -160,6 +160,7 @@ class MessageHandler:
             predicate=lambda _: True,  # 匹配所有消息
             handler=self._handle_normal_message,
             name="default_message_handler",
+            priority=50
         )
 
         logger.info("MessageHandler 已向 MessageRuntime 注册处理器和钩子")
@@ -314,7 +315,7 @@ class MessageHandler:
             # 触发 notice 事件（可供插件监听）
             await event_manager.trigger_event(
                 EventType.ON_NOTICE_RECEIVED,
-                permission_group="USER",
+                permission_group="SYSTEM",
                 message=message,
                 notice_type=notice_type,
                 chat_stream=chat,
diff --git a/src/chat/message_receive/uni_message_sender.py b/src/chat/message_receive/uni_message_sender.py
index a149f02ba..7f3e1076f 100644
--- a/src/chat/message_receive/uni_message_sender.py
+++ b/src/chat/message_receive/uni_message_sender.py
@@ -13,7 +13,7 @@ from mofox_wire import MessageEnvelope
 from src.chat.message_receive.message_processor import process_message_from_dict
 from src.chat.message_receive.storage import MessageStorage
 from src.chat.utils.utils import calculate_typing_time, truncate_message
-from src.common.data_models.database_data_model import DatabaseMessages
+from src.common.data_models.database_data_model import DatabaseMessages, DatabaseUserInfo
 from src.common.logger import get_logger
 from src.config.config import global_config
 
@@ -27,13 +27,13 @@ logger = get_logger("sender")
 
 async def send_envelope(
     envelope: MessageEnvelope,
-    chat_stream: "ChatStream" | None = None,
+    chat_stream: ChatStream | None = None,
     db_message: DatabaseMessages | None = None,
     show_log: bool = True,
 ) -> bool:
     """发送消息"""
     message_preview = truncate_message(
-        (db_message.processed_plain_text if db_message else str(envelope.get("message_segment", ""))),
+        (db_message.processed_plain_text or "" if db_message else str(envelope.get("message_segment", ""))),
         max_length=120,
     )
 
@@ -81,6 +81,7 @@ class HeartFCSender:
         show_log: bool = True,
         thinking_start_time: float = 0.0,
         display_message: str | None = None,
+        storage_user_info: "DatabaseUserInfo | None" = None,
     ) -> bool:
         if not chat_stream:
             logger.error("消息缺少 chat_stream，无法发送")
@@ -93,6 +94,13 @@ class HeartFCSender:
                 platform=chat_stream.platform,
             )
 
+            # 如果提供了用于存储的用户信息，则覆盖
+            if storage_message and storage_user_info:
+                db_message.user_info.user_id = storage_user_info.user_id
+                db_message.user_info.user_nickname = storage_user_info.user_nickname
+                db_message.user_info.user_cardname = storage_user_info.user_cardname
+                db_message.user_info.platform = storage_user_info.platform
+
             # 使用调用方指定的展示文本
             if display_message:
                 db_message.display_message = display_message
@@ -125,9 +133,13 @@ class HeartFCSender:
 
                 # 将发送的消息写入上下文历史
                 try:
-                    if chat_stream.context:
+                    if chat_stream and chat_stream.context and global_config.chat:
                         context = chat_stream.context
-                        max_context_size = getattr(global_config.chat, "max_context_size", 40)
+                        chat_config = global_config.chat
+                        if chat_config:
+                            max_context_size = getattr(chat_config, "max_context_size", 40)
+                        else:
+                            max_context_size = 40
 
                         if len(context.history_messages) >= max_context_size:
                             context.history_messages = context.history_messages[1:]
diff --git a/src/chat/utils/utils_image.py b/src/chat/utils/utils_image.py
index 8e649f197..f0ae224c8 100644
--- a/src/chat/utils/utils_image.py
+++ b/src/chat/utils/utils_image.py
@@ -54,6 +54,7 @@ class ImageManager:
             self._ensure_image_dir()
 
             self._initialized = True
+            assert model_config is not None
             self.vlm = LLMRequest(model_set=model_config.model_task_config.vlm, request_type="image")
 
             # try:
@@ -189,7 +190,7 @@ class ImageManager:
                 return "[表情包(描述生成失败)]"
 
             # 4. (可选) 如果启用了“偷表情包”，则将图片和完整描述存入待注册区
-            if global_config.emoji.steal_emoji:
+            if global_config and global_config.emoji and global_config.emoji.steal_emoji:
                 logger.debug(f"偷取表情包功能已开启，保存待注册表情包: {image_hash}")
                 try:
                     image_format = (Image.open(io.BytesIO(image_bytes)).format or "jpeg").lower()
@@ -226,6 +227,22 @@ class ImageManager:
             image_bytes = base64.b64decode(image_base64)
             image_hash = hashlib.md5(image_bytes).hexdigest()
 
+            # 1.5. 如果是GIF，先转换为JPG
+            try:
+                image_format_check = (Image.open(io.BytesIO(image_bytes)).format or "jpeg").lower()
+                if image_format_check == "gif":
+                    logger.info(f"检测到GIF图片 (Hash: {image_hash[:8]}...)，正在转换为JPG...")
+                    if transformed_b64 := self.transform_gif(image_base64):
+                        image_base64 = transformed_b64
+                        image_bytes = base64.b64decode(image_base64)
+                        logger.info("GIF转换成功，将使用转换后的图片进行描述")
+                    else:
+                        logger.error("GIF转换失败，无法生成描述")
+                        return "[图片(GIF转换失败)]"
+            except Exception as e:
+                logger.warning(f"图片格式检测失败: {e!s}，将按原格式处理")
+
+
             # 2. 优先查询 Images 表缓存
             async with get_db_session() as session:
                 result = await session.execute(select(Images).where(Images.emoji_hash == image_hash))
@@ -242,6 +259,8 @@ class ImageManager:
             # 4. 如果都未命中，则同步调用VLM生成新描述
             logger.info(f"[新图片识别] 无缓存 (Hash: {image_hash[:8]}...)，调用VLM生成描述")
             description = None
+            assert global_config is not None
+            assert global_config.custom_prompt is not None
             prompt = global_config.custom_prompt.image_prompt
             logger.info(f"[识图VLM调用] Prompt: {prompt}")
             for i in range(3):  # 重试3次
diff --git a/src/common/database/core/session.py b/src/common/database/core/session.py
index b033088f9..9cd701fc2 100644
--- a/src/common/database/core/session.py
+++ b/src/common/database/core/session.py
@@ -126,6 +126,12 @@ async def get_db_session_direct() -> AsyncGenerator[AsyncSession, None]:
     用于特殊场景，如需要完全独立的连接时。
     一般情况下应使用 get_db_session()。
 
+    事务管理说明：
+    - 正常退出时自动提交事务
+    - 发生异常时自动回滚事务
+    - 如果用户代码已手动调用 commit/rollback，再次调用是安全的
+    - 适用于所有数据库类型（SQLite, MySQL, PostgreSQL）
+
     Yields:
         AsyncSession: SQLAlchemy异步会话对象
     """
@@ -139,8 +145,16 @@ async def get_db_session_direct() -> AsyncGenerator[AsyncSession, None]:
             await _apply_session_settings(session, global_config.database.database_type)
 
             yield session
+
+            # 正常退出时提交事务
+            # 这对所有数据库都很重要，因为 SQLAlchemy 默认不是 autocommit 模式
+            # 检查事务是否活动，避免在已回滚的事务上提交
+            if session.is_active:
+                await session.commit()
         except Exception:
-            await session.rollback()
+            # 检查是否需要回滚（事务是否活动）
+            if session.is_active:
+                await session.rollback()
             raise
         finally:
             await session.close()
diff --git a/src/common/database/optimization/batch_scheduler.py b/src/common/database/optimization/batch_scheduler.py
index 821d0b1af..f2d2591fb 100644
--- a/src/common/database/optimization/batch_scheduler.py
+++ b/src/common/database/optimization/batch_scheduler.py
@@ -17,7 +17,7 @@ from typing import Any, TypeVar
 
 from sqlalchemy import delete, insert, select, update
 
-from src.common.database.core.session import get_db_session
+from src.common.database.core.session import get_db_session_direct
 from src.common.logger import get_logger
 from src.common.memory_utils import estimate_size_smart
 
@@ -330,7 +330,7 @@ class AdaptiveBatchScheduler:
         operations: list[BatchOperation],
     ) -> None:
         """批量执行查询操作"""
-        async with get_db_session() as session:
+        async with get_db_session_direct() as session:
             for op in operations:
                 try:
                     # 构建查询
@@ -371,7 +371,7 @@ class AdaptiveBatchScheduler:
         operations: list[BatchOperation],
     ) -> None:
         """批量执行插入操作"""
-        async with get_db_session() as session:
+        async with get_db_session_direct() as session:
             try:
                 # 收集数据，并过滤掉 id=None 的情况（让数据库自动生成）
                 all_data = []
@@ -387,7 +387,7 @@ class AdaptiveBatchScheduler:
                 # 批量插入
                 stmt = insert(operations[0].model_class).values(all_data)
                 await session.execute(stmt)
-                await session.commit()
+                # 注意：commit 由 get_db_session_direct 上下文管理器自动处理
 
                 # 设置结果
                 for op in operations:
@@ -402,20 +402,21 @@ class AdaptiveBatchScheduler:
 
             except Exception as e:
                 logger.error(f"批量插入失败: {e}")
-                await session.rollback()
+                # 注意：rollback 由 get_db_session_direct 上下文管理器自动处理
                 for op in operations:
                     if op.future and not op.future.done():
                         op.future.set_exception(e)
+                raise  # 重新抛出异常以触发 rollback
 
     async def _execute_update_batch(
         self,
         operations: list[BatchOperation],
     ) -> None:
         """批量执行更新操作"""
-        async with get_db_session() as session:
+        async with get_db_session_direct() as session:
             results = []
             try:
-                # 🔧 修复：收集所有操作后一次性commit，而不是循环中多次commit
+                # 🔧 收集所有操作后一次性commit，而不是循环中多次commit
                 for op in operations:
                     # 构建更新语句
                     stmt = update(op.model_class)
@@ -430,8 +431,7 @@ class AdaptiveBatchScheduler:
                     result = await session.execute(stmt)
                     results.append((op, result.rowcount))
 
-                # 所有操作成功后，一次性commit
-                await session.commit()
+                # 注意：commit 由 get_db_session_direct 上下文管理器自动处理
 
                 # 设置所有操作的结果
                 for op, rowcount in results:
@@ -446,21 +446,22 @@ class AdaptiveBatchScheduler:
 
             except Exception as e:
                 logger.error(f"批量更新失败: {e}")
-                await session.rollback()
+                # 注意：rollback 由 get_db_session_direct 上下文管理器自动处理
                 # 所有操作都失败
                 for op in operations:
                     if op.future and not op.future.done():
                         op.future.set_exception(e)
+                raise  # 重新抛出异常以触发 rollback
 
     async def _execute_delete_batch(
         self,
         operations: list[BatchOperation],
     ) -> None:
         """批量执行删除操作"""
-        async with get_db_session() as session:
+        async with get_db_session_direct() as session:
             results = []
             try:
-                # 🔧 修复：收集所有操作后一次性commit，而不是循环中多次commit
+                # 🔧 收集所有操作后一次性commit，而不是循环中多次commit
                 for op in operations:
                     # 构建删除语句
                     stmt = delete(op.model_class)
@@ -472,8 +473,7 @@ class AdaptiveBatchScheduler:
                     result = await session.execute(stmt)
                     results.append((op, result.rowcount))
 
-                # 所有操作成功后，一次性commit
-                await session.commit()
+                # 注意：commit 由 get_db_session_direct 上下文管理器自动处理
 
                 # 设置所有操作的结果
                 for op, rowcount in results:
@@ -488,11 +488,12 @@ class AdaptiveBatchScheduler:
 
             except Exception as e:
                 logger.error(f"批量删除失败: {e}")
-                await session.rollback()
+                # 注意：rollback 由 get_db_session_direct 上下文管理器自动处理
                 # 所有操作都失败
                 for op in operations:
                     if op.future and not op.future.done():
                         op.future.set_exception(e)
+                raise  # 重新抛出异常以触发 rollback
 
     async def _adjust_parameters(self) -> None:
         """根据性能自适应调整参数"""
diff --git a/src/common/database/optimization/connection_pool.py b/src/common/database/optimization/connection_pool.py
index 4030bc061..ed7d3e5ef 100644
--- a/src/common/database/optimization/connection_pool.py
+++ b/src/common/database/optimization/connection_pool.py
@@ -123,6 +123,12 @@ class ConnectionPoolManager:
         """
         获取数据库会话的透明包装器
         如果有可用连接则复用，否则创建新连接
+
+        事务管理说明：
+        - 正常退出时自动提交事务
+        - 发生异常时自动回滚事务
+        - 如果用户代码已手动调用 commit/rollback，再次调用是安全的（空操作）
+        - 支持所有数据库类型：SQLite、MySQL、PostgreSQL
         """
         connection_info = None
 
@@ -151,21 +157,30 @@ class ConnectionPoolManager:
 
             yield connection_info.session
 
-            # 🔧 修复：正常退出时提交事务
-            # 这对SQLite至关重要，因为SQLite没有autocommit
+            # 🔧 正常退出时提交事务
+            # 这对所有数据库（SQLite、MySQL、PostgreSQL）都很重要
+            # 因为 SQLAlchemy 默认使用事务模式，不会自动提交
+            # 注意：如果用户代码已调用 commit()，这里的 commit() 是安全的空操作
             if connection_info and connection_info.session:
                 try:
-                    await connection_info.session.commit()
+                    # 检查事务是否处于活动状态，避免在已回滚的事务上提交
+                    if connection_info.session.is_active:
+                        await connection_info.session.commit()
                 except Exception as commit_error:
                     logger.warning(f"提交事务时出错: {commit_error}")
-                    await connection_info.session.rollback()
+                    try:
+                        await connection_info.session.rollback()
+                    except Exception:
+                        pass  # 忽略回滚错误，因为事务可能已经结束
                     raise
 
         except Exception:
             # 发生错误时回滚连接
             if connection_info and connection_info.session:
                 try:
-                    await connection_info.session.rollback()
+                    # 检查是否需要回滚（事务是否活动）
+                    if connection_info.session.is_active:
+                        await connection_info.session.rollback()
                 except Exception as rollback_error:
                     logger.warning(f"回滚连接时出错: {rollback_error}")
             raise
diff --git a/src/plugin_system/apis/send_api.py b/src/plugin_system/apis/send_api.py
index 3aac3d76c..d5aecb070 100644
--- a/src/plugin_system/apis/send_api.py
+++ b/src/plugin_system/apis/send_api.py
@@ -192,6 +192,7 @@ def _build_message_envelope(
     timestamp: float,
 ) -> MessageEnvelope:
     """构建发送的 MessageEnvelope 数据结构"""
+    # 这里的 user_info 决定了消息要发给谁，所以在私聊场景下必须是目标用户
     target_user_info = target_stream.user_info or bot_user_info
     message_info: dict[str, Any] = {
         "message_id": message_id,
@@ -212,7 +213,7 @@ def _build_message_envelope(
             "platform": target_stream.group_info.platform,
         }
 
-    return {
+    return {  # type: ignore
         "id": str(uuid.uuid4()),
         "direction": "outgoing",
         "platform": target_stream.platform,
@@ -257,9 +258,14 @@ async def _send_to_target(
         current_time = time.time()
         message_id = f"send_api_{int(current_time * 1000)}"
 
+        bot_config = global_config.bot
+        if not bot_config:
+            logger.error("机器人配置丢失，无法构建机器人用户信息")
+            return False
+
         bot_user_info = DatabaseUserInfo(
-            user_id=str(global_config.bot.qq_account),
-            user_nickname=global_config.bot.nickname,
+            user_id=str(bot_config.qq_account),
+            user_nickname=bot_config.nickname,
             platform=target_stream.platform,
         )
 
@@ -328,6 +334,7 @@ async def _send_to_target(
             show_log=show_log,
             thinking_start_time=current_time,
             display_message=display_message_for_db,
+            storage_user_info=bot_user_info,
         )
 
         if sent_msg:
diff --git a/src/plugin_system/core/event_manager.py b/src/plugin_system/core/event_manager.py
index f9e4de553..93f63642b 100644
--- a/src/plugin_system/core/event_manager.py
+++ b/src/plugin_system/core/event_manager.py
@@ -433,6 +433,7 @@ class EventManager:
             EventType.AFTER_LLM,
             EventType.POST_SEND,
             EventType.AFTER_SEND,
+            EventType.ON_NOTICE_RECEIVED
         ]
 
         for event_name in default_events:
diff --git a/src/plugins/built_in/core_actions/emoji.py b/src/plugins/built_in/core_actions/emoji.py
index e3bbb79bf..e8ff4cc23 100644
--- a/src/plugins/built_in/core_actions/emoji.py
+++ b/src/plugins/built_in/core_actions/emoji.py
@@ -49,14 +49,6 @@ class EmojiAction(BaseAction):
     ----------------------------------------
     """
 
-    # ========== 以下使用旧的激活配置（已废弃但兼容） ==========
-    # 激活设置
-    if global_config.emoji.emoji_activate_type == "llm":
-        activation_type = ActionActivationType.LLM_JUDGE
-        random_activation_probability = 0
-    else:
-        activation_type = ActionActivationType.RANDOM
-        random_activation_probability = global_config.emoji.emoji_chance
     mode_enable = ChatMode.ALL
     parallel_action = True
 
@@ -88,6 +80,15 @@ class EmojiAction(BaseAction):
     # 关联类型
     associated_types: ClassVar[list[str]] = ["emoji"]
 
+    async def go_activate(self, chat_content: str = "", llm_judge_model=None) -> bool:
+        """根据配置选择激活方式"""
+        assert global_config is not None
+        if global_config.emoji.emoji_activate_type == "llm":
+            return await self._llm_judge_activation(
+                judge_prompt=self.llm_judge_prompt, llm_judge_model=llm_judge_model
+            )
+        return await self._random_activation(global_config.emoji.emoji_chance)
+
     async def execute(self) -> tuple[bool, str]:
         """执行表情动作"""
         logger.info(f"{self.log_prefix} 决定发送表情")
@@ -95,6 +96,7 @@ class EmojiAction(BaseAction):
         try:
             # 1. 获取发送表情的原因
             reason = self.action_data.get("reason", "表达当前情绪")
+            main_reply_content = self.action_data.get("main_reply_content", "")
             logger.info(f"{self.log_prefix} 发送表情原因: {reason}")
 
             # 2. 获取所有有效的表情包对象
@@ -108,7 +110,7 @@ class EmojiAction(BaseAction):
 
             # 3. 根据历史记录筛选表情
             try:
-                recent_emojis_desc = get_recent_emojis(self.chat_id, limit=10)
+                recent_emojis_desc = get_recent_emojis(self.chat_id, limit=20)
                 if recent_emojis_desc:
                     filtered_emojis = [emoji for emoji in all_emojis_obj if emoji.description not in recent_emojis_desc]
                     if filtered_emojis:
@@ -120,8 +122,8 @@ class EmojiAction(BaseAction):
                 logger.error(f"{self.log_prefix} 获取或处理表情发送历史时出错: {e}")
 
             # 4. 准备情感数据和后备列表
-            emotion_map: ClassVar = {}
-            all_emojis_data: ClassVar = []
+            emotion_map = {}
+            all_emojis_data = []
 
             for emoji in all_emojis_obj:
                 b64 = image_path_to_base64(emoji.full_path)
@@ -146,14 +148,15 @@ class EmojiAction(BaseAction):
             chosen_emotion = "表情包"  # 默认描述，避免变量未定义错误
 
             # 4. 根据配置选择不同的表情选择模式
+            assert global_config is not None
             if global_config.emoji.emoji_selection_mode == "emotion":
                 # --- 情感标签选择模式 ---
                 if not available_emotions:
                     logger.warning(f"{self.log_prefix} 获取到的表情包均无情感标签, 将随机发送")
                     emoji_base64, emoji_description = random.choice(all_emojis_data)
                 else:
-                    # 获取最近的5条消息内容用于判断
-                    recent_messages = await message_api.get_recent_messages(chat_id=self.chat_id, limit=5)
+                    # 获取最近的20条消息内容用于判断
+                    recent_messages = await message_api.get_recent_messages(chat_id=self.chat_id, limit=20)
                     messages_text = ""
                     if recent_messages:
                         messages_text = await message_api.build_readable_messages(
@@ -164,8 +167,15 @@ class EmojiAction(BaseAction):
                         )
 
                     # 构建prompt让LLM选择情感
+                    prompt_addition = ""
+                    if main_reply_content:
+                        prompt_addition = f"""
+                    这是你刚刚生成、准备发送的消息：
+                    "{main_reply_content}"
+                    """
                     prompt = f"""
-                    你是一个正在进行聊天的网友，你需要根据一个理由和最近的聊天记录，从一个情感标签列表中选择最匹配的一个。
+                    你是一个正在进行聊天的网友，你需要根据一个理由、最近的聊天记录以及你自己将要发送的消息，从一个情感标签列表中选择最匹配的一个。
+                    {prompt_addition}
                     这是最近的聊天记录：
                     {messages_text}
 
@@ -174,10 +184,8 @@ class EmojiAction(BaseAction):
                     请直接返回最匹配的那个情感标签，不要进行任何解释或添加其他多余的文字。
                     """
 
-                    if global_config.debug.show_prompt:
-                        logger.info(f"{self.log_prefix} 生成的LLM Prompt: {prompt}")
-                    else:
-                        logger.debug(f"{self.log_prefix} 生成的LLM Prompt: {prompt}")
+                    assert global_config is not None
+                    logger.debug(f"{self.log_prefix} 生成的LLM Prompt: {prompt}")
 
                     # 调用LLM
                     models = llm_api.get_available_models()
@@ -211,10 +219,11 @@ class EmojiAction(BaseAction):
                             )
                             emoji_base64, emoji_description = random.choice(all_emojis_data)
 
-            elif global_config.emoji.emoji_selection_mode == "description":
+            assert global_config is not None
+            if global_config.emoji.emoji_selection_mode == "description":
                 # --- 详细描述选择模式 ---
                 # 获取最近的5条消息内容用于判断
-                recent_messages = await message_api.get_recent_messages(chat_id=self.chat_id, limit=5)
+                recent_messages = await message_api.get_recent_messages(chat_id=self.chat_id, limit=20)
                 messages_text = ""
                 if recent_messages:
                     messages_text = await message_api.build_readable_messages(
@@ -234,8 +243,15 @@ class EmojiAction(BaseAction):
                 emoji_descriptions = [extract_refined_info(desc) for _, desc in all_emojis_data]
 
                 # 构建prompt让LLM选择描述
+                prompt_addition = ""
+                if main_reply_content:
+                    prompt_addition = f"""
+                这是你刚刚生成、准备发送的消息：
+                "{main_reply_content}"
+                """
                 prompt = f"""
-                你是一个正在进行聊天的网友，你需要根据一个理由和最近的聊天记录，从一个表情包描述列表中选择最匹配的一个。
+                你是一个正在进行聊天的网友，你需要根据一个理由、最近的聊天记录以及你自己将要发送的消息，从一个表情包描述列表中选择最匹配的一个。
+                {prompt_addition}
                 这是最近的聊天记录：
                 {messages_text}
 
@@ -264,44 +280,22 @@ class EmojiAction(BaseAction):
                     chosen_emotion = chosen_description  # 在描述模式下，用描述作为情感标签
                     logger.info(f"{self.log_prefix} LLM选择的描述: {chosen_description}")
 
-                    # 优化匹配逻辑：优先在精炼描述中精确匹配，然后进行关键词匹配
-                    def extract_refined_info(full_desc: str) -> str:
-                        return full_desc.split(" Desc:")[0].strip()
-
-                    # 1. 尝试在精炼描述中找到最匹配的表情
-                    # 我们假设LLM返回的是精炼描述的一部分或全部
+                    # 使用更鲁棒的子字符串匹配逻辑
                     matched_emoji = None
-                    best_match_score = 0
-
-                    for item in all_emojis_data:
-                        refined_info = extract_refined_info(item[1])
-                        # 计算一个简单的匹配分数
-                        score = 0
-                        if chosen_description.lower() in refined_info.lower():
-                            score += 2 # 包含匹配
-                        if refined_info.lower() in chosen_description.lower():
-                            score += 2 # 包含匹配
-
-                        # 关键词匹配加分
-                        chosen_keywords = re.findall(r"\w+", chosen_description.lower())
-                        item_keywords = re.findall(r"\[(.*?)\]", refined_info)
-                        if item_keywords:
-                            item_keywords_set = {k.strip().lower() for k in item_keywords[0].split(",")}
-                            for kw in chosen_keywords:
-                                if kw in item_keywords_set:
-                                    score += 1
-
-                        if score > best_match_score:
-                            best_match_score = score
-                            matched_emoji = item
+                    for b64, desc in all_emojis_data:
+                        # 检查LLM返回的描述是否是数据库中某个表情完整描述的一部分
+                        if chosen_description in desc:
+                            matched_emoji = (b64, desc)
+                            break
 
                     if matched_emoji:
                         emoji_base64, emoji_description = matched_emoji
-                        logger.info(f"{self.log_prefix} 找到匹配描述的表情包: {extract_refined_info(emoji_description)}")
+                        logger.info(f"{self.log_prefix} 找到匹配描述的表情包: {emoji_description}")
                     else:
                         logger.warning(f"{self.log_prefix} LLM选择的描述无法匹配任何表情包, 将随机选择")
                         emoji_base64, emoji_description = random.choice(all_emojis_data)
             else:
+                assert global_config is not None
                 logger.error(f"{self.log_prefix} 无效的表情选择模式: {global_config.emoji.emoji_selection_mode}")
                 return False, "无效的表情选择模式"