hud-python 0.4.34__tar.gz → 0.4.35__tar.gz

{hud_python-0.4.34 → hud_python-0.4.35}/.gitignore

@@ -50,4 +50,6 @@ test/
 /checkpoints/
 /checkpoints_test/
 hud/rl/checkpoints/
-hud/rl/checkpoints_test/
+hud/rl/checkpoints_test/
+
+.ck/

{hud_python-0.4.34 → hud_python-0.4.35}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.34
+Version: 0.4.35
 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-0.4.34 → hud_python-0.4.35}/hud/agents/claude.py

@@ -6,7 +6,7 @@ import copy
 import logging
 from typing import TYPE_CHECKING, Any, ClassVar, cast
 
-from anthropic import AsyncAnthropic, BadRequestError
+from anthropic import Anthropic, AsyncAnthropic, BadRequestError
 from anthropic.types.beta import BetaContentBlockParam, BetaImageBlockParam, BetaTextBlockParam
 
 import hud
@@ -54,6 +54,7 @@ class ClaudeAgent(MCPAgent):
         model: str = "claude-sonnet-4-20250514",
         max_tokens: int = 4096,
         use_computer_beta: bool = True,
+        validate_api_key: bool = True,
         **kwargs: Any,
     ) -> None:
         """
@@ -75,6 +76,13 @@ class ClaudeAgent(MCPAgent):
                 raise ValueError("Anthropic API key not found. Set ANTHROPIC_API_KEY.")
             model_client = AsyncAnthropic(api_key=api_key)
 
+        # validate api key if requested
+        if validate_api_key:
+            try:
+                Anthropic(api_key=model_client.api_key).models.list()
+            except Exception as e:
+                raise ValueError(f"Anthropic API key is invalid: {e}") from e
+
         self.anthropic_client = model_client
         self.model = model
         self.max_tokens = max_tokens

{hud_python-0.4.34 → hud_python-0.4.35}/hud/agents/openai.py

@@ -6,7 +6,7 @@ import logging
 from typing import Any, ClassVar, Literal
 
 import mcp.types as types
-from openai import AsyncOpenAI
+from openai import AsyncOpenAI, OpenAI
 from openai.types.responses import (
     ResponseComputerToolCall,
     ResponseInputMessageContentListParam,
@@ -45,6 +45,7 @@ class OperatorAgent(MCPAgent):
         model_client: AsyncOpenAI | None = None,
         model: str = "computer-use-preview",
         environment: Literal["windows", "mac", "linux", "browser"] = "linux",
+        validate_api_key: bool = True,
         **kwargs: Any,
     ) -> None:
         """
@@ -76,6 +77,13 @@ class OperatorAgent(MCPAgent):
         self.pending_call_id: str | None = None
         self.pending_safety_checks: list[Any] = []
 
+        # validate api key if requested
+        if validate_api_key:
+            try:
+                OpenAI(api_key=self.openai_client.api_key).models.list()
+            except Exception as e:
+                raise ValueError(f"OpenAI API key is invalid: {e}") from e
+
         self.model_name = "openai-" + self.model
 
         # Append OpenAI-specific instructions to the base system prompt

{hud_python-0.4.34 → hud_python-0.4.35}/hud/cli/build.py

@@ -489,7 +489,7 @@ def build_environment(
         hud_console.warning("Could not retrieve image ID for lock file")
 
     # Remove temp image after we're done
-    subprocess.run(["docker", "rmi", temp_tag], capture_output=True)  # noqa: S603, S607
+    subprocess.run(["docker", "rmi", "-f", temp_tag], capture_output=True)  # noqa: S603, S607
 
     # Add to local registry
     if image_id:

{hud_python-0.4.34 → hud_python-0.4.35}/hud/cli/eval.py

@@ -295,7 +295,7 @@ async def run_full_dataset(
     agent_type: Literal["claude", "openai", "vllm"] = "claude",
     model: str | None = None,
     allowed_tools: list[str] | None = None,
-    max_concurrent: int = 50,
+    max_concurrent: int = 30,
     max_steps: int = 10,
     parallel: bool = False,
     max_workers: int | None = None,

{hud_python-0.4.34 → hud_python-0.4.35}/hud/datasets/runner.py

@@ -22,7 +22,7 @@ async def run_dataset(
     dataset: str | Dataset | list[dict[str, Any]],
     agent_class: type[MCPAgent],
     agent_config: dict[str, Any] | None = None,
-    max_concurrent: int = 50,
+    max_concurrent: int = 30,
     metadata: dict[str, Any] | None = None,
     max_steps: int = 10,
     split: str = "train",

{hud_python-0.4.34 → hud_python-0.4.35}/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__ == "0.4.34"
+    assert hud.__version__ == "0.4.35"

{hud_python-0.4.34 → hud_python-0.4.35}/hud/utils/tool_shorthand.py

@@ -10,7 +10,8 @@ def _is_call_like(obj: Any) -> bool:
         return True
     if len(obj) == 1:
         _, v = next(iter(obj.items()))
-        return isinstance(v, dict)
+        if isinstance(v, dict):
+            return "name" in v or (len(v) == 1 and isinstance(next(iter(v.values())), dict))
     return False
 
 
@@ -19,9 +20,9 @@ def _to_call_dict(obj: Any) -> Any:
 
     Rules:
     - If obj is a dict with {name, arguments}: return {name, arguments: recurse(arguments)}
-    - Else if obj is a single-key dict {k: v}: return {name: k, arguments: recurse(v)}
+    - Else if obj is a single-key dict {k: v} where v looks call-like: return {name: k, arguments: recurse(v)}
     - Else: return obj unchanged (leaf arguments/value)
-    """
+    """  # noqa: E501
     if isinstance(obj, dict):
         if "name" in obj and "arguments" in obj:
             args = obj.get("arguments")
@@ -31,8 +32,10 @@ def _to_call_dict(obj: Any) -> Any:
             return {"name": obj.get("name"), "arguments": args}
         if len(obj) == 1:
             k, v = next(iter(obj.items()))
-            if isinstance(v, dict):
+            # Only convert single-key dicts if the value looks like it could be a call
+            if isinstance(v, dict) and _is_call_like(v):
                 return {"name": k, "arguments": _to_call_dict(v)}
+            # Otherwise, leave it as-is (this is the innermost arguments dict)
             return obj
     return obj
 

{hud_python-0.4.34 → hud_python-0.4.35}/hud/version.py

@@ -4,4 +4,4 @@ Version information for the HUD SDK.
 
 from __future__ import annotations
 
-__version__ = "0.4.34"
+__version__ = "0.4.35"

{hud_python-0.4.34 → hud_python-0.4.35}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "hud-python"
-version = "0.4.34"
+version = "0.4.35"
 description = "SDK for the HUD platform."
 readme = "README.md"
 requires-python = ">=3.11, <3.13"