Desktop Control

桌面控制技能

此技能通过 PyAutoGUI 提供全面的桌面自动化功能，让 AI Agent 能够控制鼠标和键盘、截取屏幕截图，并与桌面环境交互。

如何使用此技能

作为 AI Agent，你可以使用 uvx desktop-agent CLI 调用桌面自动化命令。

命令结构

所有命令均遵循以下模式：

uvx desktop-agent <category> <command> [arguments] [options]

类别：

• mouse - 鼠标控制
• keyboard - 键盘输入
• screen - 截图和屏幕分析
• message - 用户对话框
• app - 应用程序控制（打开、聚焦、列出窗口）

可用命令

🖱️ 鼠标控制（mouse）

控制光标移动和点击。

# Move cursor to coordinates
uvx desktop-agent mouse move <x> <y> [--duration SECONDS]

# Click at current position or specific coordinates
uvx desktop-agent mouse click [x] [y] [--button left|right|middle] [--clicks N]

# Specialized clicks
uvx desktop-agent mouse double-click [x] [y]
uvx desktop-agent mouse right-click [x] [y]
uvx desktop-agent mouse middle-click [x] [y]

# Drag to coordinates
uvx desktop-agent mouse drag <x> <y> [--duration SECONDS] [--button BUTTON]

# Scroll (positive=up, negative=down)
uvx desktop-agent mouse scroll <clicks> [x] [y]

# Get current mouse position
uvx desktop-agent mouse position

示例：

# Move to center of 1920x1080 screen
uvx desktop-agent mouse move 960 540 --duration 0.5

# Right-click at specific location
uvx desktop-agent mouse right-click 500 300

# Scroll down 5 clicks
uvx desktop-agent mouse scroll -5

⌨️ 键盘控制（keyboard）

输入文本并执行键盘快捷键。

# Type text
uvx desktop-agent keyboard write "<text>" [--interval SECONDS]

# Press keys
uvx desktop-agent keyboard press <key> [--presses N] [--interval SECONDS]

# Execute hotkey combination (comma-separated)
uvx desktop-agent keyboard hotkey "<key1>,<key2>,..."

# Hold/release keys
uvx desktop-agent keyboard keydown <key>
uvx desktop-agent keyboard keyup <key>

示例：

# Type text with natural delay
uvx desktop-agent keyboard write "Hello World" --interval 0.05

# Copy selected text
uvx desktop-agent keyboard hotkey "ctrl,c"

# Open Task Manager
uvx desktop-agent keyboard hotkey "ctrl,shift,esc"

# Press Enter 3 times
uvx desktop-agent keyboard press enter --presses 3

常见按键名称：

• 修饰键：ctrl、shift、alt、win
• 特殊键：enter、tab、esc、space、backspace、delete
• 功能键：f1 至 f12
• 方向键：up、down、left、right

🖼️ 屏幕与截图（screen）

截取屏幕截图并分析屏幕内容。支持指定特定窗口。

# Take screenshot
uvx desktop-agent screen screenshot <filename> [--region "x,y,width,height"] [--window <title>] [--active]

# Locate image on screen or within window
uvx desktop-agent screen locate <image_path> [--confidence 0.0-1.0] [--window <title>] [--active]
uvx desktop-agent screen locate-center <image_path> [--confidence 0.0-1.0] [--window <title>] [--active]

# Locate text using OCR within window
uvx desktop-agent screen locate-text-coordinates <text> [--window <title>] [--active]
uvx desktop-agent screen read-all-text [--window <title>] [--active]

# Utility commands
uvx desktop-agent screen pixel <x> <y>
uvx desktop-agent screen size
uvx desktop-agent screen on-screen <x> <y>

示例：

# Screenshot of active window
uvx desktop-agent screen screenshot active.png --active

# Screenshot of a specific application
uvx desktop-agent screen screenshot chrome.png --window "Google Chrome"

# Locate image within Notepad
uvx desktop-agent screen locate-center button.png --window "Notepad"

💬 消息对话框（message）

显示用户交互对话框。

# Show alert
uvx desktop-agent message alert "<text>" [--title TITLE] [--button BUTTON]

# Show confirmation dialog
uvx desktop-agent message confirm "<text>" [--title TITLE] [--buttons "OK,Cancel"]

# Prompt for input
uvx desktop-agent message prompt "<text>" [--title TITLE] [--default TEXT]

# Password input
uvx desktop-agent message password "<text>" [--title TITLE] [--mask CHAR]

示例：

# Simple alert
uvx desktop-agent message alert "Task completed!"

# Get user confirmation
uvx desktop-agent message confirm "Continue with operation?"

# Ask for user input
uvx desktop-agent message prompt "Enter your name:"

📱 应用程序控制（app）

在 Windows、macOS 和 Linux 上控制应用程序。

# Open an application by name
uvx desktop-agent app open <name> [--arg ARGS...]

# Focus on a window by title/name
uvx desktop-agent app focus <name>

# List all visible windows
uvx desktop-agent app list

示例：

# Windows: Open Notepad
uvx desktop-agent app open notepad

# Windows: Open Chrome with a URL
uvx desktop-agent app open "chrome" --arg "https://google.com"

# macOS: Open Safari
uvx desktop-agent app open "Safari"

# Focus on a specific window
uvx desktop-agent app focus "Untitled - Notepad"

# List all open windows
uvx desktop-agent app list

常见自动化工作流

工作流 1：打开应用程序并输入文本

# Open notepad directly (cross-platform)
uvx desktop-agent app open notepad

# Wait for app to open, then focus it
uvx desktop-agent app focus notepad

# Type some text
uvx desktop-agent keyboard write "Hello from Desktop Skill!"

工作流 2：截图 + 分析

# Get screen size first
uvx desktop-agent screen size

# Take full screenshot
uvx desktop-agent screen screenshot current_screen.png

# Check if specific UI element is visible
uvx desktop-agent screen locate save_button.png

工作流 3：填写表单

# Click first field
uvx desktop-agent mouse click 300 200

# Fill field
uvx desktop-agent keyboard write "John Doe"

# Tab to next field
uvx desktop-agent keyboard press tab

# Fill second field
uvx desktop-agent keyboard write "[email protected]"

# Submit form (Enter)
uvx desktop-agent keyboard press enter

工作流 4：复制/粘贴操作

# Select all text
uvx desktop-agent keyboard hotkey "ctrl,a"

# Copy
uvx desktop-agent keyboard hotkey "ctrl,c"

# Click destination
uvx desktop-agent mouse click 500 600

# Paste
uvx desktop-agent keyboard hotkey "ctrl,v"

安全注意事项

使用此技能时，AI Agent 应当：

1. 确认坐标：点击前使用 screen size 和 on-screen
2. 添加延迟：在命令之间加入适当的延迟，确保 UI 响应
3. 验证图像：使用 locate 命令前，确保图像文件存在
4. 处理失败情况：窗口变化或元素移动时，命令可能会失败
5. 用户安全：始终通过 message confirm 向用户确认破坏性操作

故障排除

PyAutoGUI 故障保护

PyAutoGUI 具有故障保护机制：将鼠标移至屏幕角落会中止操作。这是一项安全功能。

未找到图像

使用 screen locate 时，请确保：

• 图像文件存在且路径正确
• 调整 --confidence（尝试 0.7-0.9）
• 图像与屏幕上的实际显示完全一致（分辨率、颜色）

获取帮助

# Show all available commands
uvx desktop-agent --help

# Show commands for specific category
uvx desktop-agent mouse --help
uvx desktop-agent keyboard --help
uvx desktop-agent screen --help
uvx desktop-agent message --help

# Show help for specific command
uvx desktop-agent mouse move --help

面向 AI Agent 的集成技巧

1. 使用绝对坐标时，务必先检查屏幕尺寸
2. 尽可能使用相对定位（例如，获取当前位置并计算偏移量）
3. 组合命令以完成复杂工作流
4. 执行前进行验证（例如，检查图像是否显示在屏幕上）
5. 通过消息对话框向用户反馈重要操作
6. 妥善处理错误——UI 状态变化时，命令可能会失败

性能说明

• 使用 --duration 的鼠标移动会以动画方式进行，因此需要时间
• 在大屏幕上定位图像（locate）可能较慢，尽可能使用区域
• 键盘命令通常执行较快（< 100ms）
• 截图效果取决于屏幕分辨率和区域大小

输出格式

默认情况下，所有命令都会输出结构化 JSON，非常适合 AI Agent 以编程方式使用：

uvx desktop-agent mouse position
# Output: {"success": true, "command": "mouse.position", "timestamp": "2026-01-31T10:00:00Z", "duration_ms": 5, "data": {"position": {"x": 960, "y": 540}}}

响应架构

所有 JSON 响应均遵循以下架构：

{
  "success": true,
  "command": "category.command",
  "timestamp": "2026-01-31T10:00:00Z",
  "duration_ms": 150,
  "data": { ... },
  "error": null
}

错误响应架构

{
  "success": false,
  "command": "category.command",
  "timestamp": "2026-01-31T10:00:00Z",
  "duration_ms": 50,
  "data": null,
  "error": {
    "code": "image_not_found",
    "message": "Image file 'button.png' not found",
    "details": {},
    "recoverable": true
  }
}

错误代码

代码	描述
success	命令执行成功
invalid_argument	命令参数无效
coordinates_out_of_bounds	坐标超出屏幕范围
image_not_found	未找到图像文件或屏幕上不存在该图像
window_not_found	未找到目标窗口
ocr_failed	OCR 操作失败
application_not_found	未找到应用程序
permission_denied	权限不足
platform_not_supported	不支持该平台
timeout	操作超时
unknown_error	未知错误

鼠标移动：

uvx desktop-agent mouse move 960 540

{"success": true, "command": "mouse.move", "timestamp": "...", "duration_ms": 150, "data": {"x": 960, "y": 540, "duration": 0}, "error": null}

屏幕尺寸：

uvx desktop-agent screen size

{"success": true, "command": "screen.size", "timestamp": "...", "duration_ms": 5, "data": {"size": {"width": 1920, "height": 1080}}, "error": null}

查找图像：

uvx desktop-agent screen locate button.png

{"success": true, "command": "screen.locate", "timestamp": "...", "duration_ms": 250, "data": {"image_found": true, "bounding_box": {"left": 100, "top": 200, "width": 50, "height": 30, "center_x": 125, "center_y": 215}}, "error": null}

列出窗口：

uvx desktop-agent app list

{"success": true, "command": "app.list", "timestamp": "...", "duration_ms": 100, "data": {"windows": ["Untitled - Notepad", "Google Chrome", "Visual Studio Code"]}, "error": null}

错误示例：

uvx desktop-agent screen locate missing.png

{"success": false, "command": "screen.locate", "timestamp": "...", "duration_ms": 50, "data": null, "error": {"code": "image_not_found", "message": "Image file 'missing.png' not found", "details": {}, "recoverable": true}}

AI Agent 高效使用指南

本节将介绍 AI Agent 如何通过优化的命令序列和最佳实践，有效使用此技能。

🎯 核心策略：先观察，再操作

执行操作前务必了解当前状态。这样可以避免点击错误的坐标或在错误的窗口中输入内容。

推荐的初始操作序列：

# 1. Get screen dimensions to understand your workspace
uvx desktop-agent screen size
uvx desktop-agent app list
uvx desktop-agent mouse position

📋 按任务划分的推荐命令序列

打开并与应用程序交互

# ✅ CORRECT: Open, wait, verify, then interact
uvx desktop-agent app open notepad              # Step 1: Open app
uvx desktop-agent app list
uvx desktop-agent app focus "Notepad"
uvx desktop-agent keyboard write "Hello World"  # Step 4: Now safe to type

# ❌ WRONG: Type immediately without verification
uvx desktop-agent app open notepad
uvx desktop-agent keyboard write "Hello World"  # May type in wrong window!

查找并点击 UI 元素（基于图像）

# ✅ CORRECT: Locate first, click if found
uvx desktop-agent screen locate-center button.png --confidence 0.8
# Check if success=true and coordinates are valid
uvx desktop-agent mouse click 125 215  # Use returned coordinates

# ❌ WRONG: Click without verifying element exists
uvx desktop-agent mouse click 125 215  # Might click wrong area!

查找并点击 UI 元素（基于 OCR 文本）

# ✅ CORRECT: Read screen text, then locate specific text
uvx desktop-agent screen read-all-text --active
uvx desktop-agent screen locate-text-coordinates "Save" --active
# Use returned coordinates to click

# For window-specific OCR:
uvx desktop-agent screen locate-text-coordinates "OK" --window "Dialog Title"

填写包含多个字段的表单

# ✅ CORRECT: Click each field explicitly before typing
uvx desktop-agent mouse click 300 200           # Click first field
uvx desktop-agent keyboard write "John Doe"
uvx desktop-agent mouse click 300 250           # Click second field (more reliable)
uvx desktop-agent keyboard write "[email protected]"
uvx desktop-agent mouse click 300 300           # Click third field
uvx desktop-agent keyboard write "555-1234"

# OR use Tab navigation (less reliable if field order changes)
uvx desktop-agent mouse click 300 200
uvx desktop-agent keyboard write "John Doe"
uvx desktop-agent keyboard press tab
uvx desktop-agent keyboard write "[email protected]"
uvx desktop-agent keyboard press tab
uvx desktop-agent keyboard write "555-1234"
uvx desktop-agent keyboard press enter          # Submit

截取指定区域的屏幕截图以供分析

# ✅ CORRECT: Screenshot specific windows for faster processing
uvx desktop-agent app list --json                           # Find exact window title
uvx desktop-agent screen screenshot app.png --window "Google Chrome"

# For active window only
uvx desktop-agent screen screenshot active.png --active

# Full screen only when necessary (slower, larger file)
uvx desktop-agent screen size
uvx desktop-agent screen screenshot full.png

安全地拖放

# ✅ CORRECT: Move to start, verify position, then drag
uvx desktop-agent mouse move 100 200                 # Move to source
uvx desktop-agent mouse position              # Verify position
uvx desktop-agent mouse drag 500 400 --duration 0.5  # Drag to destination

# For precision, use slower duration
uvx desktop-agent mouse drag 500 400 --duration 1.0

🔄 错误恢复模式

未找到窗口时

# Pattern: List windows, find closest match, retry
uvx desktop-agent app focus "Chrome"             # Fails with window_not_found
uvx desktop-agent app list                # See actual window titles
# Output shows: "Google Chrome - My Page"
uvx desktop-agent app focus "Google Chrome"      # Use correct title

未找到图像时

# Pattern: Adjust confidence or take new screenshot
uvx desktop-agent screen locate button.png --confidence 0.9
uvx desktop-agent screen locate button.png --confidence 0.7
# If still failing, capture current state for analysis
uvx desktop-agent screen screenshot current.png --active

点击似乎未命中时

# Pattern: Verify coordinates are on screen
uvx desktop-agent screen size             # Get screen bounds
uvx desktop-agent screen on-screen 1500 900      # Check if coords are valid
uvx desktop-agent mouse move 1500 900            # Move first to visualize
uvx desktop-agent mouse click                    # Then click at current position

⚡ 性能优化

尽量减少屏幕截图

# ✅ GOOD: Screenshot only the region you need
uvx desktop-agent screen screenshot button_area.png --region "100,200,200,100"

# ✅ GOOD: Screenshot specific window instead of full screen  
uvx desktop-agent screen screenshot chrome.png --window "Google Chrome"

# ❌ SLOW: Full screen capture when you only need a small area
uvx desktop-agent screen screenshot full.png

批量输入键盘内容

# ✅ FASTER: Write entire text at once
uvx desktop-agent keyboard write "This is a complete sentence with all the text."

# ❌ SLOWER: Multiple write commands
uvx desktop-agent keyboard write "This is "
uvx desktop-agent keyboard write "a complete "
uvx desktop-agent keyboard write "sentence."

尽可能使用快捷键代替鼠标

# ✅ FASTER: Use keyboard shortcuts
uvx desktop-agent keyboard hotkey "ctrl,s"       # Save
uvx desktop-agent keyboard hotkey "ctrl,a"       # Select all
uvx desktop-agent keyboard hotkey "ctrl,shift,s" # Save as

# ❌ SLOWER: Navigate menu with mouse
uvx desktop-agent mouse click 50 30              # Click File menu
uvx desktop-agent mouse click 60 80              # Click Save option

🛡️ 防御式编程模式

始终验证关键操作

# Before destructive action, confirm with user
uvx desktop-agent message confirm "This will delete all files. Continue?" --title "Warning"
# Check output: if "Cancel" was clicked, abort operation

使用 JSON 模式以可靠地解析

# ✅ RELIABLE: Parse structured JSON output
uvx desktop-agent screen locate button.png
# Parse: {"success": true, "data": {"center_x": 125, "center_y": 215}}

# ❌ FRAGILE: Parse text output
uvx desktop-agent screen locate button.png
# Parse: "Found at: Box(left=100, top=200, width=50, height=30)"

多步骤操作前进行验证

# Multi-step file operation with validation
uvx desktop-agent app list
uvx desktop-agent screen locate-text-coordinates "File" --active
uvx desktop-agent mouse click <returned_x> <returned_y>
uvx desktop-agent screen locate-text-coordinates "Save As" --active
uvx desktop-agent mouse click <returned_x> <returned_y>

🎮 平台特定注意事项

Windows

# Common Windows shortcuts
uvx desktop-agent keyboard hotkey "win,d"        # Show desktop
uvx desktop-agent keyboard hotkey "win,e"        # Open Explorer
uvx desktop-agent keyboard hotkey "alt,tab"      # Switch windows
uvx desktop-agent keyboard hotkey "win,r"        # Run dialog

# Open apps by name
uvx desktop-agent app open notepad
uvx desktop-agent app open calc
uvx desktop-agent app open mspaint

macOS

# Common macOS shortcuts (use 'command' for Cmd key)
uvx desktop-agent keyboard hotkey "command,space"   # Spotlight
uvx desktop-agent keyboard hotkey "command,tab"     # App switcher
uvx desktop-agent keyboard hotkey "command,q"       # Quit app
uvx desktop-agent keyboard hotkey "command,shift,3" # Screenshot

# Open apps
uvx desktop-agent app open "Safari"
uvx desktop-agent app open "TextEdit"

Linux

# Open apps (uses xdg-open or direct command)
uvx desktop-agent app open firefox
uvx desktop-agent app open gedit

# Common shortcuts may vary by DE
uvx desktop-agent keyboard hotkey "alt,f2"       # Run dialog (many DEs)

📊 决策树：选择合适的命令

 Want to interact with an app?
├── App not running → `app open <name>`
├── App running but not focused → `app focus <name>` 
└── Need to verify windows → `app list`

Want to find a UI element?
├── Have reference image → `screen locate-center <image>`
├── Know the text label → `screen locate-text-coordinates "<text>"`
└── Need to see all text → `screen read-all-text --active`

Want to click something?
├── Know exact coordinates → `mouse click <x> <y>`
├── Need to find first → Use locate commands above, then click returned coords
└── Not sure if on screen → `screen on-screen <x> <y>` first

Want to type something?
├── Regular text → `keyboard write "<text>"`
├── Keyboard shortcut → `keyboard hotkey "<key1>,<key2>"`
├── Single key press → `keyboard press <key>`
└── Multiple of same key → `keyboard press <key> --presses N`

面向 AI Agent 的集成技巧

1. 使用绝对坐标时，务必先检查屏幕尺寸
2. 尽可能使用相对定位（例如，获取当前位置并计算偏移量）
3. 组合命令以处理复杂工作流
4. 执行前先验证（例如，检查屏幕上是否存在图像）
5. 使用消息对话框为重要操作向用户提供反馈
6. 妥善处理错误——UI 状态发生变化时，命令可能会失败