feat：maisak正确解析图片原始数据，reply信息

src/maisaka/builtin_tool/context.py

@@ -6,7 +6,6 @@ from base64 import b64decode
 from datetime import datetime
 from typing import TYPE_CHECKING, Any, Dict, List, Optional
 
-from src.chat.message_receive.message import SessionMessage
 from src.chat.utils.utils import process_llm_response
 from src.common.data_models.message_component_data_model import EmojiComponent, MessageSequence, TextComponent
 from src.config.config import global_config

src/maisaka/chat_loop_service.py

@@ -15,7 +15,6 @@ from rich.panel import Panel
 
 from src.cli.console import console
 from src.common.data_models.llm_service_data_models import LLMGenerationOptions
-from src.common.data_models.message_component_data_model import MessageSequence, TextComponent
 from src.common.logger import get_logger
 from src.common.prompt_i18n import load_prompt
 from src.common.utils.utils_session import SessionUtils
@@ -38,9 +37,7 @@ from src.plugin_runtime.host.hook_spec_registry import HookSpec, HookSpecRegistr
 from src.services.llm_service import LLMServiceClient
 
 from .builtin_tool import get_builtin_tools
-from .context_messages import AssistantMessage, LLMContextMessage, SessionBackedMessage, ToolResultMessage
-from .message_adapter import format_speaker_content
-from .planner_message_utils import build_session_backed_text_message
+from .context_messages import AssistantMessage, LLMContextMessage, ToolResultMessage
 from .prompt_cli_renderer import PromptCLIVisualizer
 
 
@@ -324,7 +321,7 @@ class MaisakaChatLoopService:
         if not prompt_lines:
             return ""
 
-        return f"在该聊天中的注意事项：\n" + "\n\n".join(prompt_lines) + "\n"
+        return "在该聊天中的注意事项：\n" + "\n\n".join(prompt_lines) + "\n"
 
     @staticmethod
     def _get_chat_prompt_for_chat(chat_id: str, is_group_chat: Optional[bool]) -> str:

src/maisaka/context_messages.py

@@ -11,7 +11,13 @@ import base64
 from PIL import Image as PILImage
 
 from src.chat.message_receive.message import SessionMessage
-from src.common.data_models.message_component_data_model import EmojiComponent, ImageComponent, MessageSequence, TextComponent
+from src.common.data_models.message_component_data_model import (
+    EmojiComponent,
+    ImageComponent,
+    MessageSequence,
+    ReplyComponent,
+    TextComponent,
+)
 from src.llm_models.payload_content.message import Message, MessageBuilder, RoleType
 from src.llm_models.payload_content.tool_option import ToolCall
 
@@ -27,11 +33,42 @@ def _guess_image_format(image_bytes: bytes) -> Optional[str]:
         return None
 
 
-def _build_binary_component_type_text(component: EmojiComponent | ImageComponent) -> str:
-    """为图片类消息组件构造显式的消息类型标记。"""
-    if isinstance(component, EmojiComponent):
-        return "[消息类型]表情包"
-    return "[消息类型]图片"
+def _append_emoji_component(builder: MessageBuilder, component: EmojiComponent) -> bool:
+    """将表情组件追加到 LLM 消息构建器。"""
+    image_format = _guess_image_format(component.binary_data)
+    if image_format and component.binary_data:
+        builder.add_text_content("[消息类型]表情包")
+        builder.add_image_content(image_format, base64.b64encode(component.binary_data).decode("utf-8"))
+        return True
+
+    if component.content:
+        builder.add_text_content(component.content)
+        return True
+    return False
+
+
+def _append_image_component(builder: MessageBuilder, component: ImageComponent) -> bool:
+    """将图片组件追加到 LLM 消息构建器。"""
+    image_format = _guess_image_format(component.binary_data)
+    if image_format and component.binary_data:
+        builder.add_text_content("[消息类型]图片")
+        builder.add_image_content(image_format, base64.b64encode(component.binary_data).decode("utf-8"))
+        return True
+
+    if component.content:
+        builder.add_text_content(component.content)
+        return True
+    return False
+
+
+def _append_reply_component(builder: MessageBuilder, component: ReplyComponent) -> bool:
+    """将回复组件追加到 LLM 消息构建器。"""
+    target_message_id = component.target_message_id.strip()
+    if not target_message_id:
+        return False
+
+    builder.add_text_content(f"[引用回复]({target_message_id})")
+    return True
 
 
 def _build_message_from_sequence(
@@ -57,17 +94,17 @@ def _build_message_from_sequence(
                 has_content = True
             continue
 
-        if isinstance(component, (EmojiComponent, ImageComponent)):
-            image_format = _guess_image_format(component.binary_data)
-            if image_format and component.binary_data:
-                builder.add_text_content(_build_binary_component_type_text(component))
-                builder.add_image_content(image_format, base64.b64encode(component.binary_data).decode("utf-8"))
-                has_content = True
-                continue
+        if isinstance(component, EmojiComponent):
+            has_content = _append_emoji_component(builder, component) or has_content
+            continue
 
-            if component.content:
-                builder.add_text_content(component.content)
-                has_content = True
+        if isinstance(component, ImageComponent):
+            has_content = _append_image_component(builder, component) or has_content
+            continue
+
+        if isinstance(component, ReplyComponent):
+            has_content = _append_reply_component(builder, component) or has_content
+            continue
 
     if not has_content and fallback_text:
         builder.add_text_content(fallback_text)

src/maisaka/message_adapter.py

@@ -5,7 +5,13 @@ from datetime import datetime
 from typing import Optional
 import re
 
-from src.common.data_models.message_component_data_model import EmojiComponent, ImageComponent, MessageSequence, TextComponent
+from src.common.data_models.message_component_data_model import (
+    EmojiComponent,
+    ImageComponent,
+    MessageSequence,
+    ReplyComponent,
+    TextComponent,
+)
 
 SPEAKER_PREFIX_PATTERN = re.compile(
     r"^(?:(?P<timestamp>\d{2}:\d{2}:\d{2}))?(?:\[msg_id:(?P<message_id>[^\]]+)\])?\[(?P<speaker>[^\]]+)\](?P<content>.*)$",
@@ -65,5 +71,11 @@ def build_visible_text_from_sequence(message_sequence: MessageSequence) -> str:
 
         if isinstance(component, ImageComponent):
             parts.append("[图片]")
+            continue
+
+        if isinstance(component, ReplyComponent):
+            target_message_id = component.target_message_id.strip()
+            if target_message_id:
+                parts.append(f"[引用回复]({target_message_id})")
 
     return "".join(parts)

src/maisaka/reasoning_engine.py

@@ -12,7 +12,7 @@ import traceback
 from src.chat.heart_flow.heartFC_utils import CycleDetail
 from src.chat.message_receive.message import SessionMessage
 from src.chat.utils.utils import process_llm_response
-from src.common.data_models.message_component_data_model import MessageSequence, TextComponent
+from src.common.data_models.message_component_data_model import EmojiComponent, ImageComponent, MessageSequence, TextComponent
 from src.common.logger import get_logger
 from src.config.config import global_config
 from src.core.tooling import ToolExecutionContext, ToolExecutionResult, ToolInvocation, ToolSpec
@@ -230,12 +230,10 @@ class MaisakaReasoningEngine:
         planner_prefix = build_planner_user_prefix_from_session_message(message)
 
         appended_component = False
-        if global_config.maisaka.direct_image_input:
-            source_sequence = getattr(message, "maisaka_original_raw_message", message.raw_message)
-        else:
-            source_sequence = message.raw_message
-
+        source_sequence = message.raw_message
         planner_components = clone_message_sequence(source_sequence).components
+        if global_config.maisaka.direct_image_input:
+            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
         else:
@@ -256,6 +254,24 @@ class MaisakaReasoningEngine:
 
         return message_sequence, legacy_visible_text
 
+    async def _hydrate_visual_components(self, planner_components: list[object]) -> None:
+        """在 Maisaka 真正需要图片或表情时，按需回填二进制数据。"""
+        load_tasks: list[asyncio.Task[None]] = []
+        for component in planner_components:
+            if isinstance(component, ImageComponent) and not component.binary_data:
+                load_tasks.append(asyncio.create_task(component.load_image_binary()))
+                continue
+            if isinstance(component, EmojiComponent) and not component.binary_data:
+                load_tasks.append(asyncio.create_task(component.load_emoji_binary()))
+
+        if not load_tasks:
+            return
+
+        results = await asyncio.gather(*load_tasks, return_exceptions=True)
+        for result in results:
+            if isinstance(result, Exception):
+                logger.warning(f"{self._runtime.log_prefix} 回填图片或表情二进制数据失败，Maisaka 将退化为文本占位: {result}")
+
     def _build_legacy_visible_text(self, message: SessionMessage, source_sequence: MessageSequence) -> str:
         user_info = message.message_info.user_info
         speaker_name = user_info.user_cardname or user_info.user_nickname or user_info.user_id