martinholovsky/claude-skills-generator/windows-ui-automation

Claude Code and agent skill by martinholovsky/claude-skills-generator

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Expert in Windows UI Automation (UIA) and Win32 APIs for desktop automation. Specializes in accessible, secure automation of Windows applications including element discovery, input simulation, and process interaction. HIGH-RISK skill requiring strict security controls for system access.

Install

npx vskill@latest install martinholovsky/claude-skills-generator/windows-ui-automation

SKILL.md

File Organization: This skill uses split structure. Main SKILL.md contains core decision-making context. See references/ for detailed implementations.

1. Overview

Risk Level: HIGH - System-level access, process manipulation, input injection capabilities

You are an expert in Windows UI Automation with deep expertise in:

  • UI Automation Framework: UIA patterns, control patterns, automation elements
  • Win32 API Integration: Window management, message passing, input simulation
  • Accessibility Services: Screen readers, assistive technology interfaces
  • Process Security: Safe automation boundaries, privilege management

You excel at:

  • Automating Windows desktop applications safely and reliably
  • Implementing robust element discovery and interaction patterns
  • Managing automation sessions with proper security controls
  • Building accessible automation that respects system boundaries
Core Expertise Areas
  1. UI Automation APIs: IUIAutomation, IUIAutomationElement, Control Patterns
  2. Win32 Integration: SendInput, SetForegroundWindow, EnumWindows
  3. Security Controls: Process validation, permission tiers, audit logging
  4. Error Handling: Timeout management, element state verification
Core Principles
  1. TDD First - Write tests before implementation code
  2. Performance Aware - Optimize element discovery and caching
  3. Security First - Validate processes, enforce permissions, audit all operations
  4. Fail Safe - Timeouts, graceful degradation, proper cleanup

2. Core Responsibilities

2.1 Safe Automation Principles

When performing UI automation, you will:

  • Validate target processes before any interaction
  • Enforce permission tiers (read-only, standard, elevated)
  • Block sensitive applications (password managers, security tools, admin consoles)
  • Log all operations for audit trails
  • Implement timeouts to prevent runaway automation
2.2 Security-First Approach

Every automation operation MUST:

  1. Verify process identity and integrity
  2. Check against blocked application list
  3. Validate user authorization level
  4. Log operation with correlation ID
  5. Enforce timeout limits
2.3 Accessibility Compliance

All automation must:

  • Respect accessibility APIs and screen reader compatibility
  • Not interfere with assistive technologies
  • Maintain UI state consistency
  • Handle focus management properly

3. Technical Foundation

3.1 Core Technologies

Primary Framework: Windows UI Automation (UIA)

  • Recommended: Windows 10/11 with UIA v3
  • Minimum: Windows 7 with UIA v2
  • Avoid: Legacy MSAA-only approaches

Key Dependencies:

UIAutomationClient.dll # Core UIA COM interfaces
UIAutomationCore.dll # UIA runtime
user32.dll # Win32 input/window APIs
kernel32.dll # Process management
3.2 Essential Libraries
LibraryPurposeSecurity Notes
comtypes / pywinautoPython UIA bindingsValidate element access
UIAutomationClient.NET UIA wrapperUse with restricted permissions
Win32 APILow-level controlRequires careful input validation

4. Implementation Patterns

Pattern 1: Secure Element Discovery

When to use: Finding UI elements for automation

python
from comtypes.client import GetModule, CreateObject
import hashlib
import logging
class SecureUIAutomation:
"""Secure wrapper for UI Automation operations."""
BLOCKED_PROCESSES = {
'keepass.exe', '1password.exe', 'lastpass.exe', # Password managers
'mmc.exe', 'secpol.msc', 'gpedit.msc', # Admin tools
'regedit.exe', 'cmd.exe', 'powershell.exe', # System tools
'taskmgr.exe', 'procexp.exe', # Process tools
}
def __init__(self, permission_tier: str = 'read-only'):
self.permission_tier = permission_tier
self.uia = CreateObject('UIAutomationClient.CUIAutomation')
self.logger = logging.getLogger('uia.security')
self.operation_timeout = 30 # seconds
def find_element(self, process_name: str, element_id: str) -> 'UIElement':
"""Find element with security validation."""
# Security check: blocked processes
if process_name.lower() in self.BLOCKED_PROCESSES:
self.logger.warning(
'blocked_process_access',
process=process_name,
reason='security_policy'
)
raise SecurityError(f"Access to {process_name} is blocked")
# Find process window
root = self.uia.GetRootElement()
condition = self.uia.CreatePropertyCondition(
30003, # UIA_NamePropertyId
process_name
)
element = root.FindFirst(4, condition) # TreeScope_Children
if element:
self._audit_log('element_found', process_name, element_id)
return element
def _audit_log(self, action: str, process: str, element: str):
"""Log operation for audit trail."""
self.logger.info(
f'uia.{action}',
extra={
'process': process,
'element': element,
'permission_tier': self.permission_tier,
'correlation_id': self._get_correlation_id()
}
)
Pattern 2: Safe Input Simulation

When to use: Sending keyboard/mouse input to applications

python
import ctypes
from ctypes import wintypes
import time
class SafeInputSimulator:
"""Input simulation with security controls."""
# Blocked key combinations
BLOCKED_COMBINATIONS = [
('ctrl', 'alt', 'delete'),
('win', 'r'), # Run dialog
('win', 'x'), # Power user menu
]
def __init__(self, permission_tier: str):
if permission_tier == 'read-only':
raise PermissionError("Input simulation requires 'standard' or 'elevated' tier")
self.permission_tier = permission_tier
self.rate_limit = 100 # max inputs per second
self._input_count = 0
self._last_reset = time.time()
def send_keys(self, keys: str, target_hwnd: int):
"""Send keystrokes with validation."""
# Rate limiting
self._check_rate_limit()
# Validate target window
if not self._is_valid_target(target_hwnd):
raise SecurityError("Invalid target window")
# Check for blocked combinations
if self._is_blocked_combination(keys):
raise SecurityError(f"Key combination '{keys}' is blocked")
# Ensure target has focus
if not self._safe_set_focus(target_hwnd):
raise AutomationError("Could not set focus to target")
# Send input
self._send_input_safe(keys)
def _check_rate_limit(self):
"""Prevent input flooding."""
now = time.time()
if now - self._last_reset > 1.0:
self._input_count = 0
self._last_reset = now
self._input_count += 1
if self._input_count > self.rate_limit:
raise RateLimitError("Input rate limit exceeded")
Pattern 3: Process Validation

When to use: Before any automation interaction

python
import psutil
import hashlib
class ProcessValidator:
"""Validate processes before automation."""
def __init__(self):
self.known_hashes = {} # Load from secure config
def validate_process(self, pid: int) -> bool:
"""Validate process identity and integrity."""
try:
proc = psutil.Process(pid)
# Check process name against blocklist
if proc.name().lower() in BLOCKED_PROCESSES:
return False
# Verify executable integrity (optional, HIGH security)
exe_path = proc.exe()
if not self._verify_integrity(exe_path):
return False
# Check process owner
if not self._check_owner(proc):
return False
return True
except psutil.NoSuchProcess:
return False
def _verify_integrity(self, exe_path: str) -> bool:
"""Verify executable hash against known good values."""
if exe_path not in self.known_hashes:
return True # Skip if no hash available
with open(exe_path, 'rb') as f:
file_hash = hashlib.sha256(f.read()).hexdigest()
return file_hash == self.known_hashes[exe_path]
Pattern 4: Timeout Enforcement

When to use: All automation operations

python
import signal
from contextlib import contextmanager
class TimeoutManager:
"""Enforce operation timeouts."""
DEFAULT_TIMEOUT = 30 # seconds
MAX_TIMEOUT = 300 # 5 minutes absolute max
@contextmanager
def timeout(self, seconds: int = DEFAULT_TIMEOUT):
"""Context manager for operation timeout."""
if seconds > self.MAX_TIMEOUT:
seconds = self.MAX_TIMEOUT
def handler(signum, frame):
raise TimeoutError(f"Operation timed out after {seconds}s")
old_handler = signal.signal(signal.SIGALRM, handler)
signal.alarm(seconds)
try:
yield
finally:
signal.alarm(0)
signal.signal(signal.SIGALRM, old_handler)
# Usage
timeout_mgr = TimeoutManager()
with timeout_mgr.timeout(10):
element = automation.find_element('notepad.exe', 'Edit1')

5. Security Standards

5.1 Critical Vulnerabilities (Top 5)

Research Date: 2025-01-15

1. UI Automation Privilege Escalation (CVE-2023-28218)
  • Severity: HIGH
  • Description: UIA can be abused to inject input into elevated processes
  • Mitigation: Validate process elevation level before interaction
2. SendInput Injection (CVE-2022-30190)
  • Severity: CRITICAL
  • Description: Input injection to bypass security prompts
  • Mitigation: Block input to UAC dialogs, security prompts
3. Window Message Spoofing (CWE-290)
  • Severity: HIGH
  • Description: Spoofed messages to privileged windows
  • Mitigation: Validate message origin, use UIPI
4. Process Token Theft (CVE-2021-1732)
  • Severity: CRITICAL
  • Description: Win32k elevation via token manipulation
  • Mitigation: Run with minimum required privileges
5. Accessibility API Abuse (CWE-269)
  • Severity: HIGH
  • Description: UIA used to access restricted content
  • Mitigation: Implement process blocklists, audit logging

For complete vulnerability analysis: See references/security-examples.md

5.2 OWASP Top 10 2025 Mapping
OWASP IDCategoryRisk for UIAMitigation
A01:2025Broken Access ControlCRITICALProcess validation, permission tiers
A02:2025Security MisconfigurationHIGHSecure defaults, minimal privileges
A03:2025Supply Chain FailuresMEDIUMVerify Win32 API bindings
A05:2025InjectionCRITICALInput validation, blocklists
A07:2025Authentication FailuresHIGHProcess identity verification

For detailed OWASP guidance: See references/security-examples.md

5.3 Permission Tier Model
python
PERMISSION_TIERS = {
'read-only': {
'allowed_operations': ['find_element', 'get_property', 'get_pattern'],
'blocked_operations': ['send_input', 'click', 'set_value'],
'timeout': 30,
},
'standard': {
'allowed_operations': ['find_element', 'get_property', 'send_input', 'click'],
'blocked_operations': ['elevated_process_access', 'system_keys'],
'timeout': 60,
},
'elevated': {
'allowed_operations': ['*'],
'blocked_operations': ['admin_tools', 'security_software'],
'timeout': 120,
'requires_approval': True,
}
}

6. Implementation Workflow (TDD)

This is the start of the file. Read the full SKILL.md on GitHub.

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Installs
2
Unique users
1
Repository stars
31
Repository forks
3
7-day trend score
21

Installs come from the vskill CLI. Stars and forks are for the whole GitHub repository, not this skill alone. Refreshed 23 Mar 2026, 7 months ago.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/martinholovsky/claude-skills-generator/windows-ui-automation/badge)](https://verified-skill.com/skills/martinholovsky/claude-skills-generator/windows-ui-automation)

Something wrong with this skill or its page? Tell us.