macos-control

Control macOS GUI applications via mouse automation, keyboard input, screenshots, image recognition, and AppleScript execution. Use when you need to interact with macOS app UIs, take screenshots, clic

SKILL.md

GitHub

macos-control

The sagui CLI provides mouse control, keyboard input, screenshot capture, image recognition, and screen queries for macOS automation.

Preflight: ensure sagui is installed

Before issuing any sagui command, run:

command -v sagui

If the command prints a path, you're ready — continue to the sections below.

If it prints nothing (exit code 1), sagui is not installed. Stop and ask the user for permission to install it before doing anything else. Show them these options and wait for confirmation:

Recommended — Homebrew:

brew install NakaokaRei/tap/sagui

Fallback — build from source (requires Xcode / Swift 6.2+ toolchain; verify with swift --version):

git clone https://github.com/NakaokaRei/SwiftAutoGUI.git /tmp/SwiftAutoGUI
cd /tmp/SwiftAutoGUI
swift build -c release
# Copy the binary somewhere on $PATH. /usr/local/bin requires sudo:
sudo cp .build/release/sagui /usr/local/bin/sagui
# Or, no sudo, into a user dir already on PATH:
# cp .build/release/sagui ~/.local/bin/sagui

After install, re-run command -v sagui to confirm the binary is reachable, then continue.

Permissions

The first time sagui issues an event or screenshot, macOS will prompt for these permissions. Without them, commands silently fail or return blank screenshots.

  • Accessibility — required for sagui key and sagui mouse. Enable in System Settings → Privacy & Security → Accessibility for the app running Claude Code (Terminal.app, iTerm, etc.).
  • Screen Recording — required for sagui screen screenshot, sagui screen pixel, and sagui screen locate*. Enable in System Settings → Privacy & Security → Screen Recording for the same app.

If a command unexpectedly fails or returns empty output, ask the user to verify both permissions are granted to their terminal app.

This skill is macOS-only — sagui builds on CoreGraphics. On other platforms, tell the user the skill cannot run and stop.

Coordinates

All coordinates use CGWindow coordinate system — logical points with origin at the top-left of the primary display.

  • x increases to the right
  • y increases downward
  • Values are in logical points (not pixels)

Use sagui screen size to get screen dimensions.

Commands

key — Simulate keyboard input

sagui key shortcut command c             # Keyboard shortcut (Cmd+C)
sagui key shortcut command shift a       # Multi-modifier shortcut
sagui key down shift                     # Press key without releasing
sagui key up shift                       # Release a held key
sagui key type "Hello, World!"           # Type text character by character
sagui key type "slow" --interval 0.1     # Type with delay between keystrokes
sagui key shortcut return                # Return key (alias for returnKey)
sagui key shortcut backspace             # Backspace (alias for delete)
sagui key list                           # List every valid key name
Subcommand Arguments Optional
shortcut <keys...> (key names)
down <key>
up <key>
type <text> --interval <seconds> (default: 0)
list

Supported key names

Modifiers: command, shift, rightShift, control, rightControl, option, rightOption, capsLock, function

Letters: a-z

Numbers: zero-nine

Function keys: f1-f20

Arrow keys: upArrow, downArrow, leftArrow, rightArrow

Navigation: home, end, pageUp, pageDown, help

Special: return, returnKey, enter, tab, space, escape, backspace, delete, forwardDelete

Keypad: keypad0-keypad9, keypadDecimal, keypadMultiply, keypadPlus, keypadMinus, keypadDivide, keypadEnter, keypadClear, keypadEquals

Media: volumeUp, volumeDown, mute, brightnessUp, brightnessDown

JIS: jisYen, jisUnderscore, jisEisu, jisKana

mouse — Control mouse cursor

Coordinates are screen points (origin top-left).

sagui mouse position                                                  # Print current position
sagui mouse move --x 500 --y 300                                      # Move to absolute position
sagui mouse move-relative --dx 50 --dy -30                            # Move relative to current
sagui mouse click                                                     # Left click at current position
sagui mouse click --right                                             # Right click
sagui mouse click --double                                            # Double-click
sagui mouse click --triple                                            # Triple-click
sagui mouse click --x 500 --y 300                                    # Click an exact position
sagui mouse click --right --x 500 --y 300                            # Right-click an exact position
sagui mouse drag --from-x 100 --from-y 100 --to-x 400 --to-y 400    # Drag
sagui mouse scroll --vertical 5                                       # Scroll up
sagui mouse scroll --vertical -3                                      # Scroll down
sagui mouse scroll --horizontal 2                                     # Scroll left
Subcommand Required Optional
position
move --x, --y
move-relative --dx, --dy
click --x <point> and --y <point>, --right, --double, --triple
drag --from-x, --from-y, --to-x, --to-y
scroll --vertical <clicks>, --horizontal <clicks>

screen — Screenshots and screen queries

sagui screen size                                          # Print screen dimensions
sagui screen screenshot                                    # Save screenshot to screenshot.png
sagui screen screenshot --output capture.png               # Save to specific path
sagui screen screenshot --region 0,0,500,500               # Capture specific region
sagui screen pixel --x 100 --y 200                         # Get pixel color (R G B A)
sagui screen locate button.png                             # Find image on screen (X Y W H)
sagui screen locate button.png --confidence 0.8            # Find with lower threshold
sagui screen locate-center button.png                      # Find image center (X Y)
Subcommand Required Optional
size
screenshot --output <path> (default: screenshot.png), --region <x,y,w,h>
pixel --x, --y
locate <image-path> --confidence <0.0-1.0>
locate-center <image-path> --confidence <0.0-1.0>

Image recognition uses Metal-accelerated normalized cross-correlation. Default confidence threshold is 0.95. Lower it for fuzzy matching.

agent — AI-powered automation

sagui agent "Open Safari and search for Swift"                    # Uses the on-device Apple model
sagui agent "Click the submit button" --provider openai            # Run with OpenAI
sagui agent "Fill in the form" --model gpt-5.6-sol --provider openai --max-iterations 10
sagui agent "Press the Save button" --vision-mode automatic       # Prefer semantic AX elements when available
Required Optional
<goal> (string) --api-key <key> (or OPENAI_API_KEY env), --model <model> (default: gpt-5.6-sol), `--provider <on-device

The command prints the selected provider at startup and a concise Reasoning: summary for every step. This summary explains the selected actions; it is not the model's hidden chain of thought.

Workflow

Screenshot + mouse/keyboard interaction

  1. sagui screen size — get screen dimensions.
  2. sagui screen screenshot --output /tmp/screen.png — capture current state.
  3. Read the screenshot image to identify target coordinates.
  4. sagui mouse click --x <x> --y <y> — interact at the identified position.
  5. sagui key type "text" or sagui key shortcut command a — keyboard input.
  6. sagui screen screenshot --output /tmp/verify.png — verify result.

Image-based interaction

  1. sagui screen locate-center target.png — find element position.
  2. sagui mouse click --x <x> --y <y> — click the found coordinates atomically.

Notes

  • All mouse and keyboard commands require Accessibility permissions.
  • Use sagui --version when reporting issues or checking a CI environment.
  • Screenshot and image recognition commands require Screen Recording permissions.
  • key type supports any Unicode character. Use key shortcut with modifier keys for keyboard shortcuts.
  • Scroll: positive vertical = up, negative = down; positive horizontal = left, negative = right.
  • Image recognition confidence defaults to 0.95; lower for fuzzy matching (e.g., 0.8).
  • Supported screenshot formats: PNG, JPG/JPEG, GIF, BMP, TIFF (detected from file extension).