semantio 0.0.6__py3-none-any.whl → 0.0.7__py3-none-any.whl

semantio/tools/web_browser.py

@@ -1,24 +1,11 @@
 # web_browser.py
 from typing import Dict, Any, List, Optional, Callable
 from pydantic import Field, BaseModel
-from selenium import webdriver
-from selenium.webdriver.common.by import By
-from selenium.webdriver.common.action_chains import ActionChains
-from selenium.webdriver.remote.webelement import WebElement
-from selenium.webdriver.support.ui import WebDriverWait
-from selenium.webdriver.support import expected_conditions as EC
-from selenium.webdriver.chrome.options import Options
-from selenium.webdriver.chrome.service import Service
-from webdriver_manager.chrome import ChromeDriverManager
-from bs4 import BeautifulSoup
-import json
-import time
-import re
-import logging
-import os
-import difflib
+from playwright.sync_api import sync_playwright, Page, TimeoutError as PlaywrightTimeoutError
+import json, time, re, logging, os, difflib
 from .base_tool import BaseTool
 
+# Global logger
 logger = logging.getLogger(__name__)
 
 class BrowserPlan(BaseModel):
@@ -30,50 +17,63 @@ class BrowserPlan(BaseModel):
 class WebBrowserTool(BaseTool):
     name: str = Field("WebBrowser", description="Name of the tool")
     description: str = Field(
-        "Highly advanced universal web automation tool with advanced element identification, AJAX waiting, modal dismissal, multi-tab support, and custom JS injection.",
+        "Universal web automation tool with advanced element identification (DOM and image fallback), modal analysis, AJAX waiting, multi-tab support, and custom JS injection.",
         description="Tool description"
     )
-    
-    default_timeout: int = 15  # Default wait timeout in seconds
-    max_retries: int = 3       # Increased maximum retries for any task
+    default_timeout: int = 15000  # 15 seconds in milliseconds
+    max_retries: int = 3
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        # Bypass Pydantic's restrictions for extra attributes.
+        object.__setattr__(self, "logger", logging.getLogger(__name__))
 
     def execute(self, input: Dict[str, Any]) -> Dict[str, Any]:
-        """Execute an advanced dynamic web automation workflow."""
-        driver = None
+        """
+        Execute the browser automation workflow.
+        Maintains a context string of executed tasks and passes it to fallback routines.
+        DOES NOT close the browser after successful execution.
+        """
         overall_start = time.time()
+        results = []  # to hold summaries of executed tasks (for context)
+        current_url = ""
         try:
             headless = input.get("headless", False)
-            self.default_timeout = int(input.get("timeout", self.default_timeout))
+            self.default_timeout = int(input.get("timeout", 15)) * 1000
             self.max_retries = int(input.get("max_retries", self.max_retries))
-            driver = self._init_browser(headless)
-            results = []
-            current_url = ""
-
-            plan = self._generate_plan(input.get('query', ''), current_url)
+            plan = self._generate_plan(input.get("query", ""), current_url)
             if not plan.tasks:
                 raise ValueError("No valid tasks in the generated plan.")
 
-            # Dynamic mapping: action name to handler function.
-            action_map: Dict[str, Callable[[webdriver.Chrome, Dict[str, Any]], Dict[str, Any]]] = {
-                "navigate": lambda d, task: self._handle_navigation(d, task.get("value", "")),
-                "click": lambda d, task: self._handle_click(d, task.get("selector", "")),
-                "type": lambda d, task: self._handle_typing(d, task.get("selector", ""), task.get("value", ""), task),
-                "wait": lambda d, task: self._handle_wait(task.get("value", "")),
-                "wait_for_ajax": lambda d, task: self._handle_wait_for_ajax(d, task.get("value", "30")),
-                "scroll": lambda d, task: self._handle_scroll(d, task.get("selector", "")),
-                "hover": lambda d, task: self._handle_hover(d, task.get("selector", "")),
-                "screenshot": lambda d, task: self._handle_screenshot(d, task.get("value", "screenshot.png")),
-                "switch_tab": lambda d, task: self._handle_switch_tab(d, task.get("value", "0")),
-                "execute_script": lambda d, task: self._handle_execute_script(d, task.get("value", "")),
-                "drag_and_drop": lambda d, task: self._handle_drag_and_drop(d, task.get("selector", ""), task.get("value", "")),
+            # Start Playwright without a "with" block so we can leave the browser open.
+            p = sync_playwright().start()
+            browser = p.chromium.launch(headless=headless)
+            context = browser.new_context()
+            page = context.new_page()
+
+            # Map actions to handlers.
+            action_map: Dict[str, Callable[[Page, Dict[str, Any]], Dict[str, Any]]] = {
+                "navigate": lambda p, task: self._handle_navigation(p, task.get("value", "")),
+                "click": lambda p, task: self._handle_click(p, task.get("selector", "")),
+                "type": lambda p, task: self._handle_typing(p, task.get("selector", ""), task.get("value", ""), task),
+                "wait": lambda p, task: self._handle_wait(task.get("value", "")),
+                "wait_for_ajax": lambda p, task: self._handle_wait_for_ajax(p, task.get("value", "")),
+                "scroll": lambda p, task: self._handle_scroll(p, task.get("selector", "")),
+                "hover": lambda p, task: self._handle_hover(p, task.get("selector", "")),
+                "screenshot": lambda p, task: self._handle_screenshot(p, task.get("value", "screenshot.png")),
+                "switch_tab": lambda p, task: self._handle_switch_tab(context, task.get("value", "0")),
+                "execute_script": lambda p, task: self._handle_execute_script(p, task.get("value", "")),
+                "drag_and_drop": lambda p, task: self._handle_drag_and_drop(p, task.get("selector", ""), task.get("value", "")),
             }
             
             for task in plan.tasks:
-                # Before each action, dismiss modals/overlays.
-                self._dismiss_unwanted_modals(driver)
+                self._dismiss_unwanted_modals(page, task_context=task.get("description", ""))
                 action = task.get("action", "").lower()
-                logger.info(f"Executing task: {task.get('description', action)}")
+                self.logger.info(f"Executing task: {task.get('description', action)}")
                 start_time = time.time()
+
+                # Build a context string from previously executed tasks.
+                executed_context = "\n".join([f"{r['action']}: {r['message']}" for r in results])
                 handler = action_map.get(action)
                 if not handler:
                     results.append({
@@ -83,45 +83,28 @@ class WebBrowserTool(BaseTool):
                     })
                     continue
 
-                result = self._execute_with_retries(driver, task, handler)
+                result = self._execute_with_retries(page, task, handler, executed_context)
                 elapsed = time.time() - start_time
                 result["elapsed"] = elapsed
-                logger.info(f"Action '{action}' completed in {elapsed:.2f} seconds.")
+                self.logger.info(f"Action '{action}' completed in {elapsed:.2f} seconds.")
                 results.append(result)
 
-                if not result.get('success', False):
-                    logger.error(f"Task failed: {result.get('message')}")
-                    self._capture_failure_screenshot(driver, action)
+                if not result.get("success", False):
+                    self.logger.error(f"Task failed: {result.get('message')}")
+                    self._capture_failure_screenshot(page, action)
                     break
 
-                current_url = driver.current_url
+                current_url = page.url
 
             overall_elapsed = time.time() - overall_start
-            logger.info(f"Total execution time: {overall_elapsed:.2f} seconds.")
+            self.logger.info(f"Total execution time: {overall_elapsed:.2f} seconds.")
+            # Do not close the browser.
             return {"status": "success", "results": results, "total_time": overall_elapsed}
-            
         except Exception as e:
-            logger.exception("Execution error:")
+            self.logger.exception("Execution error:")
             return {"status": "error", "message": str(e)}
-        finally:
-            if driver:
-                driver.quit()
-
-    def _init_browser(self, headless: bool) -> webdriver.Chrome:
-        """Initialize browser with advanced options."""
-        options = Options()
-        options.add_argument("--start-maximized")
-        options.add_argument("--disable-blink-features=AutomationControlled")
-        options.add_experimental_option("excludeSwitches", ["enable-automation"])
-        if headless:
-            options.add_argument("--headless=new")
-        return webdriver.Chrome(
-            service=Service(ChromeDriverManager().install()),
-            options=options
-        )
 
     def _generate_plan(self, query: str, current_url: str) -> BrowserPlan:
-        """Generate an adaptive execution plan using an LLM or other dynamic planner."""
         prompt = f"""Generate browser automation plan for: {query}
 
 Current URL: {current_url or 'No page loaded yet'}
@@ -150,7 +133,6 @@ Guidelines:
         return self._parse_plan(response)
 
     def _parse_plan(self, response: str) -> BrowserPlan:
-        """Robust JSON parsing with multiple fallback strategies."""
         try:
             json_match = re.search(r'```json\n?(.+?)\n?```', response, re.DOTALL)
             if json_match:
@@ -163,7 +145,7 @@ Guidelines:
             validated_tasks = []
             for task in plan_data.get("tasks", []):
                 if not all(key in task for key in ["action", "description"]):
-                    logger.warning(f"Skipping task due to missing keys: {task}")
+                    self.logger.warning(f"Skipping task due to missing keys: {task}")
                     continue
                 validated_tasks.append({
                     "action": task["action"],
@@ -173,267 +155,391 @@ Guidelines:
                 })
             return BrowserPlan(tasks=validated_tasks)
         except (json.JSONDecodeError, AttributeError, ValueError) as e:
-            logger.error(f"Plan parsing failed: {e}")
+            self.logger.error(f"Plan parsing failed: {e}")
             return BrowserPlan(tasks=[])
 
-    def _execute_with_retries(self, driver: webdriver.Chrome, task: Dict[str, Any],
-                                handler: Callable[[webdriver.Chrome, Dict[str, Any]], Dict[str, Any]]) -> Dict[str, Any]:
-        """Execute a task with retry logic and exponential backoff."""
+    def _execute_with_retries(self, page: Page, task: Dict[str, Any],
+                                handler: Callable[[Page, Dict[str, Any]], Dict[str, Any]],
+                                executed_context: str = "") -> Dict[str, Any]:
+        """Execute a task with retry logic. If it fails, pass the executed_context to the fallback prompt.
+        The fallback now returns a JSON array of tasks, which are executed sequentially."""
         attempts = 0
         result = {}
         while attempts < self.max_retries:
-            result = self._execute_safe_task(driver, task, handler)
+            result = self._execute_safe_task(page, task, handler)
             if result.get("success", False):
                 return result
             attempts += 1
-            logger.info(f"Retrying task '{task.get('action')}' (attempt {attempts + 1}/{self.max_retries})")
+            self.logger.info(f"Retrying task '{task.get('action')}' (attempt {attempts + 1}/{self.max_retries})")
             time.sleep(1 * attempts)
+        if task.get("action") in ["click", "type"]:
+            self.logger.info("HTML-based automation failed. Using fallback with image-based LLM.")
+            result = self._fallback_with_image_llm(page, task, executed_context)
         return result
 
-    def _execute_safe_task(self, driver: webdriver.Chrome, task: Dict[str, Any],
-                             handler: Callable[[webdriver.Chrome, Dict[str, Any]], Dict[str, Any]]) -> Dict[str, Any]:
-        """Execute a task with comprehensive error handling."""
+    def _execute_safe_task(self, page: Page, task: Dict[str, Any],
+                             handler: Callable[[Page, Dict[str, Any]], Dict[str, Any]]) -> Dict[str, Any]:
         try:
-            return handler(driver, task)
+            return handler(page, task)
         except Exception as e:
             action = task.get("action", "unknown")
-            logger.exception(f"Error executing task '{action}':")
+            self.logger.exception(f"Error executing task '{action}':")
             return {"action": action, "success": False, "message": f"Critical error: {str(e)}"}
 
-    def _dismiss_unwanted_modals(self, driver: webdriver.Chrome):
-        """
-        Dismiss or remove unwanted modals, overlays, or pop-ups.
-        First attempts to click a close button; if not available, removes the element via JS.
-        """
+    def _dismiss_unwanted_modals(self, page: Page, task_context: str = ""):
+        modal_selectors = [".modal", ".popup", '[role="dialog"]', ".overlay", ".lightbox"]
+        for selector in modal_selectors:
+            elements = page.query_selector_all(selector)
+            for modal in elements:
+                if modal.is_visible():
+                    self._handle_modal(page, modal, task_context)
+
+    def _handle_modal(self, page: Page, modal_element, task_context: str):
         try:
-            modal_selectors = [".modal", ".popup", '[role="dialog"]', ".overlay", ".lightbox"]
-            for selector in modal_selectors:
-                elements = driver.find_elements(By.CSS_SELECTOR, selector)
-                for modal in elements:
-                    if modal.is_displayed():
-                        close_selectors = [".close", ".btn-close", "[aria-label='Close']", "[data-dismiss='modal']"]
-                        dismissed = False
-                        for close_sel in close_selectors:
-                            try:
-                                close_button = modal.find_element(By.CSS_SELECTOR, close_sel)
-                                if close_button.is_displayed():
-                                    close_button.click()
-                                    dismissed = True
-                                    logger.info(f"Dismissed modal using selector {close_sel}")
-                                    time.sleep(1)
-                                    break
-                            except Exception:
-                                continue
-                        if not dismissed:
-                            # Remove overlay by setting display to none
-                            driver.execute_script("arguments[0].remove();", modal)
-                            logger.info(f"Removed overlay/modal with selector {selector}")
+            modal_screenshot = modal_element.screenshot()
+            prompt = (
+                f"A modal is displayed on the page. The content is visible in the attached image. "
+                f"The current task context is: \"{task_context}\". "
+                "Based on the content of the modal and the task context, decide whether to dismiss the modal. "
+                "Return a JSON response in the format: { \"action\": \"dismiss\" } to dismiss or { \"action\": \"ignore\" } to leave it. "
+                "Return only the JSON."
+            )
+            response_text = self.llm.generate_from_image(prompt, image_bytes=modal_screenshot)
+            self.logger.info(f"LLM response for modal analysis: {response_text}")
+            json_match = re.search(r'```json\n?(.+?)\n?```', response_text, re.DOTALL)
+            json_text = json_match.group(1).strip() if json_match else response_text.strip()
+            decision = json.loads(json_text)
+            if decision.get("action") == "dismiss":
+                close_buttons = modal_element.query_selector_all(".close, .btn-close, [aria-label='Close'], [data-dismiss='modal']")
+                for btn in close_buttons:
+                    if btn.is_visible():
+                        btn.click()
+                        self.logger.info("Modal dismissed using a close button.")
+                        return
+                page.evaluate("(modal) => modal.remove()", modal_element)
+                self.logger.info("Modal dismissed by removal.")
+            else:
+                self.logger.info("Modal left intact according to LLM analysis.")
         except Exception as e:
-            logger.debug(f"Modal dismissal error: {e}")
+            self.logger.error(f"Modal handling error: {e}")
 
-    def _advanced_find_element(self, driver: webdriver.Chrome, keyword: str) -> Optional[WebElement]:
+    def _advanced_find_element(self, page: Page, keyword: str):
+        try:
+            candidates = page.query_selector_all("input, textarea, button, a, div")
+            best_match = None
+            best_ratio = 0.0
+            for candidate in candidates:
+                attrs = page.evaluate(
+                    """(el) => {
+                        return {
+                            id: el.id,
+                            name: el.getAttribute('name'),
+                            placeholder: el.getAttribute('placeholder'),
+                            aria: el.getAttribute('aria-label'),
+                            text: el.innerText
+                        };
+                    }""",
+                    candidate,
+                )
+                combined_text = " ".join(
+                    filter(None, [
+                        attrs.get("id"),
+                        attrs.get("name"),
+                        attrs.get("placeholder"),
+                        attrs.get("aria"),
+                        attrs.get("text"),
+                    ])
+                )
+                ratio = difflib.SequenceMatcher(None, combined_text.lower(), keyword.lower()).ratio()
+                if ratio > best_ratio:
+                    best_ratio = ratio
+                    best_match = candidate
+            if best_ratio > 0.5:
+                self.logger.info(f"Advanced fallback detected element with similarity {best_ratio:.2f} for keyword '{keyword}'")
+                return best_match
+            return None
+        except Exception as e:
+            self.logger.error(f"Advanced find element error: {e}")
+            return None
+
+    def _annotate_page_with_numbers(self, page: Page, query: str = "button, a, input, [onclick]"):
+        script = f"""
+        (() => {{
+            document.querySelectorAll('.automation-annotation-overlay').forEach(el => el.remove());
+            const elements = document.querySelectorAll('{query}');
+            let counter = 1;
+            elements.forEach(el => {{
+                const rect = el.getBoundingClientRect();
+                if (rect.width === 0 || rect.height === 0) return;
+                const overlay = document.createElement('div');
+                overlay.classList.add('automation-annotation-overlay');
+                overlay.style.position = 'absolute';
+                overlay.style.left = (rect.left + window.scrollX) + 'px';
+                overlay.style.top = (rect.top + window.scrollY) + 'px';
+                overlay.style.width = rect.width + 'px';
+                overlay.style.height = rect.height + 'px';
+                overlay.style.border = '2px solid red';
+                overlay.style.zIndex = 9999;
+                overlay.style.pointerEvents = 'none';
+                overlay.textContent = counter;
+                overlay.style.fontSize = '16px';
+                overlay.style.fontWeight = 'bold';
+                overlay.style.color = 'red';
+                overlay.style.backgroundColor = 'rgba(255, 255, 255, 0.7)';
+                document.body.appendChild(overlay);
+                counter += 1;
+            }});
+        }})();
         """
-        Advanced fallback for finding an element.
-        Searches across multiple attributes and inner text using fuzzy matching.
+        page.evaluate(script)
+
+    def _click_element_by_number(self, page: Page, number: int) -> Dict[str, Any]:
+        candidates = [el for el in page.query_selector_all("button, a, input, [onclick]") if el.is_visible()]
+        index = number - 1
+        if index < len(candidates):
+            candidate = candidates[index]
+            candidate.scroll_into_view_if_needed()
+            try:
+                candidate.click()
+                return {"action": "click", "success": True, "message": f"Clicked element number {number}"}
+            except Exception as e:
+                return {"action": "click", "success": False, "message": f"Click failed: {str(e)}"}
+        else:
+            return {"action": "click", "success": False, "message": f"Element number {number} not found."}
+
+    def _fallback_with_image_llm(self, page: Page, task: Dict[str, Any], executed_context: str = "") -> Dict[str, Any]:
+        """
+        Fallback method: Annotate the page, capture a screenshot, and ask the LLM (via image analysis)
+        to generate a JSON array of tasks for the next steps.
+        Each fallback task is an object:
+          {
+            "action": "click" or "type",
+            "element_number": <number>,
+            "text": <if action is 'type', the text to type; otherwise an empty string>
+          }
+        The prompt includes the executed_context.
         """
-        candidates = driver.find_elements(By.CSS_SELECTOR, "input, textarea, button, a, div")
-        best_match = None
-        best_ratio = 0.0
-        for candidate in candidates:
-            combined_text = " ".join([
-                candidate.get_attribute("id") or "",
-                candidate.get_attribute("name") or "",
-                candidate.get_attribute("placeholder") or "",
-                candidate.get_attribute("aria-label") or "",
-                candidate.text or "",
-            ])
-            ratio = difflib.SequenceMatcher(None, combined_text.lower(), keyword.lower()).ratio()
-            if ratio > best_ratio:
-                best_ratio = ratio
-                best_match = candidate
-        if best_ratio > 0.5:
-            logger.info(f"Advanced fallback detected element with similarity {best_ratio:.2f} for keyword '{keyword}'")
-            return best_match
-        return None
-
-    def _handle_navigation(self, driver: webdriver.Chrome, url: str) -> Dict[str, Any]:
-        """Handle navigation with URL correction."""
+        query = "input, textarea" if task.get("action") == "type" else "button, a, input, [onclick]"
+        self._annotate_page_with_numbers(page, query=query)
+        time.sleep(1)
+        screenshot_bytes = page.screenshot(type="png")
+        extra = ""
+        if task.get("action") == "type":
+            extra = f"\nThe exact text to be entered is: \"{task.get('value', '').strip()}\"."
+        prompt = (
+            f"Tasks executed so far:\n{executed_context}\n\n"
+            f"The following task remains: {task.get('description', '')}.{extra}\n"
+            "I have annotated the page with numbered overlays using the appropriate query. "
+            "Based on the attached screenshot, generate a JSON array of tasks that need to be performed next. "
+            "Each task should be a JSON object with the format:\n"
+            "[\n"
+            "  {\n"
+            "    \"action\": \"click\" or \"type\",\n"
+            "    \"element_number\": <number>,\n"
+            "    \"text\": <if action is 'type', the text to type; otherwise an empty string>\n"
+            "  },\n"
+            "  ...\n"
+            "]\n"
+            "Return only the JSON array."
+        )
+        response_text = self.llm.generate_from_image(prompt, image_bytes=screenshot_bytes)
+        self.logger.info(f"LLM response for fallback: {response_text}")
+        try:
+            fallback_tasks = json.loads(response_text.strip())
+            if not isinstance(fallback_tasks, list):
+                fallback_tasks = [fallback_tasks]
+        except Exception as e:
+            json_match = re.search(r'```json\n?(.+?)\n?```', response_text, re.DOTALL)
+            if json_match:
+                json_text = json_match.group(1).strip()
+                fallback_tasks = json.loads(json_text)
+                if not isinstance(fallback_tasks, list):
+                    fallback_tasks = [fallback_tasks]
+            else:
+                return {"action": task.get("action"), "success": False, "message": f"Fallback failed to parse JSON: {str(e)}"}
+
+        fallback_results = []
+        for fb_task in fallback_tasks:
+            action = fb_task.get("action")
+            element_number = fb_task.get("element_number")
+            if action == "type":
+                returned_text = fb_task.get("text", "").strip()
+                original_text = task.get("value", "").strip()
+                if returned_text.lower() != original_text.lower():
+                    self.logger.info("Overriding LLM-provided text with original input text.")
+                    text = original_text
+                else:
+                    text = returned_text
+            else:
+                text = fb_task.get("text", "")
+            if action == "click":
+                self.logger.info(f"LLM indicated fallback click on element number {element_number}.")
+                res = self._click_element_by_number(page, element_number)
+            elif action == "type":
+                candidates = [el for el in page.query_selector_all("input, textarea") if el.is_visible()]
+                if element_number - 1 < len(candidates):
+                    candidate = candidates[element_number - 1]
+                    candidate.scroll_into_view_if_needed()
+                    try:
+                        candidate.fill(text, timeout=self.default_timeout)
+                        res = {"action": "type", "success": True, "message": f"Typed '{text}' into element number {element_number}"}
+                    except Exception as ex:
+                        res = {"action": "type", "success": False, "message": f"Typing failed on fallback element: {str(ex)}"}
+                else:
+                    res = {"action": "type", "success": False, "message": f"Element number {element_number} not found."}
+            else:
+                res = {"action": task.get("action"), "success": False, "message": "Invalid fallback action."}
+            fallback_results.append(res)
+        overall_success = any(r.get("success", False) for r in fallback_results)
+        overall_message = "; ".join([r.get("message", "") for r in fallback_results])
+        return {"action": task.get("action"), "success": overall_success, "message": overall_message}
+
+    def _handle_navigation(self, page: Page, url: str) -> Dict[str, Any]:
         if not url.startswith(("http://", "https://")):
             url = f"https://{url}"
         try:
-            driver.get(url)
-            WebDriverWait(driver, self.default_timeout).until(EC.presence_of_element_located((By.TAG_NAME, "body")))
+            page.goto(url, timeout=self.default_timeout)
+            page.wait_for_selector("body", timeout=self.default_timeout)
             return {"action": "navigate", "success": True, "message": f"Navigated to {url}"}
+        except PlaywrightTimeoutError as e:
+            self.logger.error(f"Navigation to {url} timed out: {e}")
+            return {"action": "navigate", "success": False, "message": f"Navigation timed out: {str(e)}"}
         except Exception as e:
-            logger.error(f"Navigation to {url} failed: {e}")
+            self.logger.error(f"Navigation to {url} failed: {e}")
             return {"action": "navigate", "success": False, "message": f"Navigation failed: {str(e)}"}
 
-    def _handle_click(self, driver: webdriver.Chrome, selector: str) -> Dict[str, Any]:
-        """Handle click actions with fallback using JS if needed."""
+    def _handle_click(self, page: Page, selector: str) -> Dict[str, Any]:
         try:
-            element = WebDriverWait(driver, self.default_timeout).until(
-                EC.element_to_be_clickable((By.CSS_SELECTOR, selector))
-            )
-            driver.execute_script("arguments[0].scrollIntoView({behavior: 'smooth', block: 'center'});", element)
-            try:
-                element.click()
-            except Exception:
-                driver.execute_script("arguments[0].click();", element)
+            page.wait_for_selector(selector, state="visible", timeout=self.default_timeout)
+            page.click(selector, timeout=self.default_timeout)
             return {"action": "click", "success": True, "message": f"Clicked element: {selector}"}
+        except PlaywrightTimeoutError as e:
+            self.logger.error(f"Click action timed out on selector {selector}: {e}")
+            return {"action": "click", "success": False, "message": f"Click timed out: {str(e)}"}
         except Exception as e:
-            logger.error(f"Click action failed on selector {selector}: {e}")
+            self.logger.error(f"Click action failed on selector {selector}: {e}")
             return {"action": "click", "success": False, "message": f"Click failed: {str(e)}"}
 
-    def _handle_typing(self, driver: webdriver.Chrome, selector: str, text: str, task: Dict[str, Any]) -> Dict[str, Any]:
-        """
-        Handle typing into an element.
-        If the primary selector fails, attempt advanced fallback detection.
-        """
-        try:
-            element = WebDriverWait(driver, self.default_timeout).until(
-                EC.presence_of_element_located((By.CSS_SELECTOR, selector))
-            )
-        except Exception as e:
-            # If the task seems to involve search or similar text, use advanced fallback.
-            if "search" in task.get("description", "").lower() or "search" in selector.lower():
-                logger.info("Primary selector failed; using advanced fallback for element detection.")
-                element = self._advanced_find_element(driver, "search")
-                if not element:
-                    return {"action": "type", "success": False, "message": f"Typing failed: No search-like element found; error: {str(e)}"}
-            else:
-                return {"action": "type", "success": False, "message": f"Typing failed: {str(e)}"}
+    def _handle_typing(self, page: Page, selector: str, text: str, task: Dict[str, Any]) -> Dict[str, Any]:
         try:
-            element.clear()
-            element.send_keys(text)
+            page.wait_for_selector(selector, state="attached", timeout=self.default_timeout)
+            page.fill(selector, text, timeout=self.default_timeout)
             return {"action": "type", "success": True, "message": f"Typed '{text}' into element."}
+        except PlaywrightTimeoutError as e:
+            self.logger.info("Primary selector failed; using advanced fallback for element detection.")
+            element = self._advanced_find_element(page, "search")
+            if not element:
+                return {"action": "type", "success": False, "message": f"Typing failed: No search-like element found; error: {str(e)}"}
+            try:
+                element.fill(text, timeout=self.default_timeout)
+                return {"action": "type", "success": True, "message": f"Typed '{text}' into fallback element."}
+            except Exception as ex:
+                return {"action": "type", "success": False, "message": f"Typing failed on fallback element: {str(ex)}"}
         except Exception as e:
-            logger.error(f"Typing action failed: {e}")
+            self.logger.error(f"Typing action failed: {e}")
             return {"action": "type", "success": False, "message": f"Typing failed: {str(e)}"}
 
     def _handle_wait(self, seconds: str) -> Dict[str, Any]:
-        """Handle a simple wait."""
         try:
             wait_time = float(seconds)
-            logger.info(f"Waiting for {wait_time} seconds")
+            self.logger.info(f"Waiting for {wait_time} seconds")
             time.sleep(wait_time)
             return {"action": "wait", "success": True, "message": f"Waited {wait_time} seconds"}
         except ValueError as e:
-            logger.error(f"Invalid wait time provided: {seconds}")
+            self.logger.error(f"Invalid wait time provided: {seconds}")
             return {"action": "wait", "success": False, "message": "Invalid wait time"}
 
-    def _handle_wait_for_ajax(self, driver: webdriver.Chrome, seconds: str) -> Dict[str, Any]:
-        """
-        Wait until AJAX/network activity has subsided.
-        This implementation first checks for jQuery, then falls back to a generic check.
-        """
+    def _handle_wait_for_ajax(self, page: Page, seconds: str) -> Dict[str, Any]:
         try:
-            timeout = int(seconds)
-            logger.info(f"Waiting for AJAX/network activity for up to {timeout} seconds.")
-            end_time = time.time() + timeout
+            timeout_seconds = int(seconds) if seconds.strip() != "" else 30
+            self.logger.info(f"Waiting for AJAX/network activity for up to {timeout_seconds} seconds.")
+            end_time = time.time() + timeout_seconds
             while time.time() < end_time:
-                ajax_complete = driver.execute_script("""
-                    return (window.jQuery ? jQuery.active === 0 : true) &&
-                           (typeof window.fetch === 'function' ? true : true);
+                ajax_complete = page.evaluate("""
+                    () => {
+                        return (window.jQuery ? jQuery.active === 0 : true) &&
+                               (typeof window.fetch === 'function' ? true : true);
+                    }
                 """)
                 if ajax_complete:
                     break
                 time.sleep(0.5)
             return {"action": "wait_for_ajax", "success": True, "message": "AJAX/network activity subsided."}
         except Exception as e:
-            logger.error(f"Wait for AJAX failed: {e}")
+            self.logger.error(f"Wait for AJAX failed: {e}")
             return {"action": "wait_for_ajax", "success": False, "message": f"Wait for AJAX failed: {str(e)}"}
 
-    def _handle_scroll(self, driver: webdriver.Chrome, selector: str) -> Dict[str, Any]:
-        """Handle scrolling to a specific element or page bottom."""
+    def _handle_scroll(self, page: Page, selector: str) -> Dict[str, Any]:
         try:
             if selector:
-                element = WebDriverWait(driver, self.default_timeout).until(
-                    EC.presence_of_element_located((By.CSS_SELECTOR, selector))
-                )
-                driver.execute_script("arguments[0].scrollIntoView({behavior: 'smooth', block: 'center'});", element)
+                page.wait_for_selector(selector, timeout=self.default_timeout)
+                page.eval_on_selector(selector, "el => el.scrollIntoView({behavior: 'smooth', block: 'center'})")
                 scroll_target = selector
             else:
-                driver.execute_script("window.scrollTo(0, document.body.scrollHeight);")
+                page.evaluate("window.scrollTo(0, document.body.scrollHeight);")
                 scroll_target = "page bottom"
             return {"action": "scroll", "success": True, "message": f"Scrolled to {scroll_target}"}
         except Exception as e:
-            logger.error(f"Scroll action failed on selector {selector}: {e}")
+            self.logger.error(f"Scroll action failed on selector {selector}: {e}")
             return {"action": "scroll", "success": False, "message": f"Scroll failed: {str(e)}"}
 
-    def _handle_hover(self, driver: webdriver.Chrome, selector: str) -> Dict[str, Any]:
-        """Handle mouse hover action."""
+    def _handle_hover(self, page: Page, selector: str) -> Dict[str, Any]:
         try:
-            element = WebDriverWait(driver, self.default_timeout).until(
-                EC.visibility_of_element_located((By.CSS_SELECTOR, selector))
-            )
-            ActionChains(driver).move_to_element(element).perform()
+            page.wait_for_selector(selector, state="visible", timeout=self.default_timeout)
+            page.hover(selector, timeout=self.default_timeout)
             return {"action": "hover", "success": True, "message": f"Hovered over {selector}"}
         except Exception as e:
-            logger.error(f"Hover action failed on selector {selector}: {e}")
+            self.logger.error(f"Hover action failed on selector {selector}: {e}")
             return {"action": "hover", "success": False, "message": f"Hover failed: {str(e)}"}
 
-    def _handle_screenshot(self, driver: webdriver.Chrome, filename: str) -> Dict[str, Any]:
-        """Capture a screenshot of the current browser state."""
+    def _handle_screenshot(self, page: Page, filename: str) -> Dict[str, Any]:
         try:
-            driver.save_screenshot(filename)
+            page.screenshot(path=filename)
             return {"action": "screenshot", "success": True, "message": f"Screenshot saved as {filename}"}
         except Exception as e:
-            logger.error(f"Screenshot capture failed: {e}")
+            self.logger.error(f"Screenshot capture failed: {e}")
             return {"action": "screenshot", "success": False, "message": f"Screenshot failed: {str(e)}"}
 
-    def _handle_switch_tab(self, driver: webdriver.Chrome, value: str) -> Dict[str, Any]:
-        """
-        Switch between tabs. 'value' can be an index or the keyword 'new'.
-        """
+    def _handle_switch_tab(self, context, value: str) -> Dict[str, Any]:
         try:
-            handles = driver.window_handles
+            pages = context.pages
             if value.lower() == "new":
-                target_handle = handles[-1]
+                target_page = pages[-1]
             else:
                 idx = int(value)
-                if idx < len(handles):
-                    target_handle = handles[idx]
+                if idx < len(pages):
+                    target_page = pages[idx]
                 else:
                     return {"action": "switch_tab", "success": False, "message": f"Tab index {value} out of range"}
-            driver.switch_to.window(target_handle)
             return {"action": "switch_tab", "success": True, "message": f"Switched to tab {value}"}
         except Exception as e:
-            logger.error(f"Switch tab failed: {e}")
+            self.logger.error(f"Switch tab failed: {e}")
             return {"action": "switch_tab", "success": False, "message": f"Switch tab failed: {str(e)}"}
 
-    def _handle_execute_script(self, driver: webdriver.Chrome, script: str) -> Dict[str, Any]:
-        """
-        Execute arbitrary JavaScript code.
-        """
+    def _handle_execute_script(self, page: Page, script: str) -> Dict[str, Any]:
         try:
-            result = driver.execute_script(script)
+            result = page.evaluate(script)
             return {"action": "execute_script", "success": True, "message": "Script executed successfully", "result": result}
         except Exception as e:
-            logger.error(f"Execute script failed: {e}")
+            self.logger.error(f"Execute script failed: {e}")
             return {"action": "execute_script", "success": False, "message": f"Script execution failed: {str(e)}"}
 
-    def _handle_drag_and_drop(self, driver: webdriver.Chrome, source_selector: str, target_selector: str) -> Dict[str, Any]:
-        """
-        Simulate a drag-and-drop operation.
-        """
+    def _handle_drag_and_drop(self, page: Page, source_selector: str, target_selector: str) -> Dict[str, Any]:
         try:
-            source = WebDriverWait(driver, self.default_timeout).until(
-                EC.presence_of_element_located((By.CSS_SELECTOR, source_selector))
-            )
-            target = WebDriverWait(driver, self.default_timeout).until(
-                EC.presence_of_element_located((By.CSS_SELECTOR, target_selector))
-            )
-            ActionChains(driver).drag_and_drop(source, target).perform()
+            page.wait_for_selector(source_selector, timeout=self.default_timeout)
+            page.wait_for_selector(target_selector, timeout=self.default_timeout)
+            source = page.locator(source_selector)
+            target = page.locator(target_selector)
+            source.drag_to(target, timeout=self.default_timeout)
             return {"action": "drag_and_drop", "success": True, "message": f"Dragged element from {source_selector} to {target_selector}"}
         except Exception as e:
-            logger.error(f"Drag and drop failed from {source_selector} to {target_selector}: {e}")
+            self.logger.error(f"Drag and drop failed from {source_selector} to {target_selector}: {e}")
             return {"action": "drag_and_drop", "success": False, "message": f"Drag and drop failed: {str(e)}"}
 
-    def _capture_failure_screenshot(self, driver: webdriver.Chrome, action: str):
-        """Capture a screenshot for debugging when an error occurs."""
+    def _capture_failure_screenshot(self, page: Page, action: str):
         filename = f"failure_{action}_{int(time.time())}.png"
         try:
-            driver.save_screenshot(filename)
-            logger.info(f"Failure screenshot captured: {filename}")
+            page.screenshot(path=filename)
+            self.logger.info(f"Failure screenshot captured: {filename}")
         except Exception as e:
-            logger.error(f"Failed to capture screenshot: {e}")
+            self.logger.error(f"Failed to capture screenshot: {e}")

{semantio-0.0.6.dist-info → semantio-0.0.7.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: semantio
-Version: 0.0.6
+Version: 0.0.7
 Summary: A powerful SDK for building AI agents
 Home-page: https://github.com/Syenah/semantio
 Author: Rakesh
@@ -33,7 +33,6 @@ Requires-Dist: sentence-transformers
 Requires-Dist: fuzzywuzzy
 Requires-Dist: duckduckgo-search
 Requires-Dist: yfinance
-Requires-Dist: selenium
 Requires-Dist: beautifulsoup4
 Requires-Dist: webdriver-manager
 Requires-Dist: validators

{semantio-0.0.6.dist-info → semantio-0.0.7.dist-info}/RECORD

@@ -30,16 +30,16 @@ semantio/tools/base_tool.py,sha256=xBNSa_8a8WmA4BGRLG2dE7wj9GnBcZo7-P2SyD86GvY,5
 semantio/tools/crypto.py,sha256=mut1ztvpPcUUP3b563dh_FmKtP68KmNis3Qm8WENj8w,5559
 semantio/tools/duckduckgo.py,sha256=6mGn0js0cIsVxQlAgB8AYNLP05H8WmJKnSVosiO9iH0,5034
 semantio/tools/stocks.py,sha256=BVuK61O9OmWQjj0YdiCJY6TzpiFJ_An1UJB2RkDfX2k,5393
-semantio/tools/web_browser.py,sha256=ZjE-nrRtUEUoh6LNutMjdn9_bTlnVGU2PM2w6orjUWA,22172
+semantio/tools/web_browser.py,sha256=8-_SXvu3CRnIwKBlcmNe7-9DOd4y7OC7T24RB0xKMnI,28911
 semantio/utils/__init__.py,sha256=Lx4X4iJpRhZzRmpQb80XXh5Ve8ZMOkadWAxXSmHpO_8,244
 semantio/utils/config.py,sha256=ZTwUTqxjW3-w94zoU7GzivWyJe0JJGvBfuB4RUOuEs8,1198
 semantio/utils/date_utils.py,sha256=x3oqRGv6ee_KCJ0LvCqqZh_FSgS6YGOHBwZQS4TJetY,1471
 semantio/utils/file_utils.py,sha256=b_cMuJINEGk9ikNuNHSn9lsmICWwvtnCDZ03ndH_S2I,1779
 semantio/utils/logger.py,sha256=TmGbP8BRjLMWjXi2GWzZ0RIXt70x9qX3FuIqghCNlwM,510
 semantio/utils/validation_utils.py,sha256=iwoxEb4Q5ILqV6tbesMjPWPCCoL3AmPLejGUy6q8YvQ,1284
-semantio-0.0.6.dist-info/LICENSE,sha256=mziLlfb9hZ8HKxm9V6BiHpmgJvmcDvswu1QBlDB-6vU,1074
-semantio-0.0.6.dist-info/METADATA,sha256=SeDpiywG59La7HcAQxZHxk8uXLFoyEv3otMFYkJl5W4,6913
-semantio-0.0.6.dist-info/WHEEL,sha256=ewwEueio1C2XeHTvT17n8dZUJgOvyCWCt0WVNLClP9o,92
-semantio-0.0.6.dist-info/entry_points.txt,sha256=zbPgevSLwcLpdRHqI_atE8EOt8lK2vRF1AoDflDTo18,53
-semantio-0.0.6.dist-info/top_level.txt,sha256=Yte_6mb-bh-I_lQwMjk1GijZkxPoX4Zmp3kBftC1ZlA,9
-semantio-0.0.6.dist-info/RECORD,,
+semantio-0.0.7.dist-info/LICENSE,sha256=mziLlfb9hZ8HKxm9V6BiHpmgJvmcDvswu1QBlDB-6vU,1074
+semantio-0.0.7.dist-info/METADATA,sha256=QQRzinLKReosRRthYf1bei5FDAaOPHaG4bG5gdJnMFc,6889
+semantio-0.0.7.dist-info/WHEEL,sha256=ewwEueio1C2XeHTvT17n8dZUJgOvyCWCt0WVNLClP9o,92
+semantio-0.0.7.dist-info/entry_points.txt,sha256=zbPgevSLwcLpdRHqI_atE8EOt8lK2vRF1AoDflDTo18,53
+semantio-0.0.7.dist-info/top_level.txt,sha256=Yte_6mb-bh-I_lQwMjk1GijZkxPoX4Zmp3kBftC1ZlA,9
+semantio-0.0.7.dist-info/RECORD,,