Losita/mai-bot
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat：提高工具调用成功率，移除冗余的描述中参数介绍，增加索引列表的描述，修改prompt，移除timing的wait打断

Browse Source
This commit is contained in:
SengokuCola
2026-04-10 00:45:32 +08:00
parent 0852c38e81
commit fee9341620
17 changed files with 828 additions and 450 deletions
Download Patch File Download Diff File Expand all files Collapse all files

100
src/maisaka/chat_loop_service.py
View File

@@ -408,46 +408,43 @@ class MaisakaChatLoopService:
if llm_message is not None:
messages.append(llm_message)
normalized_injected_messages: List[Message] = []
for injected_message in injected_user_messages or []:
normalized_message = str(injected_message or "").strip()
if not normalized_message:
continue
messages.append(
normalized_injected_messages.append(
MessageBuilder()
.set_role(RoleType.User)
.add_text_content(normalized_message)
.build()
)
if normalized_injected_messages:
insertion_index = self._resolve_injected_user_messages_insertion_index(messages)
messages[insertion_index:insertion_index] = normalized_injected_messages
return messages
@staticmethod
def _build_tool_names_log_text(tool_definitions: Sequence[ToolDefinitionInput]) -> str:
"""构造 planner 请求前的工具列表日志文本。
def _resolve_injected_user_messages_insertion_index(messages: Sequence[Message]) -> int:
"""计算 injected meta user messages 在请求中的插入位置。
Args:
tool_definitions: 本轮实际传给 planner 的工具定义列表。
Returns:
str: 适合直接写入日志的单行文本。
规则与 deferred attachment 更接近：
- 从尾部向前寻找最近的 stopping point；
- stopping point 为 assistant 消息或 tool 结果消息；
- 找到后插入到其后面；
- 若不存在 stopping point，则退回到 system 消息之后。
"""
tool_names: List[str] = []
for tool_definition in tool_definitions:
if not isinstance(tool_definition, dict):
continue
normalized_name = str(tool_definition.get("name") or "").strip()
if not normalized_name:
function_definition = tool_definition.get("function")
if isinstance(function_definition, dict):
normalized_name = str(function_definition.get("name") or "").strip()
if normalized_name:
tool_names.append(normalized_name)
for index in range(len(messages) - 1, -1, -1):
message = messages[index]
if message.role in {RoleType.Assistant, RoleType.Tool}:
return index + 1
if not tool_names:
return "[无工具]"
return "、".join(tool_names)
if messages and messages[0].role == RoleType.System:
return 1
return 0
async def chat_loop_step(
self,
@@ -517,11 +514,6 @@ class MaisakaChatLoopService:
if isinstance(raw_tool_definitions, list):
all_tools = [item for item in raw_tool_definitions if isinstance(item, dict)]
logger.info(
f"规划器工具列表(request_kind={request_kind}): "
f"共 {len(all_tools)} 个 -> {self._build_tool_names_log_text(all_tools)}"
)
prompt_section: RenderableType | None = None
if global_config.debug.show_maisaka_thinking:
image_display_mode: str = "path_link" if global_config.maisaka.show_image_path else "legacy"
@@ -536,13 +528,6 @@ class MaisakaChatLoopService:
tool_definitions=list(all_tools),
)
logger.info(
f"规划器请求开始(request_kind={request_kind}): "
f"已选上下文消息数={len(selected_history)} "
f"大模型消息数={len(built_messages)} "
f"工具数={len(all_tools)} "
f"启用打断={self._interrupt_flag is not None}"
)
generation_result = await self._llm_chat.generate_response_with_messages(
message_factory=message_factory,
options=LLMGenerationOptions(
@@ -609,7 +594,7 @@ class MaisakaChatLoopService:
*,
max_context_size: Optional[int] = None,
) -> tuple[List[LLMContextMessage], str]:
"""??????? LLM ???????"""
"""选择LLM上下文消息"""
effective_context_size = max(1, int(max_context_size or global_config.chat.max_context_size))
selected_indices: List[int] = []
@@ -627,7 +612,7 @@ class MaisakaChatLoopService:
break
if not selected_indices:
return [], f"???????? {effective_context_size} ? user/assistant??? 0 ??"
return [], f"没有选择到上下文消息，实际发送 {effective_context_size} 条 user/assistant 消息"
selected_indices.reverse()
selected_history = [chat_history[index] for index in selected_indices]
@@ -644,47 +629,6 @@ class MaisakaChatLoopService:
selection_reason,
)
@staticmethod
def _select_llm_context_messages(chat_history: List[LLMContextMessage]) -> tuple[List[LLMContextMessage], str]:
"""选择真正发送给 LLM 的上下文消息。
Args:
chat_history: 当前全部对话历史。
Returns:
tuple[List[LLMContextMessage], str]: `(已选上下文, 选择说明)`。
"""
max_context_size = max(1, int(global_config.chat.max_context_size))
selected_indices: List[int] = []
counted_message_count = 0
for index in range(len(chat_history) - 1, -1, -1):
message = chat_history[index]
if message.to_llm_message() is None:
continue
selected_indices.append(index)
if message.count_in_context:
counted_message_count += 1
if counted_message_count >= max_context_size:
break
if not selected_indices:
return [], f"上下文判定：最近 {max_context_size} 条 user/assistant（当前 0 条）"
selected_indices.reverse()
selected_history = [chat_history[index] for index in selected_indices]
selected_history, hidden_assistant_count = MaisakaChatLoopService._hide_early_assistant_messages(selected_history)
selected_history, _ = drop_orphan_tool_results(selected_history)
return (
selected_history,
(
f"上下文判定：最近 {max_context_size} 条 user/assistant；"
f"展示并发送窗口内消息 {len(selected_history)} 条"
),
)
@staticmethod
def _hide_early_assistant_messages(
selected_history: List[LLMContextMessage],