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
fix: index elements inside cross-origin iframes reliably
Saurav Panda3b8f2457b1f2
diff --git a/browser_use/browser/session.py b/browser_use/browser/session.pyindex 0e0f62bc0..a64b90f52 100644--- a/browser_use/browser/session.py+++ b/browser_use/browser/session.py@@ -3531,7 +3531,10 @@ class BrowserSession(BaseModel):continue # Skip if no session availableelse:# Get cached session for this target (don't change focus - iterating frames)- cdp_session = await self.get_or_create_cdp_session(target_id, focus=False)+ try:+ cdp_session = await self.get_or_create_cdp_session(target_id, focus=False)+ except ValueError:+ continue # Target may have detached between discovery and session creationif cdp_session:target_sessions[target_id] = cdp_session.session_iddiff --git a/browser_use/browser/session_manager.py b/browser_use/browser/session_manager.pyindex 57ef256c0..b380dabed 100644--- a/browser_use/browser/session_manager.py+++ b/browser_use/browser/session_manager.py@@ -401,6 +401,8 @@ class SessionManager:if '-32001' not in error_str and 'Session with given id not found' not in error_str:self.logger.debug(f'[SessionManager] Auto-attach failed for {target_type}: {e}')+ from browser_use.browser.session import Target+async with self._lock:# Track this session for the targetif target_id not in self._target_sessions:@@ -409,23 +411,22 @@ class SessionManager:self._target_sessions[target_id].add(session_id)self._session_to_target[session_id] = target_id- # Create or update Target (source of truth for url/title)- if target_id not in self._targets:- from browser_use.browser.session import Target-- target = Target(- target_id=target_id,- target_type=target_type,- url=target_info.get('url', 'about:blank'),- title=target_info.get('title', 'Unknown title'),- )- self._targets[target_id] = target- self.logger.debug(f'[SessionManager] Created target {target_id[:8]}... (type={target_type})')- else:- # Update existing target info- existing_target = self._targets[target_id]- existing_target.url = target_info.get('url', existing_target.url)- existing_target.title = target_info.get('title', existing_target.title)+ # Create or update Target inside the same lock so that get_target() is never+ # called in the window between _target_sessions being set and _targets being set.+ if target_id not in self._targets:+ target = Target(+ target_id=target_id,+ target_type=target_type,+ url=target_info.get('url', 'about:blank'),+ title=target_info.get('title', 'Unknown title'),+ )+ self._targets[target_id] = target+ self.logger.debug(f'[SessionManager] Created target {target_id[:8]}... (type={target_type})')+ else:+ # Update existing target info+ existing_target = self._targets[target_id]+ existing_target.url = target_info.get('url', existing_target.url)+ existing_target.title = target_info.get('title', existing_target.title)# Create CDPSession (communication channel)from browser_use.browser.session import CDPSessiondiff --git a/browser_use/dom/service.py b/browser_use/dom/service.pyindex 40826e378..53c305a13 100644--- a/browser_use/dom/service.py+++ b/browser_use/dom/service.py@@ -936,38 +936,57 @@ class DomService:# Use pre-fetched all_frames to find the iframe's target (no redundant CDP call)frame_id = node.get('frameId', None)++ # Fallback: if frameId is missing or not in all_frames, try URL matching via+ # the src attribute. This handles dynamically-injected iframes (e.g. HubSpot+ # popups, chat widgets) where Chrome hasn't yet registered the frameId in the+ # frame tree at DOM-snapshot time.+ if (not frame_id or frame_id not in all_frames) and attributes:+ src = attributes.get('src', '')+ if src:+ src_base = src.split('?')[0].rstrip('/')+ for fid, finfo in all_frames.items():+ frame_url = finfo.get('url', '').split('?')[0].rstrip('/')+ if frame_url and frame_url == src_base:+ frame_id = fid+ self.logger.debug(f'Matched cross-origin iframe by src URL: {src!r} -> frameId={fid}')+ break++ iframe_document_target = Noneif frame_id:frame_info = all_frames.get(frame_id)- iframe_document_target = Noneif frame_info and frame_info.get('frameTargetId'):iframe_target_id = frame_info['frameTargetId']+ # Use frameTargetId directly from all_frames — get_all_frames() already+ # validated connectivity. Do NOT gate on session_manager.get_target():+ # there is a race where _target_sessions is set (inside the lock in+ # _handle_target_attached) before _targets is populated (outside the+ # lock), so get_target() can transiently return None for a live target.iframe_target = self.browser_session.session_manager.get_target(iframe_target_id)- if iframe_target:- iframe_document_target = {- 'targetId': iframe_target.target_id,- 'url': iframe_target.url,- 'title': iframe_target.title,- 'type': iframe_target.target_type,- }- else:- iframe_document_target = None+ iframe_document_target = {+ 'targetId': iframe_target_id,+ 'url': iframe_target.url if iframe_target else frame_info.get('url', ''),+ 'title': iframe_target.title if iframe_target else frame_info.get('title', ''),+ 'type': iframe_target.target_type if iframe_target else 'iframe',+ }+# if target actually exists in one of the frames, just recursively build the dom tree for itif iframe_document_target:self.logger.debug(f'Getting content document for iframe {node.get("frameId", None)} at depth {iframe_depth + 1}')- content_document, _ = await self.get_dom_tree(- target_id=iframe_document_target['targetId'],- all_frames=all_frames,- # TODO: experiment with this values -> not sure whether the whole cross origin iframe should be ALWAYS included as soon as some part of it is visible or not.- # Current config: if the cross origin iframe is AT ALL visible, then just include everything inside of it!- # initial_html_frames=updated_html_frames,- initial_total_frame_offset=total_frame_offset,- iframe_depth=iframe_depth + 1,- )-- dom_tree_node.content_document = content_document- dom_tree_node.content_document.parent_node = dom_tree_node+ try:+ content_document, _ = await self.get_dom_tree(+ target_id=iframe_document_target['targetId'],+ all_frames=all_frames,+ # Current config: if the cross origin iframe is AT ALL visible, include everything inside it+ initial_total_frame_offset=total_frame_offset,+ iframe_depth=iframe_depth + 1,+ )+ dom_tree_node.content_document = content_document+ dom_tree_node.content_document.parent_node = dom_tree_node+ except Exception as e:+ self.logger.debug(f'Failed to get DOM tree for cross-origin iframe {frame_id}: {e}')return dom_tree_node