hud-python 0.4.25__tar.gz → 0.4.27__tar.gz

{hud_python-0.4.25 → hud_python-0.4.27}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.25
+Version: 0.4.27
 Summary: SDK for the HUD platform.
 Project-URL: Homepage, https://github.com/hud-evals/hud-python
 Project-URL: Bug Tracker, https://github.com/hud-evals/hud-python/issues
@@ -41,7 +41,7 @@ Requires-Dist: hud-mcp-python-sdk>=3.13.2
 Requires-Dist: hud-mcp-use-python-sdk>=2.3.16
 Requires-Dist: opentelemetry-api>=1.34.1
 Requires-Dist: opentelemetry-exporter-otlp-proto-http>=1.34.1
-Requires-Dist: opentelemetry-instrumentation-mcp>=0.44.1
+Requires-Dist: opentelemetry-instrumentation-mcp==0.47.0
 Requires-Dist: opentelemetry-sdk>=1.34.1
 Requires-Dist: pathspec>=0.12.1
 Requires-Dist: prompt-toolkit==3.0.51

{hud_python-0.4.25 → hud_python-0.4.27}/hud/agents/base.py

@@ -556,26 +556,10 @@ class MCPAgent(ABC):
 
         all_tools = await self.mcp_client.list_tools()
 
-        # Filter tools
-        self._available_tools = []
-        self._tool_map = {}
-
-        # Track response tools by server
         response_tools_by_server: dict[str, str] = {}  # server_name -> tool_name
-
         for tool in all_tools:
-            # Check if tool should be included
-            if self.allowed_tools and tool.name not in self.allowed_tools:
-                continue
-            if tool.name in self.disallowed_tools:
-                continue
-
-            self._available_tools.append(tool)
-            # Simplified mapping - just tool name to tool
-            self._tool_map[tool.name] = tool
-
-            # Track response tools
             if "response" in tool.name or tool.name == "response":
+                self.console.debug(f"Found response tool: '{tool.name}'")
                 # Extract server name from tool name (e.g., "grader_response" -> "grader")
                 if "_" in tool.name:
                     server_name = tool.name.split("_", 1)[0]
@@ -583,27 +567,67 @@ class MCPAgent(ABC):
                 else:
                     response_tools_by_server["_default"] = tool.name
 
-        # Find the response tool to use (prioritize last server in config)
+        # Add response tool to lifecycle tools BEFORE filtering
         if response_tools_by_server and hasattr(self.mcp_client, "mcp_config"):
             # Get server names in order from mcp_config
             server_names = list(self.mcp_client.mcp_config.keys())
+            self.console.debug(f"Server names: {server_names}")
 
             # Try to find response tool from last server first
             response_tool_name = None
             for server_name in reversed(server_names):
                 if server_name in response_tools_by_server:
                     response_tool_name = response_tools_by_server[server_name]
+                    self.console.debug(
+                        f"Found response tool '{response_tool_name}' from server '{server_name}'"
+                    )
                     break
 
             # Fallback to any response tool
             if not response_tool_name and response_tools_by_server:
                 response_tool_name = next(iter(response_tools_by_server.values()))
+                self.console.debug(f"Using fallback response tool '{response_tool_name}'")
 
             # Add to lifecycle tools if found
             if response_tool_name and response_tool_name not in self.lifecycle_tools:
                 self.console.debug(f"Auto-detected '{response_tool_name}' tool as a lifecycle tool")
                 self.response_tool_name = response_tool_name
                 self.lifecycle_tools.append(response_tool_name)
+            elif response_tool_name:
+                self.console.debug(
+                    f"Response tool '{response_tool_name}' already in lifecycle_tools"
+                )
+                self.response_tool_name = response_tool_name
+        else:
+            self.console.debug("No response tools found or no mcp_config")
+
+        # Filter tools
+        self._available_tools = []
+        self._tool_map = {}
+
+        self.console.debug(f"All tools: {[t.name for t in all_tools]}")
+        self.console.debug(f"Allowed tools: {self.allowed_tools}")
+        self.console.debug(f"Disallowed tools: {self.disallowed_tools}")
+        self.console.debug(f"Lifecycle tools: {self.lifecycle_tools}")
+
+        for tool in all_tools:
+            # Lifecycle tools (setup, evaluate, response) should always be included
+            is_lifecycle = tool.name in self.lifecycle_tools
+
+            # Check if tool should be included
+            if not is_lifecycle:
+                if self.allowed_tools and tool.name not in self.allowed_tools:
+                    self.console.debug(f"Skipping tool '{tool.name}' - not in allowed_tools")
+                    continue
+                if tool.name in self.disallowed_tools:
+                    self.console.debug(f"Skipping tool '{tool.name}' - in disallowed_tools")
+                    continue
+
+            self.console.debug(
+                f"Adding tool '{tool.name}' to available tools (lifecycle={is_lifecycle})"
+            )
+            self._available_tools.append(tool)
+            self._tool_map[tool.name] = tool
 
         # Check if all required tools are available
         if self.required_tools:

{hud_python-0.4.25 → hud_python-0.4.27}/hud/agents/misc/response_agent.py

@@ -34,7 +34,8 @@ class ResponseAgent:
         
         - STOP: If the agent indicates it has successfully completed a task, even if phrased as a question
           like "I have entered the right values into this form. Would you like me to do anything else?"
-          or "Here is the website. Is there any other information you need?"
+          or "Here is the website. Is there any other information you need?" or if the agent has
+          strongly determined it wants to stop the task.
         
         - CONTINUE: If the agent is asking for clarification before proceeding with a task
           like "I'm about to clear cookies from this website. Would you like me to proceed?"

{hud_python-0.4.25 → hud_python-0.4.27}/hud/utils/tests/test_version.py

@@ -5,4 +5,4 @@ def test_import():
     """Test that the package can be imported."""
     import hud
 
-    assert hud.__version__ == "0.4.25"
+    assert hud.__version__ == "0.4.27"

{hud_python-0.4.25 → hud_python-0.4.27}/hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.4.25"
+__version__ = "0.4.27"

{hud_python-0.4.25 → hud_python-0.4.27}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "hud-python"
-version = "0.4.25"
+version = "0.4.27"
 description = "SDK for the HUD platform."
 readme = "README.md"
 requires-python = ">=3.11, <3.14"
@@ -27,7 +27,7 @@ dependencies = [
     "questionary==2.1.0",
     "prompt-toolkit==3.0.51",
     # Telemetry
-    "opentelemetry-instrumentation-mcp>=0.44.1",
+    "opentelemetry-instrumentation-mcp==0.47.0",
     "opentelemetry-api>=1.34.1",
     "opentelemetry-sdk>=1.34.1",
     "opentelemetry-exporter-otlp-proto-http>=1.34.1",