semantio 0.0.5__tar.gz → 0.0.7__tar.gz

{semantio-0.0.5 → semantio-0.0.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: semantio
-Version: 0.0.5
+Version: 0.0.7
 Summary: A powerful SDK for building AI agents
 Home-page: https://github.com/Syenah/semantio
 Author: Rakesh

semantio-0.0.7/semantio/__init__.py

@@ -0,0 +1,4 @@
+import warnings
+
+# Suppress all warnings globally
+warnings.simplefilter("ignore")

semantio-0.0.7/semantio/tools/web_browser.py

@@ -0,0 +1,545 @@
+# web_browser.py
+from typing import Dict, Any, List, Optional, Callable
+from pydantic import Field, BaseModel
+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):
+    tasks: List[Dict[str, Any]] = Field(
+        ...,
+        description="List of automation tasks to execute"
+    )
+
+class WebBrowserTool(BaseTool):
+    name: str = Field("WebBrowser", description="Name of the tool")
+    description: str = Field(
+        "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 = 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 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", 15)) * 1000
+            self.max_retries = int(input.get("max_retries", self.max_retries))
+            plan = self._generate_plan(input.get("query", ""), current_url)
+            if not plan.tasks:
+                raise ValueError("No valid tasks in the generated plan.")
+
+            # 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:
+                self._dismiss_unwanted_modals(page, task_context=task.get("description", ""))
+                action = task.get("action", "").lower()
+                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({
+                        "action": action,
+                        "success": False,
+                        "message": f"Unsupported action: {action}"
+                    })
+                    continue
+
+                result = self._execute_with_retries(page, task, handler, executed_context)
+                elapsed = time.time() - start_time
+                result["elapsed"] = elapsed
+                self.logger.info(f"Action '{action}' completed in {elapsed:.2f} seconds.")
+                results.append(result)
+
+                if not result.get("success", False):
+                    self.logger.error(f"Task failed: {result.get('message')}")
+                    self._capture_failure_screenshot(page, action)
+                    break
+
+                current_url = page.url
+
+            overall_elapsed = time.time() - overall_start
+            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:
+            self.logger.exception("Execution error:")
+            return {"status": "error", "message": str(e)}
+
+    def _generate_plan(self, query: str, current_url: str) -> BrowserPlan:
+        prompt = f"""Generate browser automation plan for: {query}
+
+Current URL: {current_url or 'No page loaded yet'}
+
+Required JSON format:
+{{
+    "tasks": [
+        {{
+            "action": "navigate|click|type|wait|wait_for_ajax|scroll|hover|screenshot|switch_tab|execute_script|drag_and_drop",
+            "selector": "CSS selector (optional)",
+            "value": "input text/URL/seconds/filename/target-selector",
+            "description": "action purpose"
+        }}
+    ]
+}}
+
+Guidelines:
+1. Prefer IDs in selectors (#element-id) and semantic attributes.
+2. Include wait steps after navigation and wait for AJAX where applicable.
+3. Dismiss any modals/pop-ups that are not part of the task.
+4. For drag_and_drop, use source selector in 'selector' and target selector in 'value'.
+5. For execute_script, 'value' should contain valid JavaScript.
+6. For switch_tab, 'value' should be an index or keyword 'new'.
+"""
+        response = self.llm.generate(prompt=prompt)
+        return self._parse_plan(response)
+
+    def _parse_plan(self, response: str) -> BrowserPlan:
+        try:
+            json_match = re.search(r'```json\n?(.+?)\n?```', response, re.DOTALL)
+            if json_match:
+                plan_data = json.loads(json_match.group(1).strip())
+            else:
+                json_str_match = re.search(r'\{.*\}', response, re.DOTALL)
+                if not json_str_match:
+                    raise ValueError("No JSON object found in the response.")
+                plan_data = json.loads(json_str_match.group())
+            validated_tasks = []
+            for task in plan_data.get("tasks", []):
+                if not all(key in task for key in ["action", "description"]):
+                    self.logger.warning(f"Skipping task due to missing keys: {task}")
+                    continue
+                validated_tasks.append({
+                    "action": task["action"],
+                    "selector": task.get("selector", ""),
+                    "value": task.get("value", ""),
+                    "description": task["description"]
+                })
+            return BrowserPlan(tasks=validated_tasks)
+        except (json.JSONDecodeError, AttributeError, ValueError) as e:
+            self.logger.error(f"Plan parsing failed: {e}")
+            return BrowserPlan(tasks=[])
+
+    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(page, task, handler)
+            if result.get("success", False):
+                return result
+            attempts += 1
+            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, page: Page, task: Dict[str, Any],
+                             handler: Callable[[Page, Dict[str, Any]], Dict[str, Any]]) -> Dict[str, Any]:
+        try:
+            return handler(page, task)
+        except Exception as e:
+            action = task.get("action", "unknown")
+            self.logger.exception(f"Error executing task '{action}':")
+            return {"action": action, "success": False, "message": f"Critical error: {str(e)}"}
+
+    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_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:
+            self.logger.error(f"Modal handling error: {e}")
+
+    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;
+            }});
+        }})();
+        """
+        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.
+        """
+        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:
+            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:
+            self.logger.error(f"Navigation to {url} failed: {e}")
+            return {"action": "navigate", "success": False, "message": f"Navigation failed: {str(e)}"}
+
+    def _handle_click(self, page: Page, selector: str) -> Dict[str, Any]:
+        try:
+            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:
+            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, page: Page, selector: str, text: str, task: Dict[str, Any]) -> Dict[str, Any]:
+        try:
+            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:
+            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]:
+        try:
+            wait_time = float(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:
+            self.logger.error(f"Invalid wait time provided: {seconds}")
+            return {"action": "wait", "success": False, "message": "Invalid wait time"}
+
+    def _handle_wait_for_ajax(self, page: Page, seconds: str) -> Dict[str, Any]:
+        try:
+            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 = 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:
+            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, page: Page, selector: str) -> Dict[str, Any]:
+        try:
+            if selector:
+                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:
+                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:
+            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, page: Page, selector: str) -> Dict[str, Any]:
+        try:
+            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:
+            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, page: Page, filename: str) -> Dict[str, Any]:
+        try:
+            page.screenshot(path=filename)
+            return {"action": "screenshot", "success": True, "message": f"Screenshot saved as {filename}"}
+        except Exception as 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, context, value: str) -> Dict[str, Any]:
+        try:
+            pages = context.pages
+            if value.lower() == "new":
+                target_page = pages[-1]
+            else:
+                idx = int(value)
+                if idx < len(pages):
+                    target_page = pages[idx]
+                else:
+                    return {"action": "switch_tab", "success": False, "message": f"Tab index {value} out of range"}
+            return {"action": "switch_tab", "success": True, "message": f"Switched to tab {value}"}
+        except Exception as 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, page: Page, script: str) -> Dict[str, Any]:
+        try:
+            result = page.evaluate(script)
+            return {"action": "execute_script", "success": True, "message": "Script executed successfully", "result": result}
+        except Exception as 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, page: Page, source_selector: str, target_selector: str) -> Dict[str, Any]:
+        try:
+            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:
+            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, page: Page, action: str):
+        filename = f"failure_{action}_{int(time.time())}.png"
+        try:
+            page.screenshot(path=filename)
+            self.logger.info(f"Failure screenshot captured: {filename}")
+        except Exception as e:
+            self.logger.error(f"Failed to capture screenshot: {e}")

{semantio-0.0.5 → semantio-0.0.7}/semantio.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: semantio
-Version: 0.0.5
+Version: 0.0.7
 Summary: A powerful SDK for building AI agents
 Home-page: https://github.com/Syenah/semantio
 Author: Rakesh

{semantio-0.0.5 → semantio-0.0.7}/semantio.egg-info/requires.txt

@@ -15,7 +15,6 @@ sentence-transformers
 fuzzywuzzy
 duckduckgo-search
 yfinance
-selenium
 beautifulsoup4
 webdriver-manager
 validators

{semantio-0.0.5 → semantio-0.0.7}/setup.py

@@ -2,7 +2,7 @@ from setuptools import setup, find_packages
 
 setup(
     name="semantio",
-    version="0.0.5",
+    version="0.0.7",
     description="A powerful SDK for building AI agents",
     long_description=open("README.md").read(),
     long_description_content_type="text/markdown",
@@ -28,7 +28,6 @@ setup(
         "fuzzywuzzy", # For fuzzy string matching
         "duckduckgo-search", # For DuckDuckGo search
         "yfinance",   # For stock/crypto prices
-        "selenium",   # For web scraping
         "beautifulsoup4", # For HTML parsing
         "webdriver-manager", # For browser automation
         "validators", # For URL validation

semantio-0.0.5/semantio/tools/web_browser.py

@@ -1,271 +0,0 @@
-# web_browser.py
-from typing import Dict, Any, List, Optional
-from pydantic import Field, BaseModel
-from selenium import webdriver
-from selenium.webdriver.common.by import By
-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
-from .base_tool import BaseTool
-
-logger = logging.getLogger(__name__)
-
-class BrowserPlan(BaseModel):
-    tasks: List[Dict[str, Any]] = Field(
-        ...,
-        description="List of automation tasks to execute"
-    )
-
-class WebBrowserTool(BaseTool):
-    name: str = Field("WebBrowser", description="Name of the tool")
-    description: str = Field(
-        "Universal web automation tool for dynamic website interactions",
-        description="Tool description"
-    )
-    
-    def execute(self, input: Dict[str, Any]) -> Dict[str, Any]:
-        """Execute dynamic web automation workflow"""
-        driver = None
-        try:
-            driver = self._init_browser(input.get("headless", False))
-            results = []
-            current_url = ""
-
-            # Generate initial plan
-            plan = self._generate_plan(input['query'], current_url)
-            
-            for task in plan.tasks:
-                result = self._execute_safe_task(driver, task)
-                results.append(result)
-                
-                if not result['success']:
-                    break
-                    
-                # Update context for next tasks
-                current_url = driver.current_url
-
-            return {"status": "success", "results": results}
-            
-        except Exception as e:
-            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 adaptive execution plan using LLM"""
-        prompt = f"""Generate browser automation plan for: {query}
-        
-        Current URL: {current_url or 'No page loaded yet'}
-        
-        Required JSON format:
-        {{
-            "tasks": [
-                {{
-                    "action": "navigate|click|type|wait|scroll",
-                    "selector": "CSS selector (optional)",
-                    "value": "input text/URL/seconds",
-                    "description": "action purpose"
-                }}
-            ]
-        }}
-        
-        Guidelines:
-        1. Prefer IDs in selectors (#element-id)
-        2. Use semantic attributes (aria-label, name)
-        3. Include wait steps after navigation
-        4. Prioritize visible elements
-        5. Add scroll steps for hidden elements
-        """
-        
-        response = self.llm.generate(prompt=prompt)
-        return self._parse_plan(response)
-
-    def _parse_plan(self, response: str) -> BrowserPlan:
-        """Robust JSON parsing with multiple fallback strategies"""
-        try:
-            # Try extracting JSON from markdown code block
-            json_match = re.search(r'```json\n?(.+?)\n?```', response, re.DOTALL)
-            if json_match:
-                plan_data = json.loads(json_match.group(1).strip())
-            else:
-                # Fallback to extract first JSON object
-                json_str = re.search(r'\{.*\}', response, re.DOTALL).group()
-                plan_data = json.loads(json_str)
-            
-            # Validate tasks structure
-            validated_tasks = []
-            for task in plan_data.get("tasks", []):
-                if not all(key in task for key in ["action", "description"]):
-                    continue
-                validated_tasks.append({
-                    "action": task["action"],
-                    "selector": task.get("selector", ""),
-                    "value": task.get("value", ""),
-                    "description": task["description"]
-                })
-            
-            return BrowserPlan(tasks=validated_tasks)
-            
-        except (json.JSONDecodeError, AttributeError) as e:
-            logger.error(f"Plan parsing failed: {e}")
-            return BrowserPlan(tasks=[])
-
-    def _execute_safe_task(self, driver, task: Dict) -> Dict[str, Any]:
-        """Execute task with comprehensive error handling"""
-        try:
-            action = task["action"].lower()
-            selector = task.get("selector", "")
-            value = task.get("value", "")
-            
-            if action == "navigate":
-                return self._handle_navigation(driver, value)
-                
-            elif action == "click":
-                return self._handle_click(driver, selector)
-                
-            elif action == "type":
-                return self._handle_typing(driver, selector, value)
-                
-            elif action == "wait":
-                return self._handle_wait(value)
-                
-            elif action == "scroll":
-                return self._handle_scroll(driver, selector)
-                
-            return {
-                "action": action,
-                "success": False,
-                "message": f"Unsupported action: {action}"
-            }
-            
-        except Exception as e:
-            return {
-                "action": action,
-                "success": False,
-                "message": f"Critical error: {str(e)}"
-            }
-
-    def _handle_navigation(self, driver, url: str) -> Dict[str, Any]:
-        """Smart navigation handler"""
-        if not url.startswith(("http://", "https://")):
-            url = f"https://{url}"
-            
-        try:
-            driver.get(url)
-            WebDriverWait(driver, 15).until(
-                EC.presence_of_element_located((By.TAG_NAME, "body"))
-            )
-            return {
-                "action": "navigate",
-                "success": True,
-                "message": f"Navigated to {url}"
-            }
-        except Exception as e:
-            return {
-                "action": "navigate",
-                "success": False,
-                "message": f"Navigation failed: {str(e)}"
-            }
-
-    def _handle_click(self, driver, selector: str) -> Dict[str, Any]:
-        """Dynamic click handler"""
-        try:
-            element = WebDriverWait(driver, 15).until(
-                EC.element_to_be_clickable((By.CSS_SELECTOR, selector))
-            )
-            driver.execute_script("arguments[0].scrollIntoView({behavior: 'smooth'});", element)
-            element.click()
-            return {
-                "action": "click",
-                "success": True,
-                "message": f"Clicked element: {selector}"
-            }
-        except Exception as e:
-            return {
-                "action": "click",
-                "success": False,
-                "message": f"Click failed: {str(e)}"
-            }
-
-    def _handle_typing(self, driver, selector: str, text: str) -> Dict[str, Any]:
-        """Universal typing handler"""
-        try:
-            element = WebDriverWait(driver, 15).until(
-                EC.presence_of_element_located((By.CSS_SELECTOR, selector))
-            )
-            element.clear()
-            element.send_keys(text)
-            return {
-                "action": "type",
-                "success": True,
-                "message": f"Typed '{text}' into {selector}"
-            }
-        except Exception as e:
-            return {
-                "action": "type",
-                "success": False,
-                "message": f"Typing failed: {str(e)}"
-            }
-
-    def _handle_wait(self, seconds: str) -> Dict[str, Any]:
-        """Configurable wait handler"""
-        try:
-            wait_time = float(seconds)
-            time.sleep(wait_time)
-            return {
-                "action": "wait",
-                "success": True,
-                "message": f"Waited {wait_time} seconds"
-            }
-        except ValueError:
-            return {
-                "action": "wait",
-                "success": False,
-                "message": "Invalid wait time"
-            }
-
-    def _handle_scroll(self, driver, selector: str) -> Dict[str, Any]:
-        """Smart scroll handler"""
-        try:
-            if selector:
-                element = WebDriverWait(driver, 15).until(
-                    EC.presence_of_element_located((By.CSS_SELECTOR, selector))
-                )
-                driver.execute_script("arguments[0].scrollIntoView({behavior: 'smooth'});", element)
-            else:
-                driver.execute_script("window.scrollTo(0, document.body.scrollHeight);")
-                
-            return {
-                "action": "scroll",
-                "success": True,
-                "message": f"Scrolled to {selector or 'page bottom'}"
-            }
-        except Exception as e:
-            return {
-                "action": "scroll",
-                "success": False,
-                "message": f"Scroll failed: {str(e)}"
-            }