Windows Control Skill

Full desktop automation for Windows. Control mouse, keyboard, and screen like a human user.

Quick Start

All scripts are in INLINECODE0

Screenshot

py screenshot.py > output.b64

Click

Type Text

py type_text.py "Hello World"

Press Keys

Move Mouse

py mouse_move.py 500 300

Scroll

Window Management (NEW!)

Advanced Actions (NEW!)

Read Window Text

py read_window.py "Notepad"           # Read all text from Notepad
py read_window.py "Visual Studio"     # Read text from VS Code
py read_window.py "Chrome"            # Read text from browser

Read UI Elements (NEW!)

py read_ui_elements.py "Chrome"               # All interactive elements
py read_ui_elements.py "Chrome" --buttons-only  # Just buttons
py read_ui_elements.py "Chrome" --links-only    # Just links
py read_ui_elements.py "Chrome" --json          # JSON output

Read Webpage Content (NEW!)

py read_webpage.py                     # Read active browser
py read_webpage.py "Chrome"            # Target Chrome specifically
py read_webpage.py "Chrome" --buttons  # Include buttons
py read_webpage.py "Chrome" --links    # Include links with coords
py read_webpage.py "Chrome" --full     # All elements (inputs, images)
py read_webpage.py "Chrome" --json     # JSON output

Handle Dialogs (NEW!)

# List all open dialogs
py handle_dialog.py list

# Read current dialog content
py handle_dialog.py read
py handle_dialog.py read --json

# Click button in dialog
py handle_dialog.py click "OK"
py handle_dialog.py click "Save"
py handle_dialog.py click "Yes"

# Type into dialog text field
py handle_dialog.py type "myfile.txt"
py handle_dialog.py type "C:\path\to\file" --field 0

# Dismiss dialog (auto-finds OK/Close/Cancel)
py handle_dialog.py dismiss

# Wait for dialog to appear
py handle_dialog.py wait --timeout 10
py handle_dialog.py wait "Save As" --timeout 5

Click Element by Name (NEW!)

py click_element.py "Save"                    # Click "Save" anywhere
py click_element.py "OK" --window "Notepad"   # In specific window
py click_element.py "Submit" --type Button    # Only buttons
py click_element.py "File" --type MenuItem    # Menu items
py click_element.py --list                    # List clickable elements
py click_element.py --list --window "Chrome"  # List in specific window

Read Screen Region (OCR - Optional)

py read_region.py 100 100 500 300     # Read text from coordinates

Workflow Pattern

1. 1. Read window - Extract text from specific window (fast, accurate)
2. Read UI elements - Get buttons, links with coordinates
3. Screenshot (if needed) - See visual layout
4. Act - Click element by name or coordinates
5. Handle dialogs - Interact with popups/save dialogs
6. Read window - Verify changes

Screen Coordinates

• - Origin (0, 0) is top-left corner
• Your screen: 2560x1440 (check with screenshot)
• Use coordinates from screenshot analysis

Examples

Open Notepad and type

Click in VS Code

Monitor Notepad changes

Text Reading Methods

Method 1: Windows UI Automation (BEST)

• - Use read_window.py for any window
• Use read_ui_elements.py for buttons/links with coordinates
• Use read_webpage.py for browser content with structure
• Gets actual text data (not image-based)

Method 2: Click by Name (NEW)

• - Use click_element.py to click buttons/links by name
• No coordinates needed - finds elements automatically
• Works across all windows or target specific window

Method 3: Dialog Handling (NEW)

• - Use handle_dialog.py for popups, save dialogs, alerts
• Read dialog content, click buttons, type text
• Auto-dismiss with common buttons (OK, Cancel, etc.)

Method 4: Screenshot + Vision (Fallback)

• - Take full screenshot
• AI reads text visually
• Slower but works for any content

Method 5: OCR (Optional)

• - Use read_region.py with Tesseract
• Requires additional installation
• Good for images/PDFs with text

Safety Features

• - pyautogui.FAILSAFE = True (move mouse to top-left to abort)
• Small delays between actions
• Smooth mouse movements (not instant jumps)

Requirements

• - Python 3.11+
• pyautogui (installed ✅)
• pillow (installed ✅)

Tips

• - Always screenshot first to see current state
• Coordinates are absolute (not relative to windows)
• Wait briefly after clicks for UI to update
• Use ctrl+z friendly actions when possible