hud-python 0.3.4__tar.gz → 0.3.5__tar.gz

{hud_python-0.3.4 → hud_python-0.3.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.3.4
+Version: 0.3.5
 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

{hud_python-0.3.4 → hud_python-0.3.5}/examples/agents_tools/simple_task_example.py

@@ -18,7 +18,7 @@ Usage:
 import asyncio
 import logging
 from hud.mcp import ClaudeMCPAgent
-from hud.task import TaskConfig
+from hud.datasets import TaskConfig
 from mcp.types import CallToolRequestParams as MCPToolCall
 from mcp_use import MCPClient
 

{hud_python-0.3.4 → hud_python-0.3.5}/examples/environments/gmail_local.py

@@ -1,8 +1,12 @@
 import asyncio
 import hud
-from hud.mcp import ClaudeMCPAgent, OpenAIMCPAgent
-from hud.task import TaskConfig
-from mcp_use import MCPClient
+from hud.datasets import TaskConfig
+from hud.mcp import ClaudeMCPAgent, OpenAIMCPAgent, MCPClient
+from hud.mcp.base import AgentResult
+
+import logging
+
+logging.basicConfig(level=logging.INFO)
 
 
 async def main():
@@ -36,18 +40,24 @@ async def main():
             },
             "metadata": {"id": "forward-series-b-deck-to-billgates"},
         }
+        # Update the prompt to give a disclaimer that this is a testing environment (anthropic doesn't like doing some actions)
+        task_dict["prompt"] = (
+            task_dict["prompt"]
+            + "\n\nThis is a mock sandbox environment for evaluations, so feel free to execute any actions you want and do not ask me for confirmation before executing actions."
+        )
 
         # Create TaskConfig from dict
         task = TaskConfig(**task_dict)
 
         print("📡 Defining the environment...")
         print("🔴 See the agent live at http://localhost:6080/vnc.html")
-        client = MCPClient.from_dict({"mcp_config": task.mcp_config})
+        client = MCPClient(mcp_config=task.mcp_config)
 
         agent = ClaudeMCPAgent(  # or OpenAIMCPAgent
             mcp_client=client,
             model="claude-3-7-sonnet-20250219",
-            allowed_tools=["computer"],
+            # Allowing anthropic_computer tool to be used because we're using ClaudeMCPAgent
+            allowed_tools=["anthropic_computer"],
             initial_screenshot=True,
         )
 
@@ -57,16 +67,16 @@ async def main():
 
         # Run the task
         print("🚀 Running the task...")
-        eval_result = await agent.run(task, max_steps=10)
+        eval_result: AgentResult = await agent.run(task, max_steps=30)
         print(f"🎉 Task Result: {eval_result}")
 
         # Show formatted results
-        reward = eval_result.get("reward", 0.0)
+        reward = eval_result.reward
         print(f"   🏆 Reward: {reward}")
 
         # Clean up
         print("\n🧹 Cleaning up...")
-        await client.close_all_sessions()
+        await client.close()
         print("✅ Done!")
 
 

{hud_python-0.3.4 → hud_python-0.3.5}/examples/environments/gmail_remote.py

@@ -1,7 +1,7 @@
 import asyncio
 import hud
 from hud.mcp import ClaudeMCPAgent, OpenAIMCPAgent
-from hud.task import TaskConfig
+from hud.datasets import TaskConfig
 from mcp_use import MCPClient
 
 

{hud_python-0.3.4 → hud_python-0.3.5}/hud/tools/computer/anthropic.py

@@ -68,6 +68,8 @@ class AnthropicComputerTool(HudComputerTool):
         self,
         width: int = 1400,
         height: int = 850,
+        environment_width: int = 1920,
+        environment_height: int = 1080,
         display_num: int | None = None,
         platform_type: Literal["auto", "xdo", "pyautogui"] = "auto",
         rescale_images: bool = False,
@@ -79,6 +81,8 @@ class AnthropicComputerTool(HudComputerTool):
         Args:
             width: Target width for rescaling (default: 1400 for Anthropic)
             height: Target height for rescaling (default: 850 for Anthropic)
+            environment_width: Environment screen width (default: 1920)
+            environment_height: Environment screen height (default: 1080)
             display_num: X display number
             platform_type: Which executor to use:
                 - "auto": Automatically detect based on platform
@@ -91,6 +95,8 @@ class AnthropicComputerTool(HudComputerTool):
             width=width,
             height=height,
             display_num=display_num,
+            environment_width=environment_width,
+            environment_height=environment_height,
             platform_type=platform_type,
             rescale_images=rescale_images,
             **kwargs,

{hud_python-0.3.4 → hud_python-0.3.5}/hud/tools/computer/hud.py

@@ -16,9 +16,6 @@ from hud.tools.executors.xdo import XDOExecutor
 
 logger = logging.getLogger(__name__)
 
-BASE_SCREEN_WIDTH = 1920
-BASE_SCREEN_HEIGHT = 1080
-
 
 class HudComputerTool:
     """
@@ -29,6 +26,8 @@ class HudComputerTool:
         self,
         width: int | None = None,
         height: int | None = None,
+        environment_width: int = 1920,
+        environment_height: int = 1080,
         display_num: int | None = None,
         platform_type: Literal["auto", "xdo", "pyautogui"] = "auto",
         custom_executor: BaseExecutor | None = None,
@@ -38,8 +37,10 @@ class HudComputerTool:
         Initialize the HUD computer tool.
 
         Args:
-            width: Target width for rescaling (None = use actual screen width)
-            height: Target height for rescaling (None = use actual screen height)
+            width: Target width for rescaling (None = use environment width)
+            height: Target height for rescaling (None = use environment height)
+            environment_width: Base screen width
+            environment_height: Base screen height
             display_num: X display number
             platform_type: Which executor to use:
                 - "auto": Automatically detect based on platform
@@ -49,20 +50,25 @@ class HudComputerTool:
             rescale_images: If True, rescale screenshots. If False, only rescale action coordinates
         """
         # Use provided dimensions or defaults
-        self.width = width or BASE_SCREEN_WIDTH
-        self.height = height or BASE_SCREEN_HEIGHT
+        self.width = width or environment_width
+        self.environment_width = environment_width
+
+        self.height = height or environment_height
+        self.environment_height = environment_height
+
         self.rescale_images = rescale_images
 
         logger.info("Width: %s, Height: %s", self.width, self.height)
         logger.info(
-            "Base Screen Width: %s, Base Screen Height: %s",
-            BASE_SCREEN_WIDTH,
-            BASE_SCREEN_HEIGHT,
+            "Environment Screen Width: %s, Environment Screen Height: %s",
+            self.environment_width,
+            self.environment_height,
         )
 
         # Calculate scaling factors from base screen size to target size
-        self.scale_x = self.width / BASE_SCREEN_WIDTH
-        self.scale_y = self.height / BASE_SCREEN_HEIGHT
+        self.scale_x = self.width / self.environment_width
+
+        self.scale_y = self.height / self.environment_height
 
         logger.info("Scale X: %s, Scale Y: %s", self.scale_x, self.scale_y)
         self.scale = min(self.scale_x, self.scale_y)

{hud_python-0.3.4 → hud_python-0.3.5}/hud/tools/computer/openai.py

@@ -51,6 +51,8 @@ class OpenAIComputerTool(HudComputerTool):
         self,
         width: int = 1024,
         height: int = 768,
+        environment_width: int = 1920,
+        environment_height: int = 1080,
         display_num: int | None = None,
         platform_type: Literal["auto", "xdo", "pyautogui"] = "auto",
         rescale_images: bool = False,
@@ -62,6 +64,8 @@ class OpenAIComputerTool(HudComputerTool):
         Args:
             width: Target width for rescaling (default: 1024 for OpenAI)
             height: Target height for rescaling (default: 768 for OpenAI)
+            environment_width: Environment screen width (default: 1920)
+            environment_height: Environment screen height (default: 1080)
             display_num: X display number
             platform_type: Which executor to use:
                 - "auto": Automatically detect based on platform
@@ -73,6 +77,8 @@ class OpenAIComputerTool(HudComputerTool):
         super().__init__(
             width=width,
             height=height,
+            environment_width=environment_width,
+            environment_height=environment_height,
             display_num=display_num,
             platform_type=platform_type,
             rescale_images=rescale_images,

{hud_python-0.3.4 → hud_python-0.3.5}/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__ == "0.3.4"
+    assert hud.__version__ == "0.3.5"

{hud_python-0.3.4 → hud_python-0.3.5}/hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.3.4"
+__version__ = "0.3.5"

{hud_python-0.3.4 → hud_python-0.3.5}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "hud-python"
-version = "0.3.4"
+version = "0.3.5"
 description = "SDK for the HUD platform."
 readme = "README.md"
 requires-python = ">=3.11, <3.14"