nous-hermeshub Computer Use Agents
Build AI agents that interact with computers like humans do - viewing screens, moving cursors, clicking buttons, and typing text. Covers Anthropic's Computer Use, OpenAI's Operator/CUA, and open-source alternatives. Critical focus on sandboxing, security, and handling the unique challenges of vision-based control.
Patterns
Perception-Reasoning-Action Loop
The fundamental architecture of computer use agents: observe screen, reason about next action, execute action, repeat. This loop integrates vision models with action execution through an iterative pipeline.
Key components:
- PERCEPTION: Screenshot captures current screen state
- REASONING: Vision-language model analyzes and plans
- ACTION: Execute mouse/keyboard operations
- FEEDBACK: Observe result, continue or correct
Critical insight: Vision agents are completely still during "thinking" phase (1-5 seconds), creating a detectable pause pattern.
When to use: Building any computer use agent from scratch,Integrating vision models with desktop control,Understanding agent behavior patterns
from anthropic import Anthropic from PIL import Image import base64 import pyautogui import time
class ComputerUseAgent: """ Perception-Reasoning-Action loop implementation. Based on Anthropic Computer Use patterns. """
def __init__(self, client: Anthropic, model: str = "claude-sonnet-4-20250514"):
self.client = client
self.model = model
self.max_steps = 50 # Prevent runaway loops
self.action_delay = 0.5 # Seconds between actions
def capture_screenshot(self) -> str:
"""Capture screen and return base64 encoded image."""
screenshot = pyautogui.screenshot()
# Resize for token efficiency (1280x800 is good balance)
screenshot = screenshot.resize((1280, 800), Image.LANCZOS)
import io
buffer = io.BytesIO()
screenshot.save(buffer, format="PNG")
return base64.b64encode(buffer.getvalue()).decode()
def execute_action(self, action: dict) -> dict:
"""Execute mouse/keyboard action on the computer."""
action_type = action.get("type")
if action_type == "click":
x, y = action["x"], action["y"]
button = action.get("button", "left")
pyautogui.click(x, y, button=button)
return {"success": True, "action": f"clicked at ({x}, {y})"}
elif action_type == "type":
text = action["text"]
pyautogui.typewrite(text, interval=0.02)
return {"success": True, "action": f"typed {len(text)} chars"}
elif action_type == "key":
key = action["key"]
pyautogui.press(key)
return {"success": True, "action": f"pressed {key}"}
elif action_type == "scroll":
direction = action.get("direction", "down")
amount = action.get("amount", 3)
scroll = -amount if direction == "down" else amount
pyautogui.scroll(scroll)
return {"success": True, "action": f"scrolled {direction}"}
elif action_type == "move":
x, y = action["x"], action["y"]
pyautogui.moveTo(x, y)
return {"success": True, "action": f"moved to ({x}, {y})"}
else:
return {"success": False, "error": f"Unknown action: {action_type}"}
def run(self, task: str) -> dict:
"""
Run perception-reasoning-action loop until task complete.
The loop:
1. Screenshot current state
2. Send to vision model with task context
3. Parse action from response
4. Execute action
5. Repeat until done or max steps
"""
messages = []
step_count = 0
system_prompt = """You are a computer use agent. You can see the screen
and control mouse/keyboard.
Available actions (respond with JSON):
- {"type": "click", "x": 100, "y": 200, "button": "left"}
- {"type": "type", "text": "hello world"}
- {"type": "key", "key": "enter"}
- {"type": "scroll", "direction": "down", "amount": 3}
- {"type": "done", "result": "task completed successfully"}
Always respond with ONLY a JSON action object.
Be precise with coordinates - click exactly where needed.
If you see an error, try to recover.
"""
while step_count < self.max_steps:
step_count += 1
# 1. PERCEPTION: Capture current screen
screenshot_b64 = self.capture_screenshot()
# 2. REASONING: Send to vision model
user_content = [
{"type": "text", "text": f"Task: {task}\n\nStep {step_count}. What action should I take?"},
{"type": "image", "source": {
"type": "base64",
"media_type": "image/png",
"data": screenshot_b64
}}
]
messages.append({"role": "user", "content": user_content})
response = self.client.messages.create(
model=self.model,
max_tokens=1024,
system=system_prompt,
messages=messages
)
assistant_message = response.content[0].text
messages.append({"role": "assistant", "content": assistant_message})
# 3. Parse action from response
import json
try:
action = json.loads(assistant_message)
except json.JSONDecodeError:
# Try to extract JSON from response
import re
match = re.search(r'\{[^}]+\}', assistant_message)
if match:
action = json.loads(match.group())
else:
continue
# Check if done
if action.get("type") == "done":
return {
"success": True,
"result": action.get("result"),
"steps": step_count
}
# 4. ACTION: Execute
result = self.execute_action(action)
# Small delay for UI to update
time.sleep(self.action_delay)
return {
"success": False,
"error": "Max steps reached",
"steps": step_count
}
Usage
agent = ComputerUseAgent(Anthropic()) result = agent.run("Open Chrome and search for 'weather today'")
Anti_patterns
- Running without step limits (infinite loops)
- No delay between actions (UI can't keep up)
- Screenshots at full resolution (token explosion)
- Ignoring action failures (no recovery)
Sandboxed Environment Pattern
Computer use agents MUST run in isolated, sandboxed environments. Never give agents direct access to your main system - the security risks are too high. Use Docker containers with virtual desktops.
Key isolation requirements:
- NETWORK: Restrict to necessary endpoints only
- FILESYSTEM: Read-only or scoped to temp directories
- CREDENTIALS: No access to host credentials
- SYSCALLS: Filter dangerous system calls
- RESOURCES: Limit CPU, memory, time
The goal is "blast radius minimization" - if the agent goes wrong, damage is contained to the sandbox.
When to use: Deploying any computer use agent,Testing agent behavior safely,Running untrusted automation tasks
Dockerfile for sandboxed computer use environment
Based on Anthropic's reference implementation pattern
FROM ubuntu:22.04
Install desktop environment
RUN apt-get update && apt-get install -y
xvfb
x11vnc
fluxbox
xterm
firefox
python3
python3-pip
supervisor
Security: Create non-root user
RUN useradd -m -s /bin/bash agent &&
mkdir -p /home/agent/.vnc
Install Python dependencies
COPY requirements.txt /tmp/ RUN pip3 install -r /tmp/requirements.txt
Security: Drop capabilities
RUN apt-get install -y --no-install-recommends libcap2-bin &&
setcap -r /usr/bin/python3 || true
Copy agent code
COPY --chown=agent:agent . /app WORKDIR /app
Supervisor config for virtual display + VNC
COPY supervisord.conf /etc/supervisor/conf.d/
Expose VNC port only (not desktop directly)
EXPOSE 5900
Run as non-root
USER agent
CMD ["/usr/bin/supervisord", "-c", "/etc/supervisor/conf.d/supervisord.conf"]
docker-compose.yml with security constraints
version: '3.8'
services: computer-use-agent: build: . ports: - "5900:5900" # VNC for observation - "8080:8080" # API for control
# Security constraints
security_opt:
- no-new-privileges:true
- seccomp:seccomp-profile.json
# Resource limits
deploy:
resources:
limits:
cpus: '2'
memory: 4G
reservations:
cpus: '0.5'
memory: 1G
# Network isolation
networks:
- agent-network
# No access to host filesystem
volumes:
- agent-tmp:/tmp
# Read-only root filesystem
read_only: true
tmpfs:
- /run
- /var/run
# Environment
environment:
- DISPLAY=:99
- NO_PROXY=localhost
networks: agent-network: driver: bridge internal: true # No internet by default
volumes: agent-tmp:
Python wrapper with additional runtime sandboxing
import subprocess import os from dataclasses import dataclass from typing import Optional
@dataclass class SandboxConfig: """Configuration for agent sandbox.""" network_allowed: list[str] = None # Allowed domains max_runtime_seconds: int = 300 max_memory_mb: int = 2048 allow_downloads: bool = False allow_clipboard: bool = False
class SandboxedAgent: """ Run computer use agent in Docker sandbox. """
def __init__(self, config: SandboxConfig):
self.config = config
self.container_id: Optional[str] = None
def start(self):
"""Start sandboxed environment."""
# Build network rules
network_rules = ""
if self.config.network_allowed:
for domain in self.config.network_allowed:
network_rules += f"--add-host={domain}:$(dig +short {domain}) "
else:
network_rules = "--network=none"
cmd = f"""
docker run -d \
--name computer-use-sandbox-$$ \
--security-opt no-new-privileges \
--cap-drop ALL \
--memory {self.config.max_memory_mb}m \
--cpus 2 \
--read-only \
--tmpfs /tmp \
{network_rules} \
computer-use-agent:latest
"""
result = subprocess.run(cmd, shell=True, capture_output=True)
self.container_id = result.stdout.decode().strip()
# Set up kill timer
subprocess.Popen([
"sh", "-c",
f"sleep {self.config.max_runtime_seconds} && docker kill {self.container_id}"
])
return self.container_id
def execute_task(self, task: str) -> dict:
"""Execute task in sandbox."""
if not self.container_id:
self.start()
# Send task to agent via API
import requests
response = requests.post(
f"http://localhost:8080/task",
json={"task": task},
timeout=self.config.max_runtime_seconds
)
return response.json()
def stop(self):
"""Stop and remove sandbox."""
if self.container_id:
subprocess.run(f"docker rm -f {self.container_id}", shell=True)
self.container_id = None
Anti_patterns
- Running agents on host system directly
- Giving sandbox full network access
- Running as root in container
- No resource limits (denial of service)
- Persistent storage (data can leak between runs)
Anthropic Computer Use Implementation
Official implementation pattern using Claude's computer use capability. Claude 3.5 Sonnet was the first frontier model to offer computer use. Claude Opus 4.5 is now the "best model in the world for computer use."
Key capabilities:
- screenshot: Capture current screen state
- mouse: Click, move, drag operations
- keyboard: Type text, press keys
- bash: Run shell commands
- text_editor: View and edit files
Tool versions:
- computer_20251124 (Opus 4.5): Adds zoom action for detailed inspection
- computer_20250124 (All other models): Standard capabilities
Critical limitation: "Some UI elements (like dropdowns and scrollbars) might be tricky for Claude to manipulate" - Anthropic docs
When to use: Building production computer use agents,Need highest quality vision understanding,Full desktop control (not just browser)
from anthropic import Anthropic from anthropic.types.beta import ( BetaToolComputerUse20241022, BetaToolBash20241022, BetaToolTextEditor20241022, ) import subprocess import base64 from PIL import Image import io
class AnthropicComputerUse: """ Official Anthropic Computer Use implementation.
Requires:
- Docker container with virtual display
- VNC for viewing agent actions
- Proper tool implementations
"""
def __init__(self):
self.client = Anthropic()
self.model = "claude-sonnet-4-20250514" # Best for computer use
self.screen_size = (1280, 800)
def get_tools(self) -> list:
"""Define computer use tools."""
return [
BetaToolComputerUse20241022(
type="computer_20241022",
name="computer",
display_width_px=self.screen_size[0],
display_height_px=self.screen_size[1],
),
BetaToolBash20241022(
type="bash_20241022",
name="bash",
),
BetaToolTextEditor20241022(
type="text_editor_20241022",
name="str_replace_editor",
),
]
def execute_tool(self, name: str, input: dict) -> dict:
"""Execute a tool and return result."""
if name == "computer":
return self._handle_computer_action(input)
elif name == "bash":
return self._handle_bash(input)
elif name == "str_replace_editor":
return self._handle_editor(input)
else:
return {"error": f"Unknown tool: {name}"}
def _handle_computer_action(self, input: dict) -> dict:
"""Handle computer control actions."""
action = input.get("action")
if action == "screenshot":
# Capture via xdotool/scrot
subprocess.run(["scrot", "/tmp/screenshot.png"])
with open("/tmp/screenshot.png", "rb") as f:
img_data = f.read()
# Resize for efficiency
img = Image.open(io.BytesIO(img_data))
img = img.resize(self.screen_size, Image.LANCZOS)
buffer = io.BytesIO()
img.save(buffer, format="PNG")
return {
"type": "image",
"source": {
"type": "base64",
"media_type": "image/png",
"data": base64.b64encode(buffer.getvalue()).decode()
}
}
elif action == "mouse_move":
x, y = input.get("coordinate", [0, 0])
subprocess.run(["xdotool", "mousemove", str(x), str(y)])
return {"success": True}
elif action == "left_click":
subprocess.run(["xdotool", "click", "1"])
return {"success": True}
elif action == "right_click":
subprocess.run(["xdotool", "click", "3"])
return {"success": True}
elif action == "double_click":
subprocess.run(["xdotool", "click", "--repeat", "2", "1"])
return {"success": True}
elif action == "type":
text = input.get("text", "")
# Use xdotool type with delay for reliability
subprocess.run(["xdotool", "type", "--delay", "50", text])
return {"success": True}
elif action == "key":
key = input.get("key", "")
# Map common key names
key_map = {
"return": "Return",
"enter": "Return",
"tab": "Tab",
"escape": "Escape",
"backspace": "BackSpace",
}
xdotool_key = key_map.get(key.lower(), key)
subprocess.run(["xdotool", "key", xdotool_key])
return {"success": True}
elif action == "scroll":
direction = input.get("direction", "down")
amount = input.get("amount", 3)
button = "5" if direction == "down" else "4"
for _ in range(amount):
subprocess.run(["xdotool", "click", button])
return {"success": True}
return {"error": f"Unknown action: {action}"}
def _handle_bash(self, input: dict) -> dict:
"""Execute bash command."""
command = input.get("command", "")
# Security: Sanitize and limit commands
dangerous_patterns = ["rm -rf", "mkfs", "dd if=", "> /dev/"]
for pattern in dangerous_patterns:
if pattern in command:
return {"error": "Dangerous command blocked"}
try:
result = subprocess.run(
command,
shell=True,
capture_output=True,
text=True,
timeout=30
)
return {
"stdout": result.stdout[:10000], # Limit output
"stderr": result.stderr[:1000],
"returncode": result.returncode
}
except subprocess.TimeoutExpired:
return {"error": "Command timed out"}
def _handle_editor(self, input: dict) -> dict:
"""Handle text editor operations."""
command = input.get("command")
path = input.get("path")
if command == "view":
try:
with open(path, "r") as f:
content = f.read()
return {"content": content[:50000]} # Limit size
except Exception as e:
return {"error": str(e)}
elif command == "str_replace":
old_str = input.get("old_str")
new_str = input.get("new_str")
try:
with open(path, "r") as f:
content = f.read()
if old_str not in content:
return {"error": "old_str not found in file"}
content = content.replace(old_str, new_str, 1)
with open(path, "w") as f:
f.write(content)
return {"success": True}
except Exception as e:
return {"error": str(e)}
return {"error": f"Unknown editor command: {command}"}
def run_task(self, task: str, max_steps: int = 50) -> dict:
"""Run computer use task with agentic loop."""
messages = [{"role": "user", "content": task}]
tools = self.get_tools()
for step in range(max_steps):
response = self.client.beta.messages.create(
model=self.model,
max_tokens=4096,
tools=tools,
messages=messages,
betas=["computer-use-2024-10-22"]
)
# Check for completion
if response.stop_reason == "end_turn":
return {
"success": True,
"result": response.content[0].text if response.content else "",
"steps": step + 1
}
# Handle tool use
if response.stop_reason == "tool_use":
messages.append({"role": "assistant", "content": response.content})
tool_results = []
for block in response.content:
if block.type == "tool_use":
result = self.execute_tool(block.name, block.input)
tool_results.append({
"type": "tool_result",
"tool_use_id": block.id,
"content": result
})
messages.append({"role": "user", "content": tool_results})
return {"success": False, "error": "Max steps reached"}
Anti_patterns
- Not using betas=['computer-use-2024-10-22'] flag
- Full resolution screenshots (wasteful)
- No command sanitization for bash tool
- Unbounded execution time
Browser-Use Pattern (Playwright-based)
For browser-only automation, using structured DOM access is more efficient than pixel-based computer use. Playwright MCP allows LLMs to control browsers using accessibility snapshots rather than screenshots.
Advantages over vision-based:
- Faster: No image processing required
- Cheaper: Text tokens vs image tokens
- More precise: Direct element targeting
- More reliable: No coordinate drift
When to use vision vs structured:
- Vision: Desktop apps, complex UIs, visual verification
- Structured: Web automation, form filling, data extraction
When to use: Browser-only automation tasks,Form filling and web interactions,When speed and cost matter more than visual understanding
from playwright.async_api import async_playwright from dataclasses import dataclass from typing import Optional import asyncio
@dataclass class BrowserAction: """Structured browser action.""" action: str # click, type, navigate, scroll, extract selector: Optional[str] = None text: Optional[str] = None url: Optional[str] = None
class BrowserUseAgent: """ Browser automation using Playwright with structured commands. More efficient than pixel-based for web tasks. """
def __init__(self):
self.browser = None
self.page = None
async def start(self, headless: bool = True):
"""Start browser session."""
self.playwright = await async_playwright().start()
self.browser = await self.playwright.chromium.launch(headless=headless)
self.page = await self.browser.new_page()
async def get_page_snapshot(self) -> dict:
"""
Get structured snapshot of page for LLM.
Uses accessibility tree for efficiency.
"""
# Get accessibility tree
snapshot = await self.page.accessibility.snapshot()
# Get simplified DOM info
elements = await self.page.evaluate('''() => {
const interactable = [];
const selector = 'a, button, input, select, textarea, [role="button"]';
document.querySelectorAll(selector).forEach((el, i) => {
const rect = el.getBoundingClientRect();
if (rect.width > 0 && rect.height > 0) {
interactable.push({
index: i,
tag: el.tagName.toLowerCase(),
text: el.textContent?.trim().slice(0, 100),
type: el.type,
placeholder: el.placeholder,
name: el.name,
id: el.id,
class: el.className
});
}
});
return interactable;
}''')
return {
"url": self.page.url,
"title": await self.page.title(),
"accessibility_tree": snapshot,
"interactable_elements": elements[:50] # Limit for token efficiency
}
async def execute_action(self, action: BrowserAction) -> dict:
"""Execute structured browser action."""
try:
if action.action == "navigate":
await self.page.goto(action.url, wait_until="domcontentloaded")
return {"success": True, "url": self.page.url}
elif action.action == "click":
await self.page.click(action.selector, timeout=5000)
await self.page.wait_for_load_state("networkidle", timeout=5000)
return {"success": True}
elif action.action == "type":
await self.page.fill(action.selector, action.text)
return {"success": True}
elif action.action == "scroll":
direction = action.text or "down"
distance = 500 if direction == "down" else -500
await self.page.evaluate(f"window.scrollBy(0, {distance})")
return {"success": True}
elif action.action == "extract":
# Extract text content
if action.selector:
text = await self.page.text_content(action.selector)
else:
text = await self.page.text_content("body")
return {"success": True, "text": text[:5000]}
elif action.action == "screenshot":
# Fall back to vision when needed
screenshot = await self.page.screenshot(type="png")
import base64
return {
"success": True,
"image": base64.b64encode(screenshot).decode()
}
except Exception as e:
return {"success": False, "error": str(e)}
return {"success": False, "error": f"Unknown action: {action.action}"}
async def run_with_llm(self, task: str, llm_client, max_steps: int = 20):
"""
Run browser task with LLM decision making.
Uses structured DOM instead of screenshots.
"""
system_prompt = """You are a browser automation agent. You receive
page snapshots with interactable elements and decide actions.
Respond with JSON action:
- {"action": "navigate", "url": "https://..."}
- {"action": "click", "selector": "button.submit"}
- {"action": "type", "selector": "input[name='email']", "text": "..."}
- {"action": "scroll", "text": "down"}
- {"action": "extract", "selector": ".results"}
- {"action": "done", "result": "task completed"}
Use CSS selectors based on the element info provided.
Prefer id > name > class > text content for selectors.
"""
messages = []
for step in range(max_steps):
# Get current page state
snapshot = await self.get_page_snapshot()
user_message = f"""Task: {task}
Current page:
URL: {snapshot['url']}
Title: {snapshot['title']}
Interactable elements:
{snapshot['interactable_elements']}
What action should I take?"""
messages.append({"role": "user", "content": user_message})
# Get LLM decision
response = llm_client.messages.create(
model="claude-sonnet-4-20250514",
max_tokens=1024,
system=system_prompt,
messages=messages
)
assistant_text = response.content[0].text
messages.append({"role": "assistant", "content": assistant_text})
# Parse and execute
import json
action_dict = json.loads(assistant_text)
if action_dict.get("action") == "done":
return {"success": True, "result": action_dict.get("result")}
action = BrowserAction(**action_dict)
result = await self.execute_action(action)
if not result.get("success"):
messages.append({
"role": "user",
"content": f"Action failed: {result.get('error')}"
})
await asyncio.sleep(0.5) # Rate limit
return {"success": False, "error": "Max steps reached"}
async def close(self):
"""Clean up browser."""
if self.browser:
await self.browser.close()
if hasattr(self, 'playwright'):
await self.playwright.stop()
Usage
async def main(): agent = BrowserUseAgent() await agent.start(headless=False)
from anthropic import Anthropic
result = await agent.run_with_llm(
"Go to weather.com and find the weather for New York",
Anthropic()
)
print(result)
await agent.close()
asyncio.run(main())
Anti_patterns
- Using screenshots when DOM access works
- Not waiting for page loads
- Hardcoded selectors that break
- No error recovery for stale elements
User Confirmation Pattern
For sensitive actions, agents should pause and ask for human confirmation. "ChatGPT agent also pauses and asks for confirmation prior to taking sensitive steps such as completing a purchase."
Sensitivity levels:
- LOW: Navigation, reading (auto-approve)
- MEDIUM: Form filling, clicking (log, maybe confirm)
- HIGH: Purchases, authentication, file operations (always confirm)
- CRITICAL: Credential entry, financial transactions (confirm + review)
When to use: Actions with real-world consequences,Financial transactions,Authentication flows,File modifications
from enum import Enum from dataclasses import dataclass from typing import Callable, Optional import asyncio
class ActionSeverity(Enum): LOW = "low" # Auto-approve MEDIUM = "medium" # Log, optional confirm HIGH = "high" # Always confirm CRITICAL = "critical" # Confirm + review details
@dataclass class SensitiveAction: """Action that may need user confirmation.""" action_type: str description: str severity: ActionSeverity details: dict
class ConfirmationGate: """ Gate sensitive actions through user confirmation. """
# Action type -> severity mapping
ACTION_SEVERITY = {
# LOW - auto-approve
"navigate": ActionSeverity.LOW,
"scroll": ActionSeverity.LOW,
"read": ActionSeverity.LOW,
"screenshot": ActionSeverity.LOW,
# MEDIUM - log and maybe confirm
"click": ActionSeverity.MEDIUM,
"type": ActionSeverity.MEDIUM,
"search": ActionSeverity.MEDIUM,
# HIGH - always confirm
"download": ActionSeverity.HIGH,
"submit_form": ActionSeverity.HIGH,
"login": ActionSeverity.HIGH,
"file_write": ActionSeverity.HIGH,
# CRITICAL - confirm with full review
"purchase": ActionSeverity.CRITICAL,
"enter_password": ActionSeverity.CRITICAL,
"enter_credit_card": ActionSeverity.CRITICAL,
"send_money": ActionSeverity.CRITICAL,
"delete": ActionSeverity.CRITICAL,
}
def __init__(
self,
confirm_callback: Callable[[SensitiveAction], bool] = None,
auto_confirm_low: bool = True,
auto_confirm_medium: bool = False
):
self.confirm_callback = confirm_callback or self._default_confirm
self.auto_confirm_low = auto_confirm_low
self.auto_confirm_medium = auto_confirm_medium
self.action_log = []
def _default_confirm(self, action: SensitiveAction) -> bool:
"""Default confirmation via CLI prompt."""
print(f"\n{'='*60}")
print(f"ACTION CONFIRMATION REQUIRED")
print(f"{'='*60}")
print(f"Type: {action.action_type}")
print(f"Severity: {action.severity.value.upper()}")
print(f"Description: {action.description}")
print(f"Details: {action.details}")
print(f"{'='*60}")
while True:
response = input("Allow this action? [y/n]: ").lower().strip()
if response in ['y', 'yes']:
return True
elif response in ['n', 'no']:
return False
def classify_action(self, action_type: str, context: dict) -> ActionSeverity:
"""Classify action severity, considering context."""
base_severity = self.ACTION_SEVERITY.get(action_type, ActionSeverity.MEDIUM)
# Escalate based on context
if context.get("involves_credentials"):
return ActionSeverity.CRITICAL
if context.get("involves_money"):
return ActionSeverity.CRITICAL
if context.get("irreversible"):
return max(base_severity, ActionSeverity.HIGH, key=lambda x: x.value)
return base_severity
def check_action(
self,
action_type: str,
description: str,
details: dict = None
) -> tuple[bool, str]:
"""
Check if action should proceed.
Returns (approved, reason).
"""
details = details or {}
severity = self.classify_action(action_type, details)
action = SensitiveAction(
action_type=action_type,
description=description,
severity=severity,
details=details
)
# Log all actions
self.action_log.append({
"action": action,
"timestamp": __import__('datetime').datetime.now().isoformat()
})
# Auto-approve low severity
if severity == ActionSeverity.LOW and self.auto_confirm_low:
return True, "auto-approved (low severity)"
# Maybe auto-approve medium
if severity == ActionSeverity.MEDIUM and self.auto_confirm_medium:
return True, "auto-approved (medium severity)"
# Request confirmation
approved = self.confirm_callback(action)
if approved:
return True, "user approved"
else:
return False, "user rejected"
class ConfirmedComputerUseAgent: """ Computer use agent with confirmation gates. """
def __init__(self, base_agent, confirmation_gate: ConfirmationGate):
self.agent = base_agent
self.gate = confirmation_gate
def execute_action(self, action: dict) -> dict:
"""Execute action with confirmation check."""
action_type = action.get("type", "unknown")
# Build description
if action_type == "click":
desc = f"Click at ({action.get('x')}, {action.get('y')})"
elif action_type == "type":
text = action.get('text', '')
# Mask if looks like password
if self._looks_sensitive(text):
desc = f"Type sensitive text ({len(text)} chars)"
else:
desc = f"Type: {text[:50]}..."
else:
desc = f"Execute: {action_type}"
# Context for severity classification
context = {
"involves_credentials": self._looks_sensitive(action.get("text", "")),
"involves_money": self._mentions_money(action),
}
# Check with gate
approved, reason = self.gate.check_action(
action_type, desc, context
)
if not approved:
return {
"success": False,
"error": f"Action blocked: {reason}",
"action": action_type
}
# Execute if approved
return self.agent.execute_action(action)
def _looks_sensitive(self, text: str) -> bool:
"""Check if text looks like sensitive data."""
if not text:
return False
# Common patterns
patterns = [
r'\b\d{16}\b', # Credit card
r'\b\d{3,4}\b.*\b\d{3,4}\b', # CVV-like
r'password',
r'secret',
r'api.?key',
r'token'
]
import re
return any(re.search(p, text.lower()) for p in patterns)
def _mentions_money(self, action: dict) -> bool:
"""Check if action involves money."""
text = str(action)
money_patterns = [
r'\$\d+', r'pay', r'purchase', r'buy', r'checkout',
r'credit', r'debit', r'invoice', r'payment'
]
import re
return any(re.search(p, text.lower()) for p in money_patterns)
Usage
gate = ConfirmationGate( auto_confirm_low=True, auto_confirm_medium=False # Confirm clicks, typing )
agent = ConfirmedComputerUseAgent(base_agent, gate) result = agent.execute_action({"type": "click", "x": 500, "y": 300})
Anti_patterns
- Auto-approving all actions
- Not logging rejected actions
- Showing full passwords in confirmation
- No timeout on confirmation (hangs forever)
Action Logging Pattern
All computer use agent actions should be logged for:
- Debugging failed automations
- Security auditing
- Reproducibility
- Compliance requirements
Log format should capture:
- Timestamp
- Action type and parameters
- Screenshot before/after
- Success/failure status
- Model reasoning (if available)
When to use: Production computer use deployments,Debugging automation failures,Security-sensitive environments
from dataclasses import dataclass, field from datetime import datetime from typing import Optional, Any import json import os
@dataclass class ActionLogEntry: """Single action log entry.""" timestamp: datetime action_type: str parameters: dict success: bool error: Optional[str] = None screenshot_before: Optional[str] = None # Path to screenshot screenshot_after: Optional[str] = None model_reasoning: Optional[str] = None duration_ms: Optional[int] = None
def to_dict(self) -> dict:
return {
"timestamp": self.timestamp.isoformat(),
"action_type": self.action_type,
"parameters": self._sanitize_params(self.parameters),
"success": self.success,
"error": self.error,
"screenshot_before": self.screenshot_before,
"screenshot_after": self.screenshot_after,
"model_reasoning": self.model_reasoning,
"duration_ms": self.duration_ms
}
def _sanitize_params(self, params: dict) -> dict:
"""Remove sensitive data from params."""
sanitized = {}
sensitive_keys = ['password', 'secret', 'token', 'key', 'credit_card']
for k, v in params.items():
if any(s in k.lower() for s in sensitive_keys):
sanitized[k] = "[REDACTED]"
elif isinstance(v, str) and len(v) > 100:
sanitized[k] = v[:100] + "...[truncated]"
else:
sanitized[k] = v
return sanitized
@dataclass class TaskSession: """A complete task execution session.""" session_id: str task: str start_time: datetime end_time: Optional[datetime] = None actions: list[ActionLogEntry] = field(default_factory=list) success: bool = False final_result: Optional[str] = None
class ActionLogger: """ Comprehensive action logging for computer use agents. """
def __init__(self, log_dir: str = "./agent_logs"):
self.log_dir = log_dir
self.screenshot_dir = os.path.join(log_dir, "screenshots")
os.makedirs(self.screenshot_dir, exist_ok=True)
self.current_session: Optional[TaskSession] = None
def start_session(self, task: str) -> str:
"""Start a new task session."""
import uuid
session_id = str(uuid.uuid4())[:8]
self.current_session = TaskSession(
session_id=session_id,
task=task,
start_time=datetime.now()
)
return session_id
def log_action(
self,
action_type: str,
parameters: dict,
success: bool,
error: Optional[str] = None,
screenshot_before: bytes = None,
screenshot_after: bytes = None,
model_reasoning: str = None,
duration_ms: int = None
):
"""Log a single action."""
if not self.current_session:
raise RuntimeError("No active session")
# Save screenshots if provided
screenshot_paths = {}
timestamp_str = datetime.now().strftime("%Y%m%d_%H%M%S_%f")
if screenshot_before:
path = os.path.join(
s
…(truncated)