from rapidfuzz import fuzz import json import os import easyocr import requests from config.config import Config configs = Config.get_instance().config_data def get_trajectory_path(): trajectory_id = configs['TRAJECTORY_ID'] log_root = configs["LOG_ROOT"] trajectory_path = f"{log_root}{trajectory_id}/" return trajectory_path def parse_actions(): trajectory_path = get_trajectory_path() actions = [] # Traverse all files ending in 'response.json' in the trajectory path. for filename in os.listdir(trajectory_path): if filename.endswith("response.json"): file_path = os.path.join(trajectory_path, filename) with open(file_path, 'r') as f: data = json.load(f) actions.append({"ControlText": data['response']['ControlText'], "Function": data['response']['Function'], "Args": data['response']['Args']}) # print(data['response']['ControlText']) # File Explorer # print(data['response']['Function']) # click_input # print(data['response']['Args']) # {'button': 'left', 'double': False} return actions def fuzzy_match(str1, str2, threshold=60): similarity = fuzz.ratio(str1, str2) if similarity >= threshold: return True return False def fuzzy_in(str1, str2, threshold=60): len_str1 = len(str1) for i in range(len(str2) - len_str1 + 1): substring = str2[i:i + len_str1] similarity = fuzz.ratio(str1, substring) if similarity >= threshold: return True return False def check_mouse_clicks(text: str) -> bool: """Checks if the mouse has clicked on the specified text. Parameters --------- text: str The text associated with the click. Returns --------- bool True if the mouse has clicked on the specified text, False otherwise. Examples --------- >>> # Evaluate if the agent has successfully set the picture 'envelope.png' as background >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion() -> EvalResult: >>> if not check_mouse_clicks(text='envelope.png'): >>> return EvalResult(False, "Subtask execution failed because did not click on 'envelope.png'.") >>> if not check_mouse_clicks(text='set as background'): >>> return EvalResult(False, "Subtask execution failed because did not click on 'set as background'.") >>> return EvalResult(True, "Subtask completed successfully") """ actions = parse_actions() for action in actions: if (action['Function'] == "click_input" and fuzzy_match(text.upper(), action['ControlText'].upper())): return True return False def check_keyboard_types(text: str) -> bool: """Checks if the keyboard has typed the specified text. Parameters --------- text: str The text to be typed. Returns --------- bool True if the keyboard has typed the specified text, False otherwise. Examples --------- >>> # Evaluate if the agent has successfully typed 'Hello, World!' >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion() -> EvalResult: >>> if not check_keyboard_types(text='Hello, World!'): >>> return EvalResult(False, "Subtask execution failed because text 'Hello, World!' was not typed.") >>> return EvalResult(True, "Subtask completed successfully") """ actions = parse_actions() for action in actions: if (action['Function'] == "keyboard_input" and fuzzy_match(text.upper(), action['Args']['keys'].upper())): return True return False def check_file_exists(file_path: str) -> bool: """Checks if the specified file exists. Parameters --------- file_path: str The path to the file to be checked. Returns --------- bool True if the file exists, False otherwise. Examples --------- >>> # Evaluate if the agent has successfully renamed 'cat.jpg' to 'cute cat.jpg' >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion() -> EvalResult: >>> # Check if original file no longer exists >>> if check_file_exists(file_path='C:/Users/user/Desktop/images/cat.jpg'): >>> return EvalResult(False, "Subtask execution failed because original file 'cat.jpg' still exists.") >>> # Check if new file exists >>> if not check_file_exists(file_path='C:/Users/user/Desktop/images/cute cat.jpg'): >>> return EvalResult(False, "Subtask execution failed because renamed file 'cute cat.jpg' was not found.") >>> # All checks passed, task completed >>> return EvalResult(True, "Subtask completed successfully") """ IP = configs['VM_IP'] command = f"IF EXIST \"{file_path}\" (echo True) ELSE (echo False)" payload = { "shell": True, "command": command } response = requests.post(f"http://{IP}:5000/execute", json=payload) response = response.json() assert response['status'] == 'success' return eval(response["output"]) def check_text_exists_via_ocr(text: str) -> bool: """Checks if the specified text is present in the last screenshot using OCR (Optical Character Recognition). Parameters --------- text: str The text to be checked. Returns --------- bool True if the text is present in the last screenshot, False otherwise. Examples --------- >>> # Evaluate if the agent has successfully set the clock to '9:00 AM' >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion() -> EvalResult: >>> # Check if the time exists in the window >>> if not check_text_exists_via_ocr(text='9:00 AM'): >>> return EvalResult(False, "Subtask execution failed because time '9:00 AM' was not found.") >>> # All checks passed, task completed >>> return EvalResult(True, "Subtask completed successfully") """ actions = parse_actions() step_num = len(actions) trajectory_path = get_trajectory_path() last_screenshot_path = os.path.join(trajectory_path, f"action_step{step_num-1}_raw.png") reader = easyocr.Reader(['en']) ocr_result = reader.readtext(last_screenshot_path, detail=0) ocr_result = ' '.join(ocr_result) return fuzzy_in(text.upper(), ocr_result.upper()) def check_text_exists_via_control(text: str) -> bool: """Checks if the specified text is present in the last screenshot through control information. Parameters --------- text: str The text to be checked. Returns --------- bool True if the text is present in the last screenshot, False otherwise. Examples --------- >>> # Evaluate if the agent has successfully input the code 'print("Hello World!")' >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion() -> EvalResult: >>> # Check if the code exists in the window >>> if not check_text_exists_via_control(text='print("Hello World!")'): >>> return EvalResult(False, "Subtask execution failed because code 'print(\"Hello World!\")' was not found.") >>> # All checks passed, task completed >>> return EvalResult(True, "Subtask completed successfully") """ actions = parse_actions() step_num = len(actions) trajectory_path = get_trajectory_path() last_control_path = os.path.join(trajectory_path, f"action_step{step_num-1}_controls_info.json") with open(last_control_path, 'r') as f: control_info = json.load(f) control_titles = [control['title'] for control in control_info] control_titles = ' '.join(control_titles) return fuzzy_in(text.upper(), control_titles.upper()) def check_text_exists(text: str) -> bool: """Checks if the specified text is included in the last screenshot. Parameters --------- text: str The text to be checked. Returns --------- bool True if the text is present in the last screenshot, False otherwise. Examples --------- >>> # Evaluate if the agent has successfully created a new folder named 'Project Files' >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion() -> EvalResult: >>> # Check if the folder name exists in the window >>> if not check_text_exists(text='Project Files'): >>> return EvalResult(False, "Subtask execution failed because folder 'Project Files' was not found.") >>> # All checks passed, task completed >>> return EvalResult(True, "Subtask completed successfully") """ return check_text_exists_via_ocr(text) or check_text_exists_via_control(text) def clipboard_contains_text() -> bool: """Checks if the clipboard contains text. Returns --------- bool True if the clipboard contains text, False otherwise. Examples --------- >>> # Evaluate if the agent has successfully opened the note and copied all the text >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion() -> EvalResult: >>> # Check if there is text in clipboard >>> if not clipboard_contains_text(): >>> return EvalResult(False, "Subtask execution failed because no text was found in clipboard.") >>> # All checks passed, task completed >>> return EvalResult(True, "Subtask completed successfully") """ IP = configs['VM_IP'] command = 'python -c "import win32clipboard; win32clipboard.OpenClipboard(); result = win32clipboard.IsClipboardFormatAvailable(win32clipboard.CF_UNICODETEXT) or win32clipboard.IsClipboardFormatAvailable(win32clipboard.CF_TEXT); win32clipboard.CloseClipboard(); print(result,end=\'\')"' payload = { "shell": True, "command": command } response = requests.post(f"http://{IP}:5000/execute", json=payload) response = response.json() assert response['status'] == 'success' return eval(response["output"]) def clipboard_contains_image() -> bool: """Checks if the clipboard contains an image. Returns --------- bool True if the clipboard contains an image, False otherwise. Examples --------- >>> # Evaluate if the agent has successfully copied the picture >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion() -> EvalResult: >>> # Check if there is an image in clipboard >>> if not clipboard_contains_image(): >>> return EvalResult(False, "Subtask execution failed because no image was found in clipboard.") >>> # All checks passed, task completed >>> return EvalResult(True, "Subtask completed successfully") """ IP = configs['VM_IP'] command = 'python -c "import win32clipboard; win32clipboard.OpenClipboard(); result = win32clipboard.IsClipboardFormatAvailable(win32clipboard.CF_DIB); win32clipboard.CloseClipboard(); print(result,end=\'\')"' payload = { "shell": True, "command": command } response = requests.post(f"http://{IP}:5000/execute", json=payload) response = response.json() assert response['status'] == 'success' return eval(response["output"]) def clipboard_contains_file() -> bool: """Checks if the clipboard contains file paths. Returns --------- bool True if the clipboard contains file paths, False otherwise. Examples --------- >>> # Evaluate if the agent has successfully copied the file >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion() -> EvalResult: >>> # Check if there is a file in clipboard >>> if not clipboard_contains_file(): >>> return EvalResult(False, "Subtask execution failed because no file was found in clipboard.") >>> # All checks passed, task completed >>> return EvalResult(True, "Subtask completed successfully") """ IP = configs['VM_IP'] command = 'python -c "import win32clipboard; win32clipboard.OpenClipboard(); result = win32clipboard.IsClipboardFormatAvailable(win32clipboard.CF_HDROP); win32clipboard.CloseClipboard(); print(result,end=\'\')"' payload = { "shell": True, "command": command } response = requests.post(f"http://{IP}:5000/execute", json=payload) response = response.json() assert response['status'] == 'success' return eval(response["output"]) def get_clipboard_text() -> str: """Retrieves text from the clipboard. Returns --------- str The text from the clipboard. Examples --------- >>> # Evaluate if the agent has successfully copied the image url >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion() -> EvalResult: >>> # Check if there is text in clipboard >>> if not clipboard_contains_text(): >>> return EvalResult(False, "Subtask execution failed because no text was found in clipboard.") >>> # Get the text from clipboard >>> clipboard_text = get_clipboard_text() >>> if not clipboard_text: >>> return EvalResult(False, "Subtask execution failed because could not get text from clipboard.") >>> # Check if text is a URL >>> if not clipboard_text.startswith(('http://', 'https://')): >>> return EvalResult(False, "Subtask execution failed because clipboard text is not a valid URL.") >>> # All checks passed, task completed >>> return EvalResult(True, "Subtask completed successfully") """ IP = configs['VM_IP'] command = 'python -c "import pyperclip; print(pyperclip.paste(), end=\'\')"' payload = { "shell": True, "command": command } response = requests.post(f"http://{IP}:5000/execute", json=payload) response = response.json() assert response['status'] == 'success' return response["output"] def get_clipboard_file_name() -> str: """Retrieves file name from the clipboard. Returns --------- str The file name from the clipboard. Examples --------- >>> # Evaluate if the agent has successfully copied the '{file_path}' >>> import os >>> from collections import namedtuple >>> EvalResult = namedtuple('EvalResult', ['success', 'message']) >>> def evaluate_agent_task_completion(file_path: str) -> EvalResult: >>> # Extract the file name >>> file_path = file_path.rstrip('/\\') >>> file_name = os.path.basename(file_path) >>> # Check if there is a file in clipboard >>> if not clipboard_contains_file(): >>> return EvalResult(False, "Subtask execution failed because no file was found in clipboard.") >>> # Get the file name from clipboard >>> file_name_in_clipboard = get_clipboard_file_name() >>> if not file_name_in_clipboard: >>> return EvalResult(False, "Subtask execution failed because could not get file name from clipboard.") >>> # If file path is provided, check if file name matches >>> expected_name = os.path.basename(file_path) >>> if expected_name != file_name_in_clipboard: >>> return EvalResult(False, f"Subtask execution failed because clipboard file name '{file_name_in_clipboard}' does not match expected '{expected_name}'.") >>> # All checks passed, task completed >>> return EvalResult(True, "Subtask completed successfully") """ IP = configs['VM_IP'] command = 'python -c "import win32clipboard; import os; win32clipboard.OpenClipboard(); file_names = win32clipboard.GetClipboardData(win32clipboard.CF_HDROP); print(os.path.basename(file_names[0]), end=\'\'); win32clipboard.CloseClipboard()"' payload = { "shell": True, "command": command } response = requests.post(f"http://{IP}:5000/execute", json=payload) response = response.json() assert response['status'] == 'success' return response["output"]