Merge pull request #208 from hud-evals/g/fix-computer-overlap

Drop already existing "computer" if the selected computer tool is renamed

Author: pimpale

hud/agents/base.py

@@ -290,6 +290,10 @@ async def run_task(self, task: Task, max_steps: int = 10) -> Trace:
                 self.console.progress_log(f"Setting up tool phase: {task.setup_tool}")
                 results = await self.call_tools(task.setup_tool)
                 if any(result.isError for result in results):
+                    for result in results:
+                        if result.isError:
+                            self.console.error_log(f"Error in setup tool: {result}")
+
                     return Trace(
                         reward=0.0,
                         done=True,

hud/agents/claude.py

@@ -267,7 +267,7 @@ def _convert_tools_for_claude(self) -> None:
             if selected_computer_tool:
                 break
 
-        def to_api_tool(tool: types.Tool) -> BetaToolUnionParam:
+        def to_api_tool(tool: types.Tool) -> BetaToolUnionParam | None:
             if tool.name == "str_replace_based_edit_tool":
                 return BetaToolTextEditor20250728Param(
                     type="text_editor_20250728",
@@ -278,14 +278,21 @@ def to_api_tool(tool: types.Tool) -> BetaToolUnionParam:
                     type="bash_20250124",
                     name="bash",
                 )
-            if selected_computer_tool and tool.name == selected_computer_tool.name:
-                return BetaToolComputerUse20250124Param(
-                    type="computer_20250124",
-                    name="computer",
-                    display_number=1,
-                    display_width_px=computer_settings.ANTHROPIC_COMPUTER_WIDTH,
-                    display_height_px=computer_settings.ANTHROPIC_COMPUTER_HEIGHT,
-                )
+            if selected_computer_tool is not None:
+                if tool.name == selected_computer_tool.name:
+                    return BetaToolComputerUse20250124Param(
+                        type="computer_20250124",
+                        name="computer",
+                        display_number=1,
+                        display_width_px=computer_settings.ANTHROPIC_COMPUTER_WIDTH,
+                        display_height_px=computer_settings.ANTHROPIC_COMPUTER_HEIGHT,
+                    )
+                elif tool.name == "computer":
+                    logger.warning(
+                        "Renamed tool %s to 'computer', dropping original 'computer' tool",
+                        selected_computer_tool.name,
+                    )
+                    return None
 
             if tool.description is None or tool.inputSchema is None:
                 raise ValueError(
@@ -295,7 +302,7 @@ def to_api_tool(tool: types.Tool) -> BetaToolUnionParam:
                     2. Using pydantic Field() annotations on function parameters for the schema
                     """)
                 )
-            """Convert a tool to the API format"""
+
             return BetaToolParam(
                 name=tool.name,
                 description=tool.description,
@@ -307,6 +314,8 @@ def to_api_tool(tool: types.Tool) -> BetaToolUnionParam:
         self.claude_tools = []
         for tool in available_tools:
             claude_tool = to_api_tool(tool)
+            if claude_tool is None:
+                continue
             if claude_tool["name"] == "computer":
                 self.has_computer_tool = True
             self.tool_mapping[claude_tool["name"]] = tool.name

pyproject.toml

@@ -40,7 +40,7 @@ dependencies = [
     "numpy>=1.24.0",
     "pillow>=11.1.0",
     # AI providers
-    "anthropic",
+    "anthropic>=0.70",
     "openai",
     "google-genai",
     "tornado>=6.5.2",