fix pyautogui resolution scaling issue, tool input get coordinate issue

main.py

@@ -11,6 +11,8 @@ ANTHROPIC_API_URL = "https://api.anthropic.com/v1/messages"
 MODEL = "claude-3-7-sonnet-20250219"
 BETA_FLAG = "computer-use-2025-01-24"
 TOOL_VERSION = "20250124"
+pyautogui.PAUSE = 0.1
+SCALING_FACTOR = 1.25
 
 HEADERS = {
     "content-type": "application/json",
@@ -48,8 +50,9 @@ def execute_computer_tool(tool_input):
         buffered = BytesIO()
         screenshot.save(buffered, format="PNG")
 
-        # Check size and resize if needed (5MB = 5242880 bytes)
+        MAX_BINARY_SIZE = 5242880 * 3 // 4
-        if buffered.tell() > 4242880:
+        img_data = None
+        if buffered.tell() > MAX_BINARY_SIZE:
             # Reset buffer
             buffered.seek(0)
             screenshot = Image.open(buffered)
@@ -58,14 +61,19 @@ def execute_computer_tool(tool_input):
             while True:
                 buffered = BytesIO()
                 new_size = (int(screenshot.width * 0.8), int(screenshot.height * 0.8))
+                print('new size:', new_size)
                 screenshot = screenshot.resize(new_size, Image.Resampling.LANCZOS)
                 screenshot.save(buffered, format="PNG", optimize=True)
-                if buffered.tell() <= 4242880:
+                buffered.seek(0)
+                img_data = buffered.read()
+                if len(img_data) <= MAX_BINARY_SIZE:
                     break
-         
+                buffered.seek(0)
+                screenshot = Image.open(buffered)
+        else:
+            img_data = buffered.getvalue()
         # Convert to base64
-        buffered.seek(0)
+        img_base64 = base64.b64encode(img_data).decode('utf-8')
-        img_base64 = base64.b64encode(buffered.read()).decode('utf-8')
         
         return {
             "type": "image",
@@ -76,10 +84,10 @@ def execute_computer_tool(tool_input):
             }
         }
     
-    elif action == "click":
+    elif action == "left_click":
         # Get coordinates
-        x = tool_input.get("coordinate_x")
+        x = tool_input.get("coordinate")[0] * SCALING_FACTOR
-        y = tool_input.get("coordinate_y")
+        y = tool_input.get("coordinate")[1] * SCALING_FACTOR
         if x is None or y is None:
             return {"type": "text", "text": "Error: Missing coordinates for click action"}
         
@@ -89,13 +97,13 @@ def execute_computer_tool(tool_input):
     
     elif action == "double_click":
         # Get coordinates
-        x = tool_input.get("coordinate_x")
+        x = tool_input.get("coordinate")[0] * SCALING_FACTOR
-        y = tool_input.get("coordinate_y")
+        y = tool_input.get("coordinate")[1] * SCALING_FACTOR
         if x is None or y is None:
             return {"type": "text", "text": "Error: Missing coordinates for double click action"}
         
         # Perform double click
-        pyautogui.doubleClick(x, y)
+        pyautogui.doubleClick(x, y, interval=0.2)
         return {"type": "text", "text": f"Double-clicked at coordinates ({x}, {y})"}
     
     elif action == "type":
@@ -231,8 +239,8 @@ def send_prompt(api_key, messages):
             {
                 "type": f"computer_{TOOL_VERSION}",
                 "name": "computer",
-                "display_width_px": 1024,
+                "display_width_px": 2880,
-                "display_height_px": 768,
+                "display_height_px": 1864,
                 "display_number": 1
             },
             {
@@ -315,6 +323,12 @@ def run():
             
             # Execute the tool
             result_content = execute_tool(tool_name, tool_input)
+            print('[TOOL RESULT]')
+            for key in result_content: 
+                if key == "source":
+                    print(f"{key}: {len(result_content[key])} chars long")
+                else:
+                    print(f"{key}: {result_content[key]}")
             
             # Add to results
             tool_results.append({