feat;replyer采用多模态

src/chat/replyer/maisaka_generator_multi.py

@@ -1,7 +1,6 @@
 from dataclasses import dataclass, field
 from datetime import datetime
 from typing import Dict, List, Optional, Tuple
-
 import random
 import time
 
@@ -9,6 +8,7 @@ from sqlmodel import select
 
 from src.chat.message_receive.chat_manager import BotChatSession
 from src.common.database.database import get_db_session
+from src.common.data_models.message_component_data_model import MessageSequence, TextComponent
 from src.common.database.database_model import Expression
 from src.common.data_models.reply_generation_data_models import (
     GenerationMetrics,
@@ -19,12 +19,12 @@ from src.common.logger import get_logger
 from src.common.prompt_i18n import load_prompt
 from src.config.config import global_config
 from src.core.types import ActionInfo
+from src.llm_models.payload_content.message import ImageMessagePart, Message, MessageBuilder, RoleType, TextMessagePart
 from src.services.llm_service import LLMServiceClient
 
 from src.chat.message_receive.message import SessionMessage
-from src.llm_models.payload_content.message import Message, MessageBuilder, RoleType
 from src.maisaka.context_messages import AssistantMessage, LLMContextMessage, ReferenceMessage, SessionBackedMessage, ToolResultMessage
-from src.maisaka.message_adapter import parse_speaker_content
+from src.maisaka.message_adapter import clone_message_sequence, parse_speaker_content
 
 logger = get_logger("replyer")
 
@@ -159,6 +159,34 @@ class MaisakaReplyGenerator:
         """构建追加在上下文末尾的回复指令。"""
         return "请基于以上逐条对话消息，自然地继续回复。直接输出你要说的话，不要额外解释。"
 
+    def _build_multimodal_user_message(
+        self,
+        message: SessionBackedMessage,
+        default_user_name: str,
+    ) -> Optional[Message]:
+        """构建保留图片等多模态片段的用户消息。"""
+        speaker_name, _ = parse_speaker_content(message.processed_plain_text.strip())
+        visible_speaker = speaker_name or default_user_name
+
+        raw_message = clone_message_sequence(message.raw_message)
+        if not raw_message.components:
+            raw_message = MessageSequence([TextComponent(f"[{visible_speaker}]")])
+        elif isinstance(raw_message.components[0], TextComponent):
+            first_text = raw_message.components[0].text or ""
+            raw_message.components[0] = TextComponent(f"[{visible_speaker}]{first_text}")
+        else:
+            raw_message.components.insert(0, TextComponent(f"[{visible_speaker}]"))
+
+        multimodal_message = SessionBackedMessage(
+            raw_message=raw_message,
+            visible_text=f"[{visible_speaker}]{message.processed_plain_text}",
+            timestamp=message.timestamp,
+            message_id=message.message_id,
+            original_message=message.original_message,
+            source_kind=message.source_kind,
+        )
+        return multimodal_message.to_llm_message()
+
     def _build_history_messages(self, chat_history: List[LLMContextMessage]) -> List[Message]:
         """将 replyer 上下文拆成多条 LLM 消息。"""
         bot_nickname = global_config.bot.nickname.strip() or "Bot"
@@ -177,6 +205,11 @@ class MaisakaReplyGenerator:
                     )
                     continue
 
+                multimodal_message = self._build_multimodal_user_message(message, default_user_name)
+                if multimodal_message is not None:
+                    messages.append(multimodal_message)
+                    continue
+
                 for speaker_name, content_body in self._split_user_message_segments(message.processed_plain_text):
                     content = self._normalize_content(content_body)
                     if not content:
@@ -227,7 +260,14 @@ class MaisakaReplyGenerator:
         preview_lines: List[str] = []
         for message in messages:
             role_name = message.role.value.capitalize()
-            preview_lines.append(f"{role_name}: {message.get_text_content()}")
+            part_previews: List[str] = []
+            for part in message.parts:
+                if isinstance(part, TextMessagePart):
+                    part_previews.append(part.text)
+                    continue
+                if isinstance(part, ImageMessagePart):
+                    part_previews.append(f"[图片:{part.normalized_image_format}]")
+            preview_lines.append(f"{role_name}: {''.join(part_previews)}")
         return "\n\n".join(preview_lines)
 
     def _resolve_session_id(self, stream_id: Optional[str]) -> str:

src/chat/replyer/maisaka_replyer_factory.py

@@ -5,8 +5,8 @@ from src.config.config import global_config
 
 def get_maisaka_replyer_class() -> Type[object]:
     """根据配置返回 Maisaka replyer 类。"""
-    generator_type = global_config.maisaka.replyer_generator_type
+    generator_type = get_maisaka_replyer_generator_type()
-    if generator_type == "multi":
+    if generator_type == "multimodal":
         from .maisaka_generator_multi import MaisakaReplyGenerator
 
         return MaisakaReplyGenerator
@@ -18,4 +18,4 @@ def get_maisaka_replyer_class() -> Type[object]:
 
 def get_maisaka_replyer_generator_type() -> str:
     """返回当前配置的 Maisaka replyer 生成器类型。"""
-    return global_config.maisaka.replyer_generator_type
+    return global_config.chat.replyer_generator_type

src/config/official_configs.py

@@ -16,8 +16,6 @@ class ExampleConfig(ConfigBase):
     \"""This is an example field\"""
     - 注释前面增加_warp_标记可以实现配置文件中注释在配置项前面单独一行显示
 """
-
-
 class BotConfig(ConfigBase):
     """机器人配置类"""
 
@@ -283,7 +281,7 @@ class ChatConfig(ConfigBase):
         },
     )
 
-    direct_image_input: bool = Field(
+    multimodal_planner: bool = Field(
         default=True,
         json_schema_extra={
             "x-widget": "switch",
@@ -292,14 +290,14 @@ class ChatConfig(ConfigBase):
     )
     """是否直接输入图片"""
 
-    replyer_generator_type: Literal["legacy", "multi"] = Field(
+    replyer_generator_type: Literal["legacy", "multimodal"] = Field(
         default="legacy",
         json_schema_extra={
             "x-widget": "select",
             "x-icon": "git-branch",
         },
     )
-    """Maisaka replyer 生成器类型：legacy（旧版单 prompt）/ multi（多消息版）"""
+    """Maisaka replyer 生成器类型：legacy（旧版单 prompt）/ multimodal（多模态版，适合主循环直接展示图片）"""
 
     enable_talk_value_rules: bool = Field(
         default=True,

src/maisaka/builtin_tool/reply.py

@@ -1,5 +1,6 @@
 """reply 内置工具。"""
 
+import traceback
 from typing import Optional
 
 from src.chat.replyer.replyer_manager import replyer_manager
@@ -82,6 +83,7 @@ async def handle_tool(
         logger.exception(
             f"{tool_ctx.runtime.log_prefix} 获取回复生成器时发生异常: 目标消息编号={target_message_id}"
         )
+        logger.info(traceback.format_exc())
         return tool_ctx.build_failure_result(
             invocation.tool_name,
             "获取 Maisaka 回复生成器时发生异常。",

src/maisaka/reasoning_engine.py

@@ -266,7 +266,7 @@ class MaisakaReasoningEngine:
         source_sequence = message.raw_message
 
         planner_components = clone_message_sequence(source_sequence).components
-        if global_config.chat.direct_image_input:
+        if global_config.chat.multimodal_planner:
             await self._hydrate_visual_components(planner_components)
         if planner_components and isinstance(planner_components[0], TextComponent):
             planner_components[0].text = planner_prefix + planner_components[0].text

src/plugin_runtime/host/message_utils.py

@@ -235,8 +235,18 @@ class PluginMessageUtils:
             if isinstance(raw_forward_nodes, list):
                 for node in raw_forward_nodes:
                     if not isinstance(node, dict):
+                        logger.info(f"解析转发节点时跳过非字典节点: {node!r}")
                         continue
                     raw_content = node.get("content", [])
+                    logger.info(
+                        "开始解析转发节点: "
+                        f"message_id={node.get('message_id')!r} "
+                        f"user_id={node.get('user_id')!r} "
+                        f"user_nickname={node.get('user_nickname')!r} "
+                        f"user_cardname={node.get('user_cardname')!r} "
+                        f"raw_content_type={type(raw_content).__name__} "
+                        f"raw_content={raw_content!r}"
+                    )
                     node_components: List[StandardMessageComponents] = []
                     if isinstance(raw_content, list):
                         node_components = [
@@ -244,7 +254,17 @@ class PluginMessageUtils:
                             for content in raw_content
                             if isinstance(content, dict)
                         ]
+                    logger.info(
+                        "转发节点解析结果: "
+                        f"message_id={node.get('message_id')!r} "
+                        f"component_types={[component.__class__.__name__ for component in node_components]!r} "
+                        f"component_values={[getattr(component, 'text', None) for component in node_components]!r}"
+                    )
                     if not node_components:
+                        logger.warning(
+                            "转发节点内容为空，使用占位文本回退: "
+                            f"message_id={node.get('message_id')!r} raw_content={raw_content!r}"
+                        )
                         node_components = [TextComponent(text="[empty forward node]")]
                     forward_nodes.append(
                         ForwardComponent(