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
Revert "fix(tools): refuse evaluate() on restricted browser profiles (#4871)"
MagMuelleraf9d40641983
diff --git a/browser_use/tools/service.py b/browser_use/tools/service.pyindex be34451a1..cf999b849 100644--- a/browser_use/tools/service.py+++ b/browser_use/tools/service.py@@ -1775,14 +1775,7 @@ You will be given a query and the markdown of a webpage that has been filtered tterminates_sequence=True,)async def evaluate(code: str, browser_session: BrowserSession):- # SecurityWatchdog only hooks navigation, so evaluate() would bypass- # profile restrictions via fetch() / direct origin access.- profile = browser_session.browser_profile- if profile.allowed_domains or profile.prohibited_domains or profile.block_ip_addresses:- return ActionResult(- error='evaluate() is disabled on profiles configured with allowed_domains, '- 'prohibited_domains, or block_ip_addresses.'- )+ # Execute JavaScript with proper error handling and promise supportcdp_session = await browser_session.get_or_create_cdp_session()diff --git a/tests/ci/security/test_evaluate_restricted.py b/tests/ci/security/test_evaluate_restricted.pydeleted file mode 100644index 461d0bc38..000000000--- a/tests/ci/security/test_evaluate_restricted.py+++ /dev/null@@ -1,116 +0,0 @@-"""Tests for the evaluate() guard on restricted browser profiles.--The agent's `evaluate()` action calls `Runtime.evaluate` directly through CDP.-SecurityWatchdog only subscribes to navigation events, so JS running inside an-already-allowed page can `fetch()` arbitrary internal URLs, read cookies and-localStorage from any allowed origin's context, and otherwise act as if the-restrictions weren't there.--On profiles configured with `allowed_domains` or `block_ip_addresses`, the-operator has signalled "this agent is constrained" — exposing an unmediated-JS evaluation primitive contradicts that signal. The guard refuses-`evaluate()` outright on such profiles; agents that need JS can be run on an-unrestricted profile or the gate can be lifted explicitly later.-"""--from __future__ import annotations--from typing import Any--import pytest--from browser_use.agent.views import ActionResult-from browser_use.tools.service import Tools---class _StubProfile:- def __init__(- self,- allowed_domains: Any = None,- prohibited_domains: Any = None,- block_ip_addresses: bool = False,- ) -> None:- self.allowed_domains = allowed_domains- self.prohibited_domains = prohibited_domains- self.block_ip_addresses = block_ip_addresses---class _StubBrowserSession:- is_local = True- agent_focus_target_id: str | None = None- session_manager: Any = None- cdp_client: Any = None-- def __init__(self, profile: _StubProfile) -> None:- self.browser_profile = profile-- async def get_current_page_url(self) -> str:- return 'about:blank'-- async def get_or_create_cdp_session(self) -> Any:- raise AssertionError('evaluate() must refuse before reaching CDP when restrictions are configured')---async def _run_evaluate(profile: _StubProfile, code: str = '1 + 1') -> ActionResult:- tools = Tools()- result = await tools.registry.execute_action(- 'evaluate',- {'code': code},- browser_session=_StubBrowserSession(profile), # type: ignore[arg-type]- )- assert isinstance(result, ActionResult)- return result---async def test_evaluate_refused_when_allowed_domains_set() -> None:- """A non-empty `allowed_domains` list signals a restricted profile — refuse."""- result = await _run_evaluate(_StubProfile(allowed_domains=['example.com']))- assert result.error is not None- assert 'allowed_domains' in result.error.lower() or 'restricted' in result.error.lower()---async def test_evaluate_refused_when_block_ip_addresses_set() -> None:- """`block_ip_addresses=True` signals a restricted profile — refuse."""- result = await _run_evaluate(_StubProfile(block_ip_addresses=True))- assert result.error is not None- assert 'block_ip_addresses' in result.error.lower() or 'restricted' in result.error.lower()---async def test_evaluate_refused_when_prohibited_domains_set() -> None:- """`prohibited_domains` is the deny-list counterpart of `allowed_domains`;- SecurityWatchdog enforces both as navigation restrictions. Refuse evaluate()- when either is set — otherwise an agent can fetch() into the blocked domains."""- result = await _run_evaluate(_StubProfile(prohibited_domains=['evil.test']))- assert result.error is not None- assert 'prohibited_domains' in result.error.lower() or 'restricted' in result.error.lower()---async def test_evaluate_refused_when_both_restrictions_set() -> None:- """Belt and suspenders configuration — refuse."""- result = await _run_evaluate(_StubProfile(allowed_domains=['x.test'], block_ip_addresses=True))- assert result.error is not None---async def test_evaluate_reaches_cdp_when_unrestricted() -> None:- """No restrictions → evaluate() should proceed to CDP. Our stub raises- when `get_or_create_cdp_session` is called, so we observe via the wrapped- error that the guard did not short-circuit."""- tools = Tools()- with pytest.raises(RuntimeError, match='must refuse before reaching CDP'):- await tools.registry.execute_action(- 'evaluate',- {'code': '1 + 1'},- browser_session=_StubBrowserSession(_StubProfile()), # type: ignore[arg-type]- )---async def test_evaluate_proceeds_when_allowed_domains_is_empty_list() -> None:- """Empty allowed_domains is treated as 'no restriction' elsewhere in the- codebase (e.g. SecurityWatchdog). Be consistent: don't refuse evaluate()."""- tools = Tools()- with pytest.raises(RuntimeError, match='must refuse before reaching CDP'):- await tools.registry.execute_action(- 'evaluate',- {'code': '1 + 1'},- browser_session=_StubBrowserSession(_StubProfile(allowed_domains=[])), # type: ignore[arg-type]- )