hud-python 0.4.35__tar.gz → 0.4.36__tar.gz

{hud_python-0.4.35 → hud_python-0.4.36}/.gitignore

@@ -22,7 +22,6 @@ uv.lock
 
 # Test files
 /*.ipynb
-test.json
 TODO.md
 
 .coverage

{hud_python-0.4.35 → hud_python-0.4.36}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hud-python
-Version: 0.4.35
+Version: 0.4.36
 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
@@ -40,7 +40,7 @@ Requires-Dist: datasets>=2.14.0
 Requires-Dist: httpx<1,>=0.23.0
 Requires-Dist: hud-fastmcp-python-sdk>=0.1.2
 Requires-Dist: hud-mcp-python-sdk>=3.13.2
-Requires-Dist: hud-mcp-use-python-sdk>=2.3.16
+Requires-Dist: hud-mcp-use-python-sdk==2.3.19
 Requires-Dist: numpy>=1.24.0
 Requires-Dist: openai
 Requires-Dist: opentelemetry-api>=1.34.1
@@ -50,8 +50,8 @@ Requires-Dist: opentelemetry-sdk>=1.34.1
 Requires-Dist: pathspec>=0.12.1
 Requires-Dist: pillow>=11.1.0
 Requires-Dist: prompt-toolkit==3.0.51
-Requires-Dist: pydantic-settings<3,>=2
-Requires-Dist: pydantic<3,>=2
+Requires-Dist: pydantic-settings<3,>=2.2
+Requires-Dist: pydantic<3,>=2.6
 Requires-Dist: questionary==2.1.0
 Requires-Dist: rich>=13.0.0
 Requires-Dist: toml>=0.10.2
@@ -59,7 +59,9 @@ Requires-Dist: typer>=0.9.0
 Requires-Dist: watchfiles>=0.21.0
 Requires-Dist: wrapt>=1.14.0
 Provides-Extra: agent
+Requires-Dist: aiodocker>=0.24.0; extra == 'agent'
 Requires-Dist: dotenv>=0.9.9; extra == 'agent'
+Requires-Dist: inspect-ai>=0.3.80; extra == 'agent'
 Requires-Dist: ipykernel; extra == 'agent'
 Requires-Dist: ipython<9; extra == 'agent'
 Requires-Dist: jupyter-client; extra == 'agent'
@@ -67,8 +69,21 @@ Requires-Dist: jupyter-core; extra == 'agent'
 Requires-Dist: langchain; extra == 'agent'
 Requires-Dist: langchain-anthropic; extra == 'agent'
 Requires-Dist: langchain-openai; extra == 'agent'
+Requires-Dist: pillow>=11.1.0; extra == 'agent'
+Requires-Dist: playwright; extra == 'agent'
+Requires-Dist: pyautogui>=0.9.54; extra == 'agent'
+Requires-Dist: pyright==1.1.401; extra == 'agent'
+Requires-Dist: pytest-asyncio; extra == 'agent'
+Requires-Dist: pytest-cov; extra == 'agent'
+Requires-Dist: pytest-mock; extra == 'agent'
+Requires-Dist: pytest<9,>=8.1.1; extra == 'agent'
+Requires-Dist: ruff>=0.11.8; extra == 'agent'
+Requires-Dist: setuptools; extra == 'agent'
+Requires-Dist: textdistance<5,>=4.5.0; extra == 'agent'
 Provides-Extra: agents
+Requires-Dist: aiodocker>=0.24.0; extra == 'agents'
 Requires-Dist: dotenv>=0.9.9; extra == 'agents'
+Requires-Dist: inspect-ai>=0.3.80; extra == 'agents'
 Requires-Dist: ipykernel; extra == 'agents'
 Requires-Dist: ipython<9; extra == 'agents'
 Requires-Dist: jupyter-client; extra == 'agents'
@@ -76,6 +91,17 @@ Requires-Dist: jupyter-core; extra == 'agents'
 Requires-Dist: langchain; extra == 'agents'
 Requires-Dist: langchain-anthropic; extra == 'agents'
 Requires-Dist: langchain-openai; extra == 'agents'
+Requires-Dist: pillow>=11.1.0; extra == 'agents'
+Requires-Dist: playwright; extra == 'agents'
+Requires-Dist: pyautogui>=0.9.54; extra == 'agents'
+Requires-Dist: pyright==1.1.401; extra == 'agents'
+Requires-Dist: pytest-asyncio; extra == 'agents'
+Requires-Dist: pytest-cov; extra == 'agents'
+Requires-Dist: pytest-mock; extra == 'agents'
+Requires-Dist: pytest<9,>=8.1.1; extra == 'agents'
+Requires-Dist: ruff>=0.11.8; extra == 'agents'
+Requires-Dist: setuptools; extra == 'agents'
+Requires-Dist: textdistance<5,>=4.5.0; extra == 'agents'
 Provides-Extra: dev
 Requires-Dist: aiodocker>=0.24.0; extra == 'dev'
 Requires-Dist: dotenv>=0.9.9; extra == 'dev'
@@ -100,14 +126,6 @@ Requires-Dist: setuptools; extra == 'dev'
 Requires-Dist: textdistance<5,>=4.5.0; extra == 'dev'
 Provides-Extra: rl
 Requires-Dist: bitsandbytes>=0.41.0; (sys_platform == 'linux') and extra == 'rl'
-Requires-Dist: dotenv>=0.9.9; extra == 'rl'
-Requires-Dist: ipykernel; extra == 'rl'
-Requires-Dist: ipython<9; extra == 'rl'
-Requires-Dist: jupyter-client; extra == 'rl'
-Requires-Dist: jupyter-core; extra == 'rl'
-Requires-Dist: langchain; extra == 'rl'
-Requires-Dist: langchain-anthropic; extra == 'rl'
-Requires-Dist: langchain-openai; extra == 'rl'
 Requires-Dist: liger-kernel>=0.5.0; (sys_platform == 'linux') and extra == 'rl'
 Requires-Dist: peft>=0.17.1; extra == 'rl'
 Requires-Dist: vllm==0.10.1.1; extra == 'rl'

hud_python-0.4.36/environments/blank/README.md

@@ -0,0 +1,92 @@
+# test-test
+
+## Environment design pattern
+- Controller (Think of this as a frontend in web development)
+  - Creates the UX and manages the lifecycle of an app (in this case for an agent)
+  - Define `mcp = MCPServer()` and register `@mcp.tool` as tools the agent can interact with
+- Environment (Think of this as a backend in web development)
+  - Owns all long‑lived states of the environment and exposes the environment data structure
+  - Expose simple HTTP endpoints (`/health`, `/act`, `/reset`, `/state`)
+
+IMPORTANT: Make sure all logs are going to stderr instead of stdio, which is reserved for MCP communication
+
+### Interactive Development
+```bash
+# 1. Configure your API keys (optional - only needed for evaluation)
+# Edit .env file to add your HUD_API_KEY and ANTHROPIC_API_KEY
+
+# 2. Start the environment (optional: with --inspector or --interactive)
+hud dev --build --interactive
+
+# 3. Choose your preferred way to test:
+
+# Option A: Run the task with Claude (requires ANTHROPIC_API_KEY)
+hud eval tasks.json --agent claude
+
+# Option B: Interactive notebook test_env.ipynb (great for learning!)
+# Requires installation:
+pip install hud-python[agents]
+
+# Option C: Simple Python script (runs all tasks from tasks.json)
+python test_task.py
+```
+
+## Layout
+```
+controller/
+  __init__.py   # mcp + shared HTTP client
+  __main__.py   # python -m controller → mcp.run()
+  hooks.py      # @mcp.initialize / @mcp.shutdown
+  tools.py      # @mcp.tool act / setup / evaluate
+
+./environment
+  ├── __init__.py
+  └── server.py       # FastAPI app: /health, /act, /reset, /state
+```
+
+## Publishing Your Environment
+
+Once your environment is ready, you can share it with the community:
+
+### 1. Push to Registry
+```bash
+# Build and push your environment (requires docker hub login and hud api key)
+hud build
+hud push
+```
+
+### 2. Create a Dataset
+
+Create a dataset on HuggingFace with your tasks:
+
+**Option A: Upload manually**
+1. Upload your `tasks.json` to HuggingFace
+2. Make sure it's **public** to appear on leaderboards
+
+**Option B: Use the SDK**
+```python
+from hud.datasets import save_tasks
+import json
+
+# Load your tasks
+with open("tasks.json") as f:
+    tasks = json.load(f)
+
+# Push to HuggingFace
+save_tasks(tasks, repo_id="your-org/your-dataset")
+```
+
+### 3. Run and Track Performance
+
+```bash
+# Run Claude on your benchmark
+hud eval "your-org/your-dataset" --agent claude
+
+# View results at:
+# app.hud.so/leaderboards/your-org/your-dataset
+```
+
+**Note**: Only public HuggingFace datasets appear as leaderboards!
+
+📚 Learn more: [Creating Benchmarks](https://docs.hud.so/evaluate-agents/create-benchmarks) | [Leaderboards](https://docs.hud.so/evaluate-agents/leaderboards)
+

hud_python-0.4.36/environments/blank/controller/README.md

@@ -0,0 +1,16 @@
+# Controller
+
+Frontend for the agent: defines tools, minimal state, calls the environment over HTTP.
+
+What to implement
+- Shared client in `__init__.py` (one `httpx.AsyncClient`)
+- Lifecycle in `hooks.py` (`@mcp.initialize`/`@mcp.shutdown`)
+- Tools in `tools.py` (`@mcp.tool`) — keep logic thin; docstrings = descriptions
+
+Run
+```bash
+hud run controller --transport http --reload
+# Helper endpoints: http://localhost:8765/hud and /hud/tools
+```
+
+Principle: the controller is UX, not state. Keep long‑lived state in the environment.

hud_python-0.4.36/environments/blank/environment/README.md

@@ -0,0 +1,16 @@
+# Environment
+
+Backend service: owns state and exposes HTTP APIs the controller calls.
+
+Endpoints (FastAPI)
+- `GET /health` → {status: ok}
+- `POST /act` → increments counter and returns {count}
+- `POST /reset` → resets counter
+- `GET /state` → returns {count}
+
+Run (dev)
+```bash
+uv run uvicorn environment.server:app --reload --port 8005
+```
+
+Principle: treat like a backend. Keep long‑lived state here; add endpoints as tools need them.

hud_python-0.4.36/environments/blank/pyproject.toml

@@ -0,0 +1,19 @@
+[project]
+name = "test_test"
+version = "0.1.0"
+description = "A minimal HUD environment"
+requires-python = ">=3.11"
+dependencies = [ "hud-python==0.4.36", "fastapi", "uvicorn[standard]", "httpx>=0.28.1",]
+
+[build-system]
+requires = [ "hatchling",]
+build-backend = "hatchling.build"
+
+[tool.hud]
+image = "test_test:dev"
+
+[tool.hatch.metadata]
+allow-direct-references = true
+
+[tool.hatch.build.targets.wheel]
+packages = [ "controller", "environment",]

{hud_python-0.4.35 → hud_python-0.4.36}/environments/browser/README.md

@@ -2,100 +2,99 @@
 
 A browser automation environment for HUD that provides GUI access and web app interaction capabilities. This environment supports hot-reloading during development while maintaining persistent state.
 
-## Architecture Overview
+## Quick Start
 
-The browser environment uses a two-process architecture:
+### Interactive Development
+```bash
+# 1. Configure your API keys (optional - only needed for evaluation)
+# Edit .env file to add your HUD_API_KEY and ANTHROPIC_API_KEY
 
-1. **Context Server** (`context.py`): Long-running process that maintains persistent state
-2. **MCP Server** (`server.py`): Hot-reloadable process that handles tool requests
+# 2. Start the environment (optional: with inspector)
+hud dev --build --inspector
 
-### Key Components
+# 3. Choose your preferred way to test:
 
-- **BrowserContext**: Stores persistent state (running apps, ports, playwright instance)
-- **ServiceManager**: Manages X11, VNC, and app processes
-- **BaseHub Tools**: Setup and evaluate tools organized by app (2048, todo)
-- **Multiprocessing Proxy**: Enables state sharing between processes
+# Option A: Run the task with Claude (requires ANTHROPIC_API_KEY)
+hud eval tasks.json --agent claude
 
-## Context Management and Common Pitfalls
+# Option B: Interactive notebook test_env.ipynb (great for learning!)
+# Requires installation:
+pip install hud-python[agents]
 
-### Understanding the Proxy System
+# Option C: Simple Python script (runs all tasks from tasks.json)
+python test_task.py
+```
 
-The browser environment uses Python's `multiprocessing.Manager` to share state between the context server and MCP server. This introduces important constraints:
+## How HUD Environments Work
 
-#### ❌ Common Pitfall: Unpicklable Objects
+The environment is split into two components:
 
-```python
-# BAD: This will fail with "cannot pickle 'coroutine' object"
-@setup.tool("my_tool")
-async def my_tool():
-    env = setup.env
-    result = await env.call_app_api("app", "/api/endpoint")  # Returns coroutine
-    # The coroutine can't be serialized through the proxy!
-```
+- **`env.py`** - Stateful logic that persists across reloads
+- **`server.py`** - MCP server with tools (reloads on file changes)
 
-#### ✅ Solution: Direct HTTP Calls
+This separation is crucial for `hud dev` - it allows you to modify the MCP tools and see changes immediately without losing the environment state. The environment runs as a separate process and communicates via socket, while the server can be restarted freely.
 
-```python
-# GOOD: Make HTTP calls directly
-@setup.tool("my_tool")
-async def my_tool():
-    import httpx
-    
-    # Get the backend port from persistent context
-    persistent_ctx = setup.env
-    backend_port = persistent_ctx.get_app_backend_port("app")
-    
-    # Make API call directly
-    url = f"http://localhost:{backend_port}/api/endpoint"
-    async with httpx.AsyncClient() as client:
-        response = await client.get(url)
-        response.raise_for_status()
-        result = response.json()
-```
+If you are ever seeing issues with the environment itself, running `hud dev --full-reload` will reload both the environment and the server.
 
-### State Synchronization Issues
+## Publishing Your Environment
 
-#### ❌ Common Pitfall: Direct List/Dict Manipulation
+Once your environment is ready, you can share it with the community:
 
-```python
-# BAD: Regular Python lists don't sync through proxy
-class ServiceManager:
-    def __init__(self):
-        self._launched_apps = []  # Won't sync!
+### 1. Push to Registry
+```bash
+# Build and push your environment (requires docker hub login and hud api key)
+hud build
+hud push
 ```
 
-#### ✅ Solution: Store State in Persistent Context
+### 2. Create a Dataset
 
+Create a dataset on HuggingFace with your tasks:
+
+**Option A: Upload manually**
+1. Upload your `tasks.json` to HuggingFace
+2. Make sure it's **public** to appear on leaderboards
+
+**Option B: Use the SDK**
 ```python
-# GOOD: Use the persistent context for shared state
-class BrowserContext:
-    def __init__(self):
-        self._running_apps: List[str] = []
-        self._app_ports: Dict[str, Dict[str, int]] = {}
-    
-    def add_running_app(self, app_name: str) -> None:
-        """Add app to running list."""
-        if app_name not in self._running_apps:
-            self._running_apps.append(app_name)
+from hud.datasets import save_tasks
+import json
+
+# Load your tasks
+with open("tasks.json") as f:
+    tasks = json.load(f)
+
+# Push to HuggingFace
+save_tasks(tasks, repo_id="your-org/your-dataset")
 ```
 
-### Accessing Shared Resources
+### 3. Run and Track Performance
 
-#### ❌ Common Pitfall: Direct Attribute Access
+```bash
+# Run Claude on your benchmark
+hud eval "your-org/your-dataset" --agent claude
 
-```python
-# BAD: Direct attribute access on proxy objects
-playwright_tool = env.playwright  # May not work with proxy
+# View results at:
+# app.hud.so/leaderboards/your-org/your-dataset
 ```
 
-#### ✅ Solution: Use Getter Methods
+**Note**: Only public HuggingFace datasets appear as leaderboards!
 
-```python
-# GOOD: Use proxy-friendly getter methods
-playwright_tool = persistent_ctx.get_playwright_tool()
-```
+📚 Learn more: [Creating Benchmarks](https://docs.hud.so/evaluate-agents/create-benchmarks) | [Leaderboards](https://docs.hud.so/evaluate-agents/leaderboards)
 
-## Best Practices
+## Architecture Overview
+
+The browser environment uses a two-process architecture:
+
+1. **Context Server** (`context.py`): Long-running process that maintains persistent state
+2. **MCP Server** (`server.py`): Hot-reloadable process that handles tool requests
+
+### Key Components
+
+- **BrowserContext**: Stores persistent state (running apps, ports, playwright instance)
+- **ServiceManager**: Manages X11, VNC, and app processes
+- **BaseHub Tools**: Setup and evaluate tools organized by app (2048, todo)
+- **Multiprocessing Proxy**: Enables state sharing between processes
 
 ### 1. Tool Implementation Pattern
 
@@ -166,26 +165,6 @@ from . import setup
 # Not inside functions
 ```
 
-## Troubleshooting
-
-### "Cannot pickle 'coroutine' object"
-
-**Cause**: Trying to return an async function result through the proxy.
-
-**Fix**: Don't use async methods on proxied objects. Make direct HTTP calls instead.
-
-### "App not launched" errors
-
-**Cause**: State synchronization issue between ServiceManager and persistent context.
-
-**Fix**: Ensure `launch_app` stores app info in the persistent context, and setup/evaluate tools check the persistent context's app list.
-
-### "Object has no attribute" on proxy objects
-
-**Cause**: Direct attribute access on multiprocessing proxy objects.
-
-**Fix**: Use getter/setter methods instead of direct attribute access.
-
 ## Development Workflow
 
 1. **Start the environment**: `hud dev`

hud_python-0.4.36/environments/browser/environment/pyproject.toml

@@ -0,0 +1,20 @@
+[project]
+name = "browser-environment"
+version = "0.1.0"
+description = "Browser environment server for managing X11, VNC, and applications"
+requires-python = ">=3.11"
+dependencies = [
+    "fastapi>=0.104.1",
+    "uvicorn[standard]>=0.24.0",
+    "httpx>=0.25.2",
+    "pydantic>=2.6,<3",
+    "pydantic-settings>=2.2,<3",
+    "python-multipart>=0.0.6",
+]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["controller", "environment"]

hud_python-0.4.36/environments/browser/pyproject.toml

@@ -0,0 +1,22 @@
+[project]
+name = "hud-browser-controller"
+version = "0.1.0"
+description = "HUD Browser Controller - MCP interface for browser environments"
+requires-python = ">=3.11,<3.14"
+dependencies = [ "pydantic>=2.6,<3", "pydantic-settings>=2.2,<3", "hud-python@git+https://github.com/hud-evals/hud-python@env-cli-improvements", "playwright", "pyautogui", "httpx", "typer", "fastapi", "uvicorn",]
+
+[build-system]
+requires = [ "hatchling",]
+build-backend = "hatchling.build"
+
+[project.scripts]
+hud-browser-controller = "controller.__main__:main"
+
+[tool.hud]
+image = "hud-browser:dev"
+
+[tool.hatch.metadata]
+allow-direct-references = true
+
+[tool.hatch.build.targets.wheel]
+packages = [ "controller", "problems",]

hud_python-0.4.36/environments/deepresearch/pyproject.toml

@@ -0,0 +1,19 @@
+[project]
+name = "deepresearch"
+version = "0.1.0"
+description = "DeepResearch HUD environment with HTTP backend (EXA on server)"
+requires-python = ">=3.11"
+dependencies = [ "hud-python==0.4.36", "fastapi>=0.104.1", "uvicorn[standard]>=0.24.0", "httpx>=0.24.0",]
+
+[build-system]
+requires = [ "hatchling",]
+build-backend = "hatchling.build"
+
+[tool.hud]
+image = "deepresearch:dev"
+
+[tool.hatch.metadata]
+allow-direct-references = true
+
+[tool.hatch.build.targets.wheel]
+packages = [ "controller", "environment",]

{hud_python-0.4.35 → hud_python-0.4.36}/hud/agents/tests/test_claude.py

@@ -86,6 +86,7 @@ class TestClaudeAgent:
             model_client=mock_model_client,
             model="claude-3-opus-20240229",
             max_tokens=1000,
+            validate_api_key=False,  # Skip validation in tests
         )
 
         assert agent.model_name == "claude-3-opus-20240229"
@@ -93,10 +94,14 @@ class TestClaudeAgent:
         assert agent.anthropic_client == mock_model_client
 
     @pytest.mark.asyncio
-    async def test_init_without_model_client(self, mock_mcp_client):
+    async def test_init_without_model_client(self, mock_mcp_client, mock_anthropic):
         """Test agent initialization without model client."""
         with patch("hud.settings.settings.anthropic_api_key", "test_key"):
-            agent = ClaudeAgent(mcp_client=mock_mcp_client, model="claude-3-opus-20240229")
+            agent = ClaudeAgent(
+                mcp_client=mock_mcp_client,
+                model="claude-3-opus-20240229",
+                validate_api_key=False,  # Skip validation in tests
+            )
 
             assert agent.model_name == "claude-3-opus-20240229"
             assert agent.anthropic_client is not None
@@ -105,7 +110,11 @@ class TestClaudeAgent:
     async def test_format_blocks(self, mock_mcp_client):
         """Test formatting content blocks into Claude messages."""
         mock_model_client = MagicMock()
-        agent = ClaudeAgent(mcp_client=mock_mcp_client, model_client=mock_model_client)
+        agent = ClaudeAgent(
+            mcp_client=mock_mcp_client,
+            model_client=mock_model_client,
+            validate_api_key=False,  # Skip validation in tests
+        )
 
         # Test with text only
         text_blocks: list[types.ContentBlock] = [
@@ -141,7 +150,11 @@ class TestClaudeAgent:
     async def test_format_tool_results_method(self, mock_mcp_client):
         """Test the agent's format_tool_results method."""
         mock_model_client = MagicMock()
-        agent = ClaudeAgent(mcp_client=mock_mcp_client, model_client=mock_model_client)
+        agent = ClaudeAgent(
+            mcp_client=mock_mcp_client,
+            model_client=mock_model_client,
+            validate_api_key=False,  # Skip validation in tests
+        )
 
         tool_calls = [
             MCPToolCall(name="test_tool", arguments={}, id="id1"),
@@ -171,7 +184,11 @@ class TestClaudeAgent:
         """Test getting model response from Claude API."""
         # Disable telemetry for this test to avoid backend configuration issues
         with patch("hud.settings.settings.telemetry_enabled", False):
-            agent = ClaudeAgent(mcp_client=mock_mcp_client, model_client=mock_anthropic)
+            agent = ClaudeAgent(
+                mcp_client=mock_mcp_client,
+                model_client=mock_anthropic,
+                validate_api_key=False,  # Skip validation in tests
+            )
 
             # Mock the API response
             mock_response = MagicMock()
@@ -215,7 +232,11 @@ class TestClaudeAgent:
         """Test getting text-only response."""
         # Disable telemetry for this test to avoid backend configuration issues
         with patch("hud.settings.settings.telemetry_enabled", False):
-            agent = ClaudeAgent(mcp_client=mock_mcp_client, model_client=mock_anthropic)
+            agent = ClaudeAgent(
+                mcp_client=mock_mcp_client,
+                model_client=mock_anthropic,
+                validate_api_key=False,  # Skip validation in tests
+            )
 
             mock_response = MagicMock()
             # Create text block
@@ -242,7 +263,11 @@ class TestClaudeAgent:
         """Test handling API errors."""
         # Disable telemetry for this test to avoid backend configuration issues
         with patch("hud.settings.settings.telemetry_enabled", False):
-            agent = ClaudeAgent(mcp_client=mock_mcp_client, model_client=mock_anthropic)
+            agent = ClaudeAgent(
+                mcp_client=mock_mcp_client,
+                model_client=mock_anthropic,
+                validate_api_key=False,  # Skip validation in tests
+            )
 
             # Mock API error
             mock_anthropic.beta.messages.create = AsyncMock(

{hud_python-0.4.35 → hud_python-0.4.36}/hud/agents/tests/test_openai.py

@@ -44,7 +44,10 @@ class TestOperatorAgent:
         """Test agent initialization."""
         mock_model_client = MagicMock()
         agent = OperatorAgent(
-            mcp_client=mock_mcp_client, model_client=mock_model_client, model="gpt-4"
+            mcp_client=mock_mcp_client,
+            model_client=mock_model_client,
+            model="gpt-4",
+            validate_api_key=False,  # Skip validation in tests
         )
 
         assert agent.model_name == "openai-gpt-4"
@@ -55,7 +58,11 @@ class TestOperatorAgent:
     async def test_format_blocks(self, mock_mcp_client):
         """Test formatting content blocks."""
         mock_model_client = MagicMock()
-        agent = OperatorAgent(mcp_client=mock_mcp_client, model_client=mock_model_client)
+        agent = OperatorAgent(
+            mcp_client=mock_mcp_client,
+            model_client=mock_model_client,
+            validate_api_key=False,  # Skip validation in tests
+        )
 
         # Test with text blocks
         blocks: list[types.ContentBlock] = [
@@ -85,7 +92,11 @@ class TestOperatorAgent:
     @pytest.mark.asyncio
     async def test_format_tool_results(self, mock_mcp_client, mock_openai):
         """Test formatting tool results."""
-        agent = OperatorAgent(mcp_client=mock_mcp_client, model_client=mock_openai)
+        agent = OperatorAgent(
+            mcp_client=mock_mcp_client,
+            model_client=mock_openai,
+            validate_api_key=False,  # Skip validation in tests
+        )
 
         tool_calls = [
             MCPToolCall(name="test_tool", arguments={}, id="call_123"),  # type: ignore
@@ -111,7 +122,11 @@ class TestOperatorAgent:
     @pytest.mark.asyncio
     async def test_format_tool_results_with_error(self, mock_mcp_client, mock_openai):
         """Test formatting tool results with errors."""
-        agent = OperatorAgent(mcp_client=mock_mcp_client, model_client=mock_openai)
+        agent = OperatorAgent(
+            mcp_client=mock_mcp_client,
+            model_client=mock_openai,
+            validate_api_key=False,  # Skip validation in tests
+        )
 
         tool_calls = [
             MCPToolCall(name="failing_tool", arguments={}, id="call_error"),  # type: ignore
@@ -131,7 +146,11 @@ class TestOperatorAgent:
     @pytest.mark.asyncio
     async def test_get_model_response(self, mock_mcp_client, mock_openai):
         """Test getting model response from OpenAI API."""
-        agent = OperatorAgent(mcp_client=mock_mcp_client, model_client=mock_openai)
+        agent = OperatorAgent(
+            mcp_client=mock_mcp_client,
+            model_client=mock_openai,
+            validate_api_key=False,  # Skip validation in tests
+        )
 
         # Set up available tools so agent doesn't return "No computer use tools available"
         agent._available_tools = [
@@ -162,7 +181,11 @@ class TestOperatorAgent:
     @pytest.mark.asyncio
     async def test_handle_empty_response(self, mock_mcp_client, mock_openai):
         """Test handling empty response from API."""
-        agent = OperatorAgent(mcp_client=mock_mcp_client, model_client=mock_openai)
+        agent = OperatorAgent(
+            mcp_client=mock_mcp_client,
+            model_client=mock_openai,
+            validate_api_key=False,  # Skip validation in tests
+        )
 
         # Set up available tools
         agent._available_tools = [