Skip to content

Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API.

Sanitize direct-start URL candidates

Magnus Müller91705e22b778
diff --git a/browser_use/agent/service.py b/browser_use/agent/service.py
index 43bebc780..4d18660ab 100644
--- a/browser_use/agent/service.py
+++ b/browser_use/agent/service.py
@@ -70,6 +70,7 @@ from browser_use.utils import (
get_browser_use_version,
get_git_info,
is_placeholder_url,
+ sanitize_url_candidate,
time_execution_async,
time_execution_sync,
)
@@ -1505,7 +1506,7 @@ class Agent(Generic[Context, AgentStructuredOutput]):
original_position = match.start() # Store original position before URL modification
# Remove trailing punctuation that's not part of URLs
- url = re.sub(r'[.,;:!?()\[\]]+$', '', url)
+ url = sanitize_url_candidate(url)
if is_placeholder_url(url):
self.logger.debug(f'Excluding placeholder URL from auto-navigation: {url}')
diff --git a/browser_use/rust/service.py b/browser_use/rust/service.py
index 54107bd11..592db0de6 100644
--- a/browser_use/rust/service.py
+++ b/browser_use/rust/service.py
@@ -67,6 +67,7 @@ from browser_use.utils import (
get_browser_use_version,
get_git_info,
is_placeholder_url,
+ sanitize_url_candidate,
)
@@ -866,7 +867,7 @@ def _extract_start_url(task: str) -> str | None:
found_urls = []
for pattern in patterns:
for match in re.finditer(pattern, task_without_emails):
- url = re.sub(r'[.,;:!?()\[\]]+$', '', match.group(0))
+ url = sanitize_url_candidate(match.group(0))
url_lower = url.lower()
if is_placeholder_url(url):
continue
diff --git a/browser_use/utils.py b/browser_use/utils.py
index d822ae5e3..5e878bc9a 100644
--- a/browser_use/utils.py
+++ b/browser_use/utils.py
@@ -38,6 +38,17 @@ def is_placeholder_url(url: str) -> bool:
return len(labels) >= 2 and all(re.fullmatch(r'x+', label) for label in labels)
+
+def sanitize_url_candidate(url: str) -> str:
+ """Normalize a URL candidate captured from prose before auto-navigation."""
+ candidate = url.strip()
+ # Some benchmark tasks arrive with escaped newlines in prose, e.g.
+ # "https://example.com/search.\\n2. Next step". Those are task text,
+ # not part of the URL.
+ candidate = re.split(r'\\[nrt]', candidate, maxsplit=1)[0]
+ return re.sub(r'[.,;:!?()\[\]]+$', '', candidate)
+
+
# Import error types - these may need to be adjusted based on actual import paths
try:
from openai import BadRequestError as OpenAIBadRequestError
diff --git a/tests/ci/test_rust_agent.py b/tests/ci/test_rust_agent.py
index 3b7ae3e19..aa55fa58b 100644
--- a/tests/ci/test_rust_agent.py
+++ b/tests/ci/test_rust_agent.py
@@ -2747,6 +2747,12 @@ def test_rust_agent_exposes_task_helper_methods():
assert agent._extract_start_url('Open https://example.com/report.pdf and summarize it.') is None
assert agent._extract_start_url('Use https://XXX.XX as a placeholder in the table.') is None
assert browser_use_agent._extract_start_url('Use https://XXX.XX as a placeholder in the table.') is None
+ numbered_task = (
+ '1. Navigate to https://elibrary.ferc.gov/eLibrary/search.\\n'
+ '2. Ensure "General Search" is selected.'
+ )
+ assert agent._extract_start_url(numbered_task) == 'https://elibrary.ferc.gov/eLibrary/search'
+ assert browser_use_agent._extract_start_url(numbered_task) == 'https://elibrary.ferc.gov/eLibrary/search'
def test_rust_agent_exposes_url_text_helper_methods():