Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.
Python ai-agentsai-toolsbrowser-agentbrowser-automationbrowser-usecdpcloud-browsercomputer-useheadless-browserllmmcpplaywrightpythonremote-browserstealth-browserweb-automationweb-scraping
update available file paths on pdf download and fix some bugs in pdf detection/download. mostly coded by claude-opus-4.1
mertunsallaef48b7fa226
diff --git a/browser_use/browser/PLAYWRIGHT_MIGRATION_EXPLAINED.md b/browser_use/browser/PLAYWRIGHT_MIGRATION_EXPLAINED.mdnew file mode 100644index 000000000..7bc98524e--- /dev/null+++ b/browser_use/browser/PLAYWRIGHT_MIGRATION_EXPLAINED.md@@ -0,0 +1,166 @@+# Browser Session Refactor: From Playwright to Pure CDP++## Overview+This PR represents a **massive architectural shift** from Playwright-based browser automation to pure Chrome DevTools Protocol (CDP) with an event-driven architecture.++## Key Architectural Changes++### Old Architecture (Playwright-based)+- **Library**: Used Playwright for browser automation+- **Style**: Imperative, direct method calls+- **State Management**: Instance variables and decorators+- **Example**: `await page.goto(url)`, `await page.click(selector)`+- **File**: `old_session.py` (4856 lines)++### New Architecture (Pure CDP + Event-driven)+- **Library**: Direct CDP communication via `cdp_use` library+- **Style**: Event-driven with EventBus pattern+- **State Management**: Watchdogs + Events+- **Example**: `event_bus.dispatch(NavigateToUrlEvent(url=url))`+- **File**: `session.py` (1606 lines, much cleaner!)++## Core Components++### 1. EventBus (`bubus` library)+Central event system for all communication:+```python+event_bus = EventBus()+event_bus.on(EventType, handler)+event_bus.dispatch(EventInstance)+```++### 2. Watchdogs Pattern+Modular components that listen to specific events and perform actions:+- **CrashWatchdog**: Monitors for browser crashes+- **DownloadsWatchdog**: Handles file downloads and PDF auto-downloads+- **SecurityWatchdog**: Manages security policies+- **StorageStateWatchdog**: Handles cookies and storage+- **LocalBrowserWatchdog**: Manages local browser lifecycle+- **AboutBlankWatchdog**: Handles about:blank navigation+- **DOMWatchdog**: Manages DOM state+- **ScreenshotWatchdog**: Handles screenshots+- **DefaultActionWatchdog**: Provides default behaviors++Each watchdog:+- Inherits from `BaseWatchdog`+- Declares what events it listens to (`LISTENS_TO`)+- Declares what events it emits (`EMITS`)+- Has handlers like `on_EventName()`++### 3. CDPSession+Manages CDP connections to browser targets (tabs):+```python+class CDPSession:+ cdp_client: CDPClient # WebSocket connection to browser+ target_id: str # Browser tab/target ID+ session_id: str # CDP session ID+```++### 4. Event Types+Events flow through the system to trigger actions:+- **Lifecycle**: `BrowserStartEvent`, `BrowserStopEvent`, `BrowserStoppedEvent`+- **Navigation**: `NavigateToUrlEvent`, `NavigationStartedEvent`, `NavigationCompleteEvent`+- **Tabs**: `TabCreatedEvent`, `TabClosedEvent`, `SwitchTabEvent`+- **Downloads**: `FileDownloadedEvent`+- **State**: `BrowserStateRequestEvent`, `AgentFocusChangedEvent`+- **Errors**: `BrowserErrorEvent`, `BrowserCrashedEvent`++## PDF Auto-Download Feature Implementation++### What We Added:++1. **Configuration** (`profile.py`):+ ```python+ auto_download_pdfs: bool = Field(+ default=True,+ description='Automatically download PDFs when navigating to PDF viewer pages.'+ )+ ```++2. **Session-level Download Tracking** (`session.py`):+ ```python+ # Track files downloaded during this session+ _downloaded_files: list[str] = PrivateAttr(default_factory=list)++ # Listen to download events+ self.event_bus.on(FileDownloadedEvent, self.on_FileDownloadedEvent)++ async def on_FileDownloadedEvent(self, event: FileDownloadedEvent):+ if event.path and event.path not in self._downloaded_files:+ self._downloaded_files.append(event.path)+ ```++3. **Downloads Watchdog** (`downloads_watchdog.py`):+ - Checks for PDF viewers on navigation: `check_for_pdf_viewer()`+ - Auto-downloads PDFs if enabled: `trigger_pdf_download()`+ - Uses JavaScript fetch to download from browser cache+ - Dispatches `FileDownloadedEvent` when complete++4. **Agent Integration** (`agent/service.py`):+ - `_check_and_update_downloads()` - Updates available files each turn+ - Accesses `browser_session.downloaded_files` property+ - Makes downloaded files available to agent actions++## Event Flow Example++When navigating to a PDF:+1. User/Agent dispatches `NavigateToUrlEvent`+2. Browser navigates to URL+3. `NavigationCompleteEvent` is dispatched+4. `DownloadsWatchdog` receives event, checks if page is PDF viewer+5. If PDF and `auto_download_pdfs=True`, downloads the PDF+6. `FileDownloadedEvent` is dispatched+7. `BrowserSession` receives event, adds to `_downloaded_files` list+8. Agent's `_check_and_update_downloads()` picks up new file++## Benefits of New Architecture++1. **Modularity**: Each concern is isolated in its own watchdog+2. **Testability**: Events can be easily mocked and tested+3. **Extensibility**: New features = new watchdog + events+4. **Debugging**: Event flow can be traced and logged+5. **Async-first**: Better concurrency with event-driven design+6. **Smaller Core**: Session.py is 1/3 the size of old_session.py+7. **No Playwright Dependency**: Direct CDP = more control, less overhead++## Migration Notes++### Breaking Changes:+- No more Playwright `Page`, `Browser`, `BrowserContext` objects+- Methods like `page.goto()` replaced with events+- Decorators like `@require_healthy_browser` removed++### Compatibility Layer:+The new `BrowserSession` provides some backwards-compatible methods:+- `downloaded_files` property (though implementation changed)+- `get_browser_state_summary()` still works+- CDP-based replacements for common operations++### What's Still TODO:+- Some error handling edge cases+- Performance optimizations for event dispatching+- More comprehensive test coverage+- Documentation for watchdog development++## Testing++Run the test script to verify PDF auto-download:+```bash+python test_pdf_download.py+```++This tests:+1. PDF auto-download when enabled+2. No auto-download when disabled+3. Downloaded files tracking in session+4. Agent's available_file_paths updates++## Conclusion++This refactor is a fundamental architectural change that makes browser-use more:+- **Maintainable**: Clear separation of concerns+- **Scalable**: Event-driven architecture scales better+- **Debuggable**: Event flow is traceable+- **Extensible**: Easy to add new features via watchdogs++The PDF auto-download feature was successfully ported to work within this new event-driven paradigm, demonstrating how features should be implemented going forward.\ No newline at end of filediff --git a/browser_use/browser/downloads_watchdog.py b/browser_use/browser/downloads_watchdog.pyindex 63652d993..aedb1a16d 100644--- a/browser_use/browser/downloads_watchdog.py+++ b/browser_use/browser/downloads_watchdog.py@@ -52,12 +52,12 @@ class DownloadsWatchdog(BaseWatchdog):_cdp_event_tasks: set[asyncio.Task] = PrivateAttr(default_factory=set) # Track CDP event handler tasksasync def on_BrowserLaunchEvent(self, event: BrowserLaunchEvent) -> None:- self.logger.info(f'[DownloadsWatchdog] Received BrowserLaunchEvent, EventBus ID: {id(self.event_bus)}')+ self.logger.debug(f'[DownloadsWatchdog] Received BrowserLaunchEvent, EventBus ID: {id(self.event_bus)}')# Ensure downloads directory existsdownloads_path = self.browser_session.browser_profile.downloads_pathif downloads_path:Path(downloads_path).mkdir(parents=True, exist_ok=True)- self.logger.info(f'[DownloadsWatchdog] Ensured downloads directory exists: {downloads_path}')+ self.logger.debug(f'[DownloadsWatchdog] Ensured downloads directory exists: {downloads_path}')async def on_TabCreatedEvent(self, event: TabCreatedEvent) -> None:"""Monitor new tabs for downloads."""@@ -100,21 +100,34 @@ class DownloadsWatchdog(BaseWatchdog):async def on_NavigationCompleteEvent(self, event: NavigationCompleteEvent) -> None:"""Check for PDFs after navigation completes."""+ self.logger.debug(f'[DownloadsWatchdog] NavigationCompleteEvent received for {event.url}, tab_index={event.tab_index}')+# Clear PDF cache for the navigated URL since content may have changedif event.url in self._pdf_viewer_cache:del self._pdf_viewer_cache[event.url]# Check if auto-download is enabled- if not self._is_auto_download_enabled():+ auto_download_enabled = self._is_auto_download_enabled()+ if not auto_download_enabled:+ self.logger.debug('[DownloadsWatchdog] Skipping PDF check - auto-download disabled')returntarget_id = await self.browser_session.get_target_id_by_tab_index(event.tab_index)- if target_id and await self.check_for_pdf_viewer(target_id):- self.logger.info(f'[DownloadsWatchdog] PDF detected after navigation to {event.url}')- await self.trigger_pdf_download(target_id)+ self.logger.debug(f'[DownloadsWatchdog] Got target_id={target_id} for tab_index={event.tab_index}')++ if target_id:+ is_pdf = await self.check_for_pdf_viewer(target_id)+ if is_pdf:+ self.logger.info(f'[DownloadsWatchdog] π PDF detected at {event.url}, triggering auto-download...')+ download_path = await self.trigger_pdf_download(target_id)+ if not download_path:+ self.logger.warning(f'[DownloadsWatchdog] β οΈ PDF download failed for {event.url}')+ else:+ self.logger.debug(f'[DownloadsWatchdog] Could not get target_id for tab_index={event.tab_index}')def _is_auto_download_enabled(self) -> bool:- return True # TODO: move this to browser_profile config option+ """Check if auto-download PDFs is enabled in browser profile."""+ return self.browser_session.browser_profile.auto_download_pdfsasync def attach_to_target(self, target_id: str) -> None:"""Set up download monitoring for a specific target."""@@ -231,14 +244,14 @@ class DownloadsWatchdog(BaseWatchdog):expected_path = Path(downloads_dir) / suggested_filename# Debug: List current directory contents- self.logger.info(f'[DownloadsWatchdog] Downloads directory: {downloads_dir}')+ self.logger.debug(f'[DownloadsWatchdog] Downloads directory: {downloads_dir}')if Path(downloads_dir).exists():files_before = list(Path(downloads_dir).iterdir())- self.logger.info(f'[DownloadsWatchdog] Files before download: {[f.name for f in files_before]}')+ self.logger.debug(f'[DownloadsWatchdog] Files before download: {[f.name for f in files_before]}')# Browser.setDownloadBehavior doesn't work reliably with CDP connections# So we'll download the file manually using JavaScript fetch- self.logger.info(f'[DownloadsWatchdog] Downloading file manually via fetch: {download_url}')+ self.logger.debug(f'[DownloadsWatchdog] Downloading file manually via fetch: {download_url}')try:# Escape the URL for JavaScript@@ -330,7 +343,7 @@ class DownloadsWatchdog(BaseWatchdog):"""Handle a download event."""download_id = f'{id(download)}'self._active_downloads[download_id] = download- self.logger.info(f'[DownloadsWatchdog] β¬οΈ Handling download: {download.suggested_filename} from {download.url[:100]}...')+ self.logger.debug(f'[DownloadsWatchdog] β¬οΈ Handling download: {download.suggested_filename} from {download.url[:100]}...')# Debug: Check if download is already being handled elsewherefailure = (@@ -356,7 +369,7 @@ class DownloadsWatchdog(BaseWatchdog):# Check if Playwright already auto-downloaded the file (due to CDP setup)original_path = Path(downloads_dir) / suggested_filenameif original_path.exists() and original_path.stat().st_size > 0:- self.logger.info(+ self.logger.debug(f'[DownloadsWatchdog] File already downloaded by Playwright: {original_path} ({original_path.stat().st_size} bytes)')@@ -370,18 +383,18 @@ class DownloadsWatchdog(BaseWatchdog):unique_filename = await self._get_unique_filename(downloads_dir, suggested_filename)download_path = Path(downloads_dir) / unique_filename- self.logger.info(f'[DownloadsWatchdog] Download started: {unique_filename} from {url[:100]}...')+ self.logger.debug(f'[DownloadsWatchdog] Download started: {unique_filename} from {url[:100]}...')current_step = 'calling_save_as'# Save the download using Playwright's save_as method- self.logger.info(f'[DownloadsWatchdog] Saving download to: {download_path}')- self.logger.info(f'[DownloadsWatchdog] Download path exists: {download_path.parent.exists()}')- self.logger.info(f'[DownloadsWatchdog] Download path writable: {os.access(download_path.parent, os.W_OK)}')+ self.logger.debug(f'[DownloadsWatchdog] Saving download to: {download_path}')+ self.logger.debug(f'[DownloadsWatchdog] Download path exists: {download_path.parent.exists()}')+ self.logger.debug(f'[DownloadsWatchdog] Download path writable: {os.access(download_path.parent, os.W_OK)}')try:- self.logger.info('[DownloadsWatchdog] About to call download.save_as()...')+ self.logger.debug('[DownloadsWatchdog] About to call download.save_as()...')await download.save_as(str(download_path))- self.logger.info(f'[DownloadsWatchdog] Successfully saved download to: {download_path}')+ self.logger.debug(f'[DownloadsWatchdog] Successfully saved download to: {download_path}')current_step = 'save_as_completed'except Exception as save_error:self.logger.error(f'[DownloadsWatchdog] save_as() failed with error: {save_error}')@@ -418,7 +431,7 @@ class DownloadsWatchdog(BaseWatchdog):)self.logger.info(- f'[DownloadsWatchdog] Download completed: {suggested_filename} ({file_size} bytes) saved to {download_path}'+ f'[DownloadsWatchdog] β Download completed: {suggested_filename} ({file_size} bytes) saved to {download_path}')# File is now tracked on filesystem, no need to track in memory@@ -440,18 +453,23 @@ class DownloadsWatchdog(BaseWatchdog):Returns True if a PDF is detected and should be downloaded."""+ self.logger.debug(f'[DownloadsWatchdog] Checking if target {target_id} is PDF viewer...')+# Get target info to get URLcdp_client = self.browser_session.cdp_clienttargets = await cdp_client.send.Target.getTargets()target_info = next((t for t in targets['targetInfos'] if t['targetId'] == target_id), None)if not target_info:+ self.logger.warning(f'[DownloadsWatchdog] No target info found for {target_id}')return Falsepage_url = target_info.get('url', '')# Check cache firstif page_url in self._pdf_viewer_cache:- return self._pdf_viewer_cache[page_url]+ cached_result = self._pdf_viewer_cache[page_url]+ self.logger.debug(f'[DownloadsWatchdog] Using cached PDF check result for {page_url}: {cached_result}')+ return cached_resulttry:# Create a temporary CDP session for this target without switching focus@@ -467,10 +485,11 @@ class DownloadsWatchdog(BaseWatchdog):// Check for Chrome's built-in PDF viewer (both old and new selectors)const pdfEmbed = document.querySelector('embed[type="application/x-google-chrome-pdf"]') ||document.querySelector('embed[type="application/pdf"]');- if (pdfEmbed && pdfEmbed.src) {+ if (pdfEmbed) {+ // For Chrome PDF viewer, use window.location.href not embed.src (which is often about:blank)return {isPdf: true,- url: pdfEmbed.src,+ url: window.location.href,isChromePdfViewer: true};}@@ -526,9 +545,9 @@ class DownloadsWatchdog(BaseWatchdog):is_pdf_viewer = result.get('result', {}).get('value', {})if is_pdf_viewer.get('isPdf', False):- self.logger.info(- f'[DownloadsWatchdog] π PDF detected: {is_pdf_viewer.get("url", "unknown")} '- f'(type: {"Chrome viewer" if is_pdf_viewer.get("isChromePdfViewer") else "direct PDF"})'+ self.logger.debug(+ f'[DownloadsWatchdog] PDF detected: {is_pdf_viewer.get("url", "unknown")} '+ f'(type: {"Chrome viewer" if is_pdf_viewer.get("isChromePdfViewer") else "direct PDF" if is_pdf_viewer.get("isDirectPdf") else "PDF URL" if is_pdf_viewer.get("isPdfUrl") else "iframe PDF"})')self._pdf_viewer_cache[page_url] = Truereturn True@@ -550,14 +569,20 @@ class DownloadsWatchdog(BaseWatchdog):Returns the download path if successful, None otherwise."""+ self.logger.debug(f'[DownloadsWatchdog] trigger_pdf_download called for target_id={target_id}')+if not self.browser_session.browser_profile.downloads_path:self.logger.warning('[DownloadsWatchdog] β No downloads path configured, cannot save PDF download')return None+ downloads_path = self.browser_session.browser_profile.downloads_path+ self.logger.debug(f'[DownloadsWatchdog] Downloads path: {downloads_path}')+try:# Create a temporary CDP session for this target without switching focusimport asyncio+ self.logger.debug(f'[DownloadsWatchdog] Creating CDP session for PDF download from target {target_id}')temp_session = await self.browser_session.get_or_create_cdp_session(target_id, focus=False)# Try to get the PDF URL with timeout@@ -566,11 +591,15 @@ class DownloadsWatchdog(BaseWatchdog):params={'expression': """(() => {+ // For Chrome's PDF viewer, the actual URL is in window.location.href+ // The embed element's src is often "about:blank"const embedElement = document.querySelector('embed[type="application/x-google-chrome-pdf"]') ||document.querySelector('embed[type="application/pdf"]');- if (embedElement && embedElement.src) {- return { url: embedElement.src };+ if (embedElement) {+ // Chrome PDF viewer detected - use the page URL+ return { url: window.location.href };}+ // Fallback to window.location.href anywayreturn { url: window.location.href };})()""",@@ -594,16 +623,18 @@ class DownloadsWatchdog(BaseWatchdog):pdf_filename = os.path.basename(parsed.path) or 'document.pdf'if not pdf_filename.endswith('.pdf'):pdf_filename += '.pdf'++ self.logger.debug(f'[DownloadsWatchdog] Generated filename: {pdf_filename}')# Check if already downloaded by looking in the downloads directorydownloads_dir = str(self.browser_session.browser_profile.downloads_path)if os.path.exists(downloads_dir):existing_files = os.listdir(downloads_dir)if pdf_filename in existing_files:- self.logger.debug(f'[DownloadsWatchdog] β PDF already downloaded: {pdf_filename}')+ self.logger.debug(f'[DownloadsWatchdog] PDF already downloaded: {pdf_filename}')return None- self.logger.info(f'[DownloadsWatchdog] β¬οΈ Downloading PDF file from: {pdf_url[:100]}...')+ self.logger.debug(f'[DownloadsWatchdog] Starting PDF download from: {pdf_url[:100]}...')# Download using JavaScript fetch to leverage browser cachetry:@@ -629,10 +660,7 @@ class DownloadsWatchdog(BaseWatchdog):// Check if served from cacheconst fromCache = response.headers.has('age') ||- !response.headers.has('date') ||- performance.getEntriesByName({escaped_pdf_url}).some(entry =>- entry.transferSize === 0 || entry.transferSize < entry.encodedBodySize- );+ !response.headers.has('date');return {{data: Array.from(uint8Array),@@ -657,6 +685,8 @@ class DownloadsWatchdog(BaseWatchdog):if download_result and download_result.get('data') and len(download_result['data']) > 0:# Ensure unique filenamedownloads_dir = str(self.browser_session.browser_profile.downloads_path)+ # Ensure downloads directory exists+ os.makedirs(downloads_dir, exist_ok=True)unique_filename = await self._get_unique_filename(downloads_dir, pdf_filename)download_path = os.path.join(downloads_dir, unique_filename)@@ -664,7 +694,13 @@ class DownloadsWatchdog(BaseWatchdog):async with await anyio.open_file(download_path, 'wb') as f:await f.write(bytes(download_result['data']))- # File is now tracked on filesystem, no need to track in memory+ # Verify file was written successfully+ if os.path.exists(download_path):+ actual_size = os.path.getsize(download_path)+ self.logger.debug(f'[DownloadsWatchdog] PDF file written successfully: {download_path} ({actual_size} bytes)')+ else:+ self.logger.error(f'[DownloadsWatchdog] β Failed to write PDF file to: {download_path}')+ return None# Log cache informationcache_status = 'from cache' if download_result.get('fromCache') else 'from network'@@ -674,6 +710,7 @@ class DownloadsWatchdog(BaseWatchdog):)# Emit file downloaded event+ self.logger.debug(f'[DownloadsWatchdog] Dispatching FileDownloadedEvent for {unique_filename}')self.event_bus.dispatch(FileDownloadedEvent(url=pdf_url,diff --git a/browser_use/browser/profile.py b/browser_use/browser/profile.pyindex ded78c8e0..a82679729 100644--- a/browser_use/browser/profile.py+++ b/browser_use/browser/profile.py@@ -603,6 +603,12 @@ class BrowserProfile(BrowserConnectArgs, BrowserLaunchPersistentContextArgs, Broinclude_dynamic_attributes: bool = Field(default=True, description='Include dynamic attributes in selectors.')highlight_elements: bool = Field(default=True, description='Highlight interactive elements on the page.')viewport_expansion: int = Field(default=500, description='Viewport expansion in pixels for LLM context.')++ # --- Downloads ---+ auto_download_pdfs: bool = Field(+ default=True,+ description='Automatically download PDFs when navigating to PDF viewer pages.'+ )profile_directory: str = 'Default' # e.g. 'Profile 1', 'Profile 2', 'Custom Profile', etc.diff --git a/browser_use/browser/session.py b/browser_use/browser/session.pyindex 13098cbe7..285722cf4 100644--- a/browser_use/browser/session.py+++ b/browser_use/browser/session.py@@ -23,6 +23,7 @@ from browser_use.browser.events import (BrowserStateRequestEvent,BrowserStopEvent,BrowserStoppedEvent,+ FileDownloadedEvent,NavigateToUrlEvent,NavigationCompleteEvent,NavigationStartedEvent,@@ -186,6 +187,7 @@ class BrowserSession(BaseModel):_cdp_session_pool: dict[str, CDPSession] = PrivateAttr(default_factory=dict)_cached_browser_state_summary: Any = PrivateAttr(default=None)_cached_selector_map: dict[int, EnhancedDOMTreeNode] = PrivateAttr(default_factory=dict)+ _downloaded_files: list[str] = PrivateAttr(default_factory=list) # Track files downloaded during this session# Watchdogs_crash_watchdog: Any | None = PrivateAttr(default=None)@@ -233,6 +235,7 @@ class BrowserSession(BaseModel):self._cdp_client_root = None # type: ignoreself._cached_browser_state_summary = Noneself._cached_selector_map.clear()+ self._downloaded_files.clear()self.agent_focus = Noneif self.is_local:@@ -268,6 +271,8 @@ class BrowserSession(BaseModel):self.event_bus.on(NavigateToUrlEvent, self.on_NavigateToUrlEvent)self.event_bus.on(SwitchTabEvent, self.on_SwitchTabEvent)self.event_bus.on(TabClosedEvent, self.on_TabClosedEvent)+ # Track downloaded files+ self.event_bus.on(FileDownloadedEvent, self.on_FileDownloadedEvent)async def start(self) -> None:"""Start the browser session."""@@ -421,6 +426,7 @@ class BrowserSession(BaseModel):await asyncio.sleep(0.5)# Dispatch navigation complete+ self.logger.debug(f'Dispatching NavigationCompleteEvent for {event.url} (tab_index={tab_index})')self.event_bus.dispatch(NavigationCompleteEvent(tab_index=tab_index,@@ -511,6 +517,18 @@ class BrowserSession(BaseModel):else:self.agent_focus = None+ async def on_FileDownloadedEvent(self, event: FileDownloadedEvent) -> None:+ """Track downloaded files during this session."""+ self.logger.debug(f'FileDownloadedEvent received: {event.file_name} at {event.path}')+ if event.path and event.path not in self._downloaded_files:+ self._downloaded_files.append(event.path)+ self.logger.info(f'π Tracked download: {event.file_name} ({len(self._downloaded_files)} total downloads in session)')+ else:+ if not event.path:+ self.logger.warning(f'FileDownloadedEvent has no path: {event}')+ else:+ self.logger.debug(f'File already tracked: {event.path}')+async def on_BrowserStopEvent(self, event: BrowserStopEvent) -> None:"""Handle browser stop request."""@@ -693,6 +711,8 @@ class BrowserSession(BaseModel):# self.event_bus.on(BrowserStoppedEvent, self._downloads_watchdog.on_BrowserStoppedEvent)# self.event_bus.on(NavigationCompleteEvent, self._downloads_watchdog.on_NavigationCompleteEvent)await self._downloads_watchdog.attach_to_session()+ if self.browser_profile.auto_download_pdfs:+ self.logger.info('π PDF auto-download enabled for this session')# Initialize StorageStateWatchdogStorageStateWatchdog.model_rebuild()@@ -1109,17 +1129,12 @@ class BrowserSession(BaseModel):@propertydef downloaded_files(self) -> list[str]:- """Get list of downloaded files from the downloads directory."""- if not self.browser_profile.downloads_path:- return []-- downloads_dir = Path(self.browser_profile.downloads_path)- if not downloads_dir.exists():- return []-- # Get all files in downloads directory (not directories)- files = [str(f) for f in downloads_dir.iterdir() if f.is_file()]- return sorted(files)+ """Get list of files downloaded during this browser session.++ Returns:+ list[str]: List of absolute file paths to downloaded files in this session+ """+ return self._downloaded_files.copy()# ========== CDP-based replacements for browser_context operations ==========