Add input handling functionality to the agent. Implement API endpoints for mouse click and text input, and update the web interface to support video capture and interaction features.

lib/screenshot/screenshot.go

@@ -13,6 +13,8 @@ import (
 	"tea.chunkbyte.com/kato/go-worm/lib/models"
 )
 
+var ErrMonitorNotFound = errors.New("monitor not found")
+
 var (
 	user32                  = syscall.NewLazyDLL("user32.dll")
 	gdi32                   = syscall.NewLazyDLL("gdi32.dll")
@@ -68,19 +70,80 @@ func Capture(format string, quality int) ([]models.CapturedImage, error) {
 	}
 	result := make([]models.CapturedImage, 0, len(monitors))
 	for _, monitor := range monitors {
-		img, err := captureRect(monitor)
+		frame, err := captureAndEncode(monitor, format, quality)
 		if err != nil {
 			return nil, err
 		}
-		data, contentType, err := encodeImage(img, format, quality)
-		if err != nil {
-			return nil, err
-		}
-		result = append(result, models.CapturedImage{ContentType: contentType, Data: data})
+		result = append(result, frame)
 	}
 	return result, nil
 }
 
+func CaptureMonitor(index int, format string, quality int) (models.CapturedImage, error) {
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+	if err := attachInputDesktop(); err != nil {
+		return models.CapturedImage{}, err
+	}
+	monitors, err := enumerateMonitors()
+	if err != nil {
+		return models.CapturedImage{}, err
+	}
+	if index < 0 || index >= len(monitors) {
+		return models.CapturedImage{}, ErrMonitorNotFound
+	}
+	return captureAndEncode(monitors[index], format, quality)
+}
+
+func MonitorCount() (int, error) {
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+	if err := attachInputDesktop(); err != nil {
+		return 0, err
+	}
+	monitors, err := enumerateMonitors()
+	if err != nil {
+		return 0, err
+	}
+	return len(monitors), nil
+}
+
+func MonitorBounds(index int) (left, top, width, height int, err error) {
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+	if err := attachInputDesktop(); err != nil {
+		return 0, 0, 0, 0, err
+	}
+	monitors, err := enumerateMonitors()
+	if err != nil {
+		return 0, 0, 0, 0, err
+	}
+	if index < 0 || index >= len(monitors) {
+		return 0, 0, 0, 0, ErrMonitorNotFound
+	}
+	r := monitors[index]
+	return int(r.Left), int(r.Top), int(r.Right - r.Left), int(r.Bottom - r.Top), nil
+}
+
+func captureAndEncode(monitor rect, format string, quality int) (models.CapturedImage, error) {
+	img, err := captureRect(monitor)
+	if err != nil {
+		return models.CapturedImage{}, err
+	}
+	data, contentType, err := encodeImage(img, format, quality)
+	if err != nil {
+		return models.CapturedImage{}, err
+	}
+	return models.CapturedImage{
+		ContentType: contentType,
+		Data:        data,
+		Left:        int(monitor.Left),
+		Top:         int(monitor.Top),
+		Width:       int(monitor.Right - monitor.Left),
+		Height:      int(monitor.Bottom - monitor.Top),
+	}, nil
+}
+
 func attachInputDesktop() error {
 	h, _, err := procOpenInputDesktop.Call(0, 0, 0x0001|0x0040)
 	if h == 0 {