feat: improve tool selection prompt with clearer instructions

app/agent/middleware/tool_selection.py

@@ -1,4 +1,5 @@
 """MoviePilot 自定义工具筛选中间件。"""
+
 import json
 from collections.abc import Awaitable, Callable
 from dataclasses import dataclass
@@ -61,7 +62,8 @@ def _create_tool_selection_response(tools: list[BaseTool]) -> TypeAdapter[Any]:
     # Create a Union of Annotated Literal types for each tool name with description
     # For instance: Union[Annotated[Literal["tool1"], Field(description="...")], ...]
     literals = [
-        Annotated[Literal[tool.name], Field(description=tool.description)] for tool in tools  # noqa
+        Annotated[Literal[tool.name], Field(description=tool.description)]
+        for tool in tools  # noqa
     ]
     selected_tool_type = Union[tuple(literals)]  # type: ignore[valid-type]  # noqa: UP007
 
@@ -90,9 +92,7 @@ def _render_tool_list(tools: list[BaseTool]) -> str:
 class ToolSelectionState(AgentState):
     """工具筛选中间件私有状态。"""
 
-    selected_tool_names: NotRequired[
+    selected_tool_names: NotRequired[Annotated[list[str] | None, PrivateStateAttr]]
-        Annotated[list[str] | None, PrivateStateAttr]
-    ]
     """当前这条用户请求首轮筛选得到的工具名列表。"""
 
 
@@ -102,7 +102,9 @@ class ToolSelectionStateUpdate(TypedDict):
     selected_tool_names: list[str] | None
 
 
-class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, ResponseT]):
+class ToolSelectorMiddleware(
+    AgentMiddleware[AgentState[ResponseT], ContextT, ResponseT]
+):
     """
     为 DeepSeek 兼容端点提供更稳妥的工具筛选实现。
 
@@ -125,12 +127,14 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
 
     state_schema = ToolSelectionState
 
-    def __init__(self,
+    def __init__(
-                 model: BaseChatModel,
+            self,
-                 system_prompt: str = DEFAULT_SYSTEM_PROMPT,
+            model: BaseChatModel,
-                 selection_tools: list[Any] | None = None,
+            system_prompt: str = DEFAULT_SYSTEM_PROMPT,
-                 max_tools: int | None = None,
+            selection_tools: list[Any] | None = None,
-                 always_include: list[str] | None = None, ) -> None:
+            max_tools: int | None = None,
+            always_include: list[str] | None = None,
+    ) -> None:
         super().__init__()
         self.model = model
         self.system_prompt = system_prompt
@@ -175,7 +179,9 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
                 raise ValueError(msg)
 
         # Separate tools that are always included from those available for selection
-        available_tools = [tool for tool in base_tools if tool.name not in self.always_include]
+        available_tools = [
+            tool for tool in base_tools if tool.name not in self.always_include
+        ]
 
         # If no tools available for selection, return None
         if not available_tools:
@@ -239,9 +245,15 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
             raise ValueError(msg)
 
         # Filter tools based on selection and append always-included tools
-        selected_tools: list[BaseTool] = [
+        if selected_tool_names:
-            tool for tool in available_tools if tool.name in selected_tool_names
+            selected_tools: list[BaseTool] = [
-        ]
+                tool for tool in available_tools if tool.name in selected_tool_names
+            ]
+        else:
+            # 如果模型筛选结果为空，则不对工具进行裁剪，使用所有可用工具
+            logger.warning("工具筛选结果为空，将恢复使用所有工具。")
+            selected_tools = available_tools
+
         always_included_tools: list[BaseTool] = [
             tool
             for tool in request.tools
@@ -264,12 +276,16 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
         和 Base URL，避免只靠单一条件漏判。
         """
         module_name = type(model).__module__.lower()
-        model_name = str(
+        model_name = (
-            getattr(model, "model_name", "") or getattr(model, "model", "")
+            str(getattr(model, "model_name", "") or getattr(model, "model", ""))
-        ).strip().lower()
+            .strip()
-        base_url = str(
+            .lower()
-            getattr(model, "openai_api_base", "") or getattr(model, "api_base", "")
+        )
-        ).strip().lower()
+        base_url = (
+            str(getattr(model, "openai_api_base", "") or getattr(model, "api_base", ""))
+            .strip()
+            .lower()
+        )
 
         return (
                 "deepseek" in module_name
@@ -353,6 +369,10 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
         DeepSeek 官方文档要求在 JSON 输出模式下，提示词中必须明确包含 JSON
         约束，否则兼容端点可能返回空内容或无意义输出。
         """
+        limit_instruction = ""
+        if self.max_tools:
+            limit_instruction = f"- Select up to {self.max_tools} tools. IF NO TOOLS ARE RELEVANT, DO NOT RETURN AN EMPTY ARRAY. SELECT THE MOST APPLICABLE ONES TO ENSURE THE REQUEST IS HANDLED."
+
         return (
             f"{selection_request.system_message}\n\n"
             "Return the answer in JSON only.\n"
@@ -361,8 +381,10 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
             "- The `tools` field must be a JSON array of strings.\n"
             "- Only use tool names from the allowed list below.\n"
             "- Order tools by relevance, with the most relevant first.\n"
+            f"{limit_instruction}\n"
             "- Do not add explanations, markdown, or extra keys.\n\n"
-            f"Allowed tools:\n{self._render_tool_list(selection_request.available_tools)}"
+            "Allowed tools:\n"
+            f"{self._render_tool_list(selection_request.available_tools)}"
         )
 
     def _normalize_selection_response(self, response: Any) -> dict[str, list[str]]:
@@ -371,13 +393,17 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
         """
         content = getattr(response, "content", response)
         text = self._extract_text_content(content)
+        logger.debug(f"工具筛选原始响应: {text}")
         payload = self._parse_json_object(text)
 
         tools = payload.get("tools")
         if not isinstance(tools, list):
             raise ValueError(f"工具筛选 JSON 缺少 `tools` 数组: {payload}")
 
-        normalized_tools = [tool_name for tool_name in tools if isinstance(tool_name, str)]
+        normalized_tools = [
+            tool_name for tool_name in tools if isinstance(tool_name, str)
+        ]
+        logger.debug(f"工具筛选标准化结果: {normalized_tools}")
         return {"tools": normalized_tools}
 
     async def _aselect_tools_with_deepseek(
@@ -394,9 +420,7 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
             [
                 {
                     "role": "system",
-                    "content": self._build_deepseek_selection_prompt(
+                    "content": self._build_deepseek_selection_prompt(selection_request),
-                        selection_request
-                    ),
                 },
                 selection_request.last_user_message,
             ]
@@ -406,9 +430,7 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
     @staticmethod
     def _extract_selected_tool_names(request: ModelRequest) -> list[str]:
         """从已筛选后的请求中提取最终工具名，保留原有顺序。"""
-        return [
+        return [tool.name for tool in request.tools if not isinstance(tool, dict)]
-            tool.name for tool in request.tools if not isinstance(tool, dict)
-        ]
 
     @staticmethod
     def _apply_selected_tools(
@@ -425,9 +447,7 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
             return request
 
         current_tools_by_name = {
-            tool.name: tool
+            tool.name: tool for tool in request.tools if not isinstance(tool, dict)
-            for tool in request.tools
-            if not isinstance(tool, dict)
         }
         selected_tools = [
             current_tools_by_name[tool_name]
@@ -498,9 +518,7 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
         )
         modified_request = await self._aselect_request_once(selection_request)
         selected_tool_names = self._extract_selected_tool_names(modified_request)
-        return ToolSelectionStateUpdate(
+        return ToolSelectionStateUpdate(selected_tool_names=selected_tool_names or None)
-            selected_tool_names=selected_tool_names or None
-        )
 
     async def awrap_model_call(
             self,
@@ -516,7 +534,11 @@ class ToolSelectorMiddleware(AgentMiddleware[AgentState[ResponseT], ContextT, Re
 
         # 正常路径下，`abefore_agent()` 已经提前写入状态；这里只保留一层兜底，
         # 兼容直接单测或未来某些绕过 before_agent 的调用场景。
-        if selected_tool_names is None and self.selection_tools and self.model is not None:
+        if (
+                selected_tool_names is None
+                and self.selection_tools
+                and self.model is not None
+        ):
             request = await self._aselect_request_once(request)
             selected_tool_names = self._extract_selected_tool_names(request) or None
             request.state["selected_tool_names"] = selected_tool_names  # noqa