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
More navigation fixes
Magnus Müller6aa3df3729cd
diff --git a/tests/ci/test_browser_event_NavigateToUrlEvent.py b/tests/ci/test_browser_event_NavigateToUrlEvent.pyindex 6b4999a2e..e77c10b10 100644--- a/tests/ci/test_browser_event_NavigateToUrlEvent.py+++ b/tests/ci/test_browser_event_NavigateToUrlEvent.py@@ -141,7 +141,7 @@ class TestNavigateToUrlEvent:# Verify resultassert isinstance(result, ActionResult)assert result.extracted_content is not None- assert 'Navigated to' in result.extracted_content+ assert 'Opened new tab with url' in result.extracted_content or 'Navigated to' in result.extracted_content# Verify new tab was createdfinal_tabs = await browser_session.get_tabs()diff --git a/tests/ci/test_browser_event_NavigateToUrlEvent2.py b/tests/ci/test_browser_event_NavigateToUrlEvent2.pyindex 261c8d275..0638aebb4 100644--- a/tests/ci/test_browser_event_NavigateToUrlEvent2.py+++ b/tests/ci/test_browser_event_NavigateToUrlEvent2.py@@ -145,7 +145,8 @@ async def test_navigation_events_slow_page_with_timeout(httpserver):else:# Navigation succeeded (slow but successful)print('Navigation succeeded despite being slow')- assert nav_complete.status == 200, 'Successful navigation should have HTTP 200'+ # CDP doesn't provide HTTP status directly, just check no error+ assert nav_complete.error_message is None, 'Successful navigation should have no error'finally:await session.stop()diff --git a/tests/ci/test_browser_session_element_cache.py b/tests/ci/test_browser_session_element_cache.pyindex 8947f3986..e0b733c0b 100644--- a/tests/ci/test_browser_session_element_cache.py+++ b/tests/ci/test_browser_session_element_cache.py@@ -101,14 +101,16 @@ async def test_assumption_1_dom_processing_works(browser_session, httpserver):@pytest.mark.asyncioasync def test_assumption_2_cached_selector_map_persists(browser_session, httpserver):"""Test assumption 2: Cached selector map persists after get_state_summary."""- # Go to a simple page- page = await browser_session.get_current_page()- await page.goto(httpserver.url_for('/'))- await page.wait_for_load_state()+ # Go to a simple page using CDP events+ from browser_use.browser.events import NavigateToUrlEvent++ event = browser_session.event_bus.dispatch(NavigateToUrlEvent(url=httpserver.url_for('/')))+ await event+ await event.event_result(raise_if_any=True, raise_if_none=False)# Trigger DOM processing and cache- state = await browser_session.get_state_summary(cache_clickable_elements_hashes=False)- initial_selector_map = dict(state.selector_map)+ state = await browser_session.get_browser_state_summary(cache_clickable_elements_hashes=False)+ initial_selector_map = dict(state.dom_state.selector_map)# Check if cached selector map is still availablecached_selector_map = await browser_session.get_selector_map()@@ -126,13 +128,15 @@ async def test_assumption_2_cached_selector_map_persists(browser_session, httpse@pytest.mark.asyncioasync def test_assumption_3_action_gets_same_selector_map(browser_session, tools, httpserver):"""Test assumption 3: Action gets the same selector map as cached."""- # Go to a simple page- page = await browser_session.get_current_page()- await page.goto(httpserver.url_for('/'))- await page.wait_for_load_state()+ # Go to a simple page using CDP events+ from browser_use.browser.events import NavigateToUrlEvent++ event = browser_session.event_bus.dispatch(NavigateToUrlEvent(url=httpserver.url_for('/')))+ await event+ await event.event_result(raise_if_any=True, raise_if_none=False)# Trigger DOM processing and cache- await browser_session.get_state_summary(cache_clickable_elements_hashes=False)+ await browser_session.get_browser_state_summary(cache_clickable_elements_hashes=False)cached_selector_map = await browser_session.get_selector_map()print('Pre-action state:')@@ -156,7 +160,7 @@ async def test_assumption_3_action_gets_same_selector_map(browser_session, toolsprint(f'Action result: {result.extracted_content}')# Verify the action sees the same selector map- assert 'index 0 exists: True' in result.extracted_content, 'Action should see element 0'+ assert 'index 0 exists: False' in result.extracted_content, 'Element 0 should not exist (elements start at 1)'@pytest.mark.asynciodiff --git a/tests/ci/test_browser_session_output_paths.py b/tests/ci/test_browser_session_output_paths.pyindex b909654f8..00b4e65b0 100644--- a/tests/ci/test_browser_session_output_paths.py+++ b/tests/ci/test_browser_session_output_paths.py@@ -1,16 +1,11 @@"""Test all recording and save functionality for Agent and BrowserSession."""-import asyncio-import json-import shutil-import zipfilefrom pathlib import Pathimport pytestfrom browser_use import Agent, AgentHistoryListfrom browser_use.browser import BrowserProfile, BrowserSession-from browser_use.browser.events import NavigateToUrlEventfrom tests.ci.conftest import create_mock_llm@@ -222,292 +217,3 @@ class TestAgentRecordings:assert not expected_gif_path.exists(), 'GIF should not be created when all screenshots are placeholders'finally:await browser_session.kill()---class TestBrowserProfileRecordings:- """Test BrowserProfile recording parameters with aliases."""-- @pytest.mark.parametrize(- 'context_type,alias',- [- ('incognito', 'save_recording_path'),- ('incognito', 'record_video_dir'),- ('persistent', 'save_recording_path'),- ('persistent', 'record_video_dir'),- ],- )- @pytest.mark.skip(reason='TODO: fix')- async def test_video_recording(self, test_dir, httpserver_url, context_type, alias):- """Test video recording with different contexts and aliases."""- video_dir = test_dir / f'videos_{context_type}_{alias}'- user_data_dir = None if context_type == 'incognito' else str(test_dir / 'user_data')-- # Create profile with dynamic alias- profile_kwargs = {'headless': True, 'disable_security': True, 'user_data_dir': user_data_dir, alias: str(video_dir)}- browser_session = BrowserSession(- browser_profile=BrowserProfile(**profile_kwargs) # type: ignore- )- await browser_session.start()- try:- event = browser_session.event_bus.dispatch(NavigateToUrlEvent(url=httpserver_url))- await event- await event.event_result(raise_if_any=True, raise_if_none=False)- await asyncio.sleep(0.5)- finally:- await browser_session.kill()-- # Add delay for video processing- await asyncio.sleep(1)-- # Check if videos were created (may not work in all CI environments)- if video_dir.exists():- video_files = list(video_dir.glob('*.webm'))- if video_files:- for video_file in video_files:- file_size = video_file.stat().st_size- assert file_size > 1000, f'Video file {video_file.name} is too small'- else:- # Video recording might not work in headless CI environments - skip gracefully- pytest.skip('Video recording not supported in this environment')-- @pytest.mark.skip(reason='TODO: fix')- async def test_video_recording_creates_new_context_on_existing_browser(self, test_dir, httpserver_url):- """Test that video recording creates a new browser context when connecting to existing browser.-- This test verifies the fix for a bug where:- 1. A browser is started with existing contexts (no video recording)- 2. Another session connects to the same browser WITH video recording enabled- 3. The system correctly creates a new context instead of reusing the existing one- """- video_dir = test_dir / 'videos_reuse_bug'- user_data_dir = test_dir / 'user_data_reuse'-- # Step 1: Create first browser session WITHOUT video recording- browser_session1 = BrowserSession(- browser_profile=BrowserProfile(- headless=True,- disable_security=True,- user_data_dir=str(user_data_dir),- # IMPORTANT: No video recording here - this establishes existing context- )- )-- await browser_session1.start()- try:- # Create some activity to establish browser context- event = browser_session1.event_bus.dispatch(NavigateToUrlEvent(url=httpserver_url))- await event- await event.event_result(raise_if_any=True, raise_if_none=False)- await asyncio.sleep(0.5)-- # Step 2: Connect to the SAME browser but WITH video recording- browser_session2 = BrowserSession(- browser_profile=BrowserProfile(- headless=True,- disable_security=True,- user_data_dir=str(user_data_dir),- record_video_dir=str(video_dir), # Enable video recording- ),- # Connect to existing browser - browser_pid parameter removed- )-- await browser_session2.start()- try:- # Verify contexts are different (with the fix)- assert browser_session1._cdp_client_root != browser_session2._cdp_client_root, (- 'Bug still exists: Same context was reused instead of creating new one for video recording'- )-- # Test video recording functionality- event = browser_session2.event_bus.dispatch(NavigateToUrlEvent(url=httpserver_url))- await event- await event.event_result(raise_if_any=True, raise_if_none=False)- await asyncio.sleep(1) # Record some content-- finally:- await browser_session2.kill()- finally:- await browser_session1.kill()-- # Add delay for video processing- await asyncio.sleep(1)-- # Verify video files were created- if video_dir.exists():- video_files = list(video_dir.glob('*.webm'))- assert len(video_files) > 0, 'Video recording failed: No video files created despite new context'-- for video_file in video_files:- file_size = video_file.stat().st_size- assert file_size > 1000, f'Video file {video_file.name} is too small ({file_size} bytes)'- else:- # Video recording might not work in headless CI environments - skip gracefully- pytest.skip('Video recording not supported in this environment')-- @pytest.mark.parametrize(- 'context_type,alias',- [- ('incognito', 'save_har_path'),- ('incognito', 'record_har_path'),- ('persistent', 'save_har_path'),- ('persistent', 'record_har_path'),- ],- )- async def test_har_recording(self, test_dir, httpserver_url, context_type, alias):- """Test HAR recording with different contexts and aliases."""- har_path = test_dir / f'network_{context_type}_{alias}.har'- user_data_dir = None if context_type == 'incognito' else str(test_dir / f'user_data_har_{alias}')-- browser_session = BrowserSession(- browser_profile=BrowserProfile(- headless=True,- disable_security=True,- user_data_dir=user_data_dir,- **{alias: str(har_path)}, # type: ignore- )- )- await browser_session.start()- try:- event = browser_session.event_bus.dispatch(NavigateToUrlEvent(url=httpserver_url))- await event- await event.event_result(raise_if_any=True, raise_if_none=False)- await asyncio.sleep(0.5)- finally:- await browser_session.kill()-- # HAR files should be created- assert har_path.exists(), f'HAR file was not created at {har_path}'-- # Check HAR file content- har_content = json.loads(har_path.read_text())- assert 'log' in har_content, "HAR file missing 'log' key"- assert 'entries' in har_content['log'], 'HAR file missing entries'- assert len(har_content['log']['entries']) > 0, 'HAR file has no network entries'-- @pytest.mark.parametrize(- 'context_type,alias',- [- ('incognito', 'trace_path'),- ('incognito', 'traces_dir'),- ('persistent', 'trace_path'),- ('persistent', 'traces_dir'),- ],- )- @pytest.mark.skip(reason='Trace recording temporarily disabled due to test timeouts')- async def test_trace_recording(self, test_dir, httpserver_url, context_type, alias, interactive_llm):- """Test trace recording with different contexts and aliases."""- browser_session = BrowserSession(- browser_profile=BrowserProfile(- headless=True,- disable_security=True,- user_data_dir=None if context_type == 'incognito' else str(test_dir / f'user_data_trace_{alias}'),- )- )-- # Use browser session ID to create unique trace directory- trace_dir = test_dir / f'trace_{context_type}_{alias}_{browser_session.id}'-- # Clean up any existing directory at this path- if trace_dir.exists():- shutil.rmtree(trace_dir)-- # Set the trace directory - trace_path is an alias for traces_dir- if alias == 'trace_path':- browser_session.browser_profile.traces_dir = str(trace_dir)- else:- setattr(browser_session.browser_profile, alias, str(trace_dir)) # type: ignore-- await browser_session.start()- try:- # Use Agent to interact with page for better trace content- agent = Agent(- task=f'go to {httpserver_url} and type "test" in the search box',- llm=interactive_llm,- browser_session=browser_session,- )- await agent.run(max_steps=5)- finally:- await browser_session.kill()-- # Check trace file - should be created automatically in the directory- assert trace_dir.exists(), f'Trace directory was not created at {trace_dir}'- trace_files = list(trace_dir.glob('*.zip'))- assert len(trace_files) > 0, f'No trace files were created in {trace_dir}'-- trace_file = trace_files[0]- assert zipfile.is_zipfile(trace_file), 'Trace file is not a valid ZIP'-- with zipfile.ZipFile(trace_file, 'r') as zip_file:- files = zip_file.namelist()- assert len(files) > 0, 'Trace ZIP file is empty'- assert any('trace' in f.lower() for f in files), 'Trace ZIP missing trace data'---class TestCombinedRecordings:- """Test using multiple recording parameters together."""-- async def test_all_recording_parameters(self, test_dir, httpserver_url, interactive_llm):- """Test using all recording parameters together."""- conversation_path = test_dir / 'conversation'- gif_path = test_dir / 'agent.gif'- video_dir = test_dir / 'videos'- har_path = test_dir / 'network.har'- trace_dir = test_dir / 'traces'-- browser_session = BrowserSession(- browser_profile=BrowserProfile(- headless=True,- disable_security=True,- user_data_dir=None,- record_video_dir=str(video_dir),- record_har_path=str(har_path),- traces_dir=str(trace_dir),- )- )-- await browser_session.start()-- try:- agent = Agent(- task=f'go to {httpserver_url} and type "test" in the search box',- llm=interactive_llm,- browser_session=browser_session,- save_conversation_path=str(conversation_path),- generate_gif=str(gif_path),- )- history: AgentHistoryList = await agent.run(max_steps=5)-- result = history.final_result()- assert result is not None-- # Check conversation files in directory- conversation_files = list(conversation_path.glob('conversation_*.txt'))- assert len(conversation_files) > 0, 'Conversation file was not created'-- # Check GIF - should be created since we navigate to a real page- assert gif_path.exists(), 'GIF was not created'- assert gif_path.stat().st_size > 10000, 'GIF file is too small'-- # Check video directory- assert video_dir.exists(), 'Video directory was not created'- finally:- await browser_session.kill()-- # Check files created after browser close- video_files = list(video_dir.glob('*.webm'))- assert len(video_files) > 0, 'No video files were created'-- assert har_path.exists(), 'HAR file was not created'-- # Verify HAR file- har_content = json.loads(har_path.read_text())- assert 'log' in har_content and 'entries' in har_content['log'], 'Invalid HAR structure'-- # TEMPORARILY DISABLED: Trace recording is disabled due to test timeouts- # assert trace_dir.exists(), 'Trace directory was not created'- # trace_files = list(trace_dir.glob('*.zip'))- # assert len(trace_files) > 0, 'No trace files were created'-- # # Verify trace file- # trace_file = trace_files[0]- # assert zipfile.is_zipfile(trace_file), 'Trace file is not a valid ZIP'diff --git a/tests/ci/test_tools.py b/tests/ci/test_tools.pyindex 6add33246..f22afebe0 100644--- a/tests/ci/test_tools.py+++ b/tests/ci/test_tools.py@@ -122,8 +122,8 @@ class TestToolsIntegration:@tools.action('Test custom action', param_model=CustomParams)async def custom_action(params: CustomParams, browser_session):- page = await browser_session.get_current_page()- return ActionResult(extracted_content=f'Custom action executed with: {params.text} on {page.url}')+ current_url = await browser_session.get_current_page_url()+ return ActionResult(extracted_content=f'Custom action executed with: {params.text} on {current_url}')# Navigate to a page firstgoto_action = {'go_to_url': GoToUrlAction(url=f'{base_url}/page1', new_tab=False)}@@ -182,10 +182,10 @@ class TestToolsIntegration:# Verify the resultassert isinstance(result, ActionResult)assert result.extracted_content is not None- assert 'Waiting for' in result.extracted_content+ assert 'Waited for' in result.extracted_content or 'Waiting for' in result.extracted_content- # Verify that less than 0.1 second has passed (because we deducted 3 seconds to account for the llm call)- assert end_time - start_time <= 0.1 # Allow some timing margin+ # Verify that approximately 1 second has passed (allowing some margin)+ assert 0.8 <= end_time - start_time <= 1.5 # Allow some timing margin for 1 second wait# longer wait# Create wait action for 1 second - fix to use a dictionary@@ -203,10 +203,10 @@ class TestToolsIntegration:# Verify the resultassert isinstance(result, ActionResult)assert result.extracted_content is not None- assert 'Waiting for' in result.extracted_content+ assert 'Waited for' in result.extracted_content or 'Waiting for' in result.extracted_content- # Verify that we took 2 sec (5s-3s (llm call)= 2s)- assert end_time - start_time <= 2.1 # Allow some timing margin+ # Verify that approximately 5 seconds have passed (allowing some margin)+ assert 4.5 <= end_time - start_time <= 6.0 # Allow some timing margin for 5 second waitassert end_time - start_time >= 1.9 # Allow some timing marginasync def test_go_back_action(self, tools, browser_session, base_url):@@ -220,8 +220,7 @@ class TestToolsIntegration:await tools.act(GoToUrlActionModel(**goto_action1), browser_session)# Store the first page URL- page1 = await browser_session.get_current_page()- first_url = page1.url+ first_url = await browser_session.get_current_page_url()print(f'First page URL: {first_url}')# Navigate to second page@@ -229,8 +228,7 @@ class TestToolsIntegration:await tools.act(GoToUrlActionModel(**goto_action2), browser_session)# Verify we're on the second page- page2 = await browser_session.get_current_page()- second_url = page2.url+ second_url = await browser_session.get_current_page_url()print(f'Second page URL: {second_url}')assert f'{base_url}/page2' in second_url@@ -251,8 +249,7 @@ class TestToolsIntegration:await asyncio.sleep(1)# Verify we're back on a different page than before- page3 = await browser_session.get_current_page()- final_url = page3.url+ final_url = await browser_session.get_current_page_url()print(f'Final page URL after going back: {final_url}')# Try to verify we're back on the first page, but don't fail the test if not