🐫 CAMEL: The first and the best multi-agent framework. Finding the Scaling Law of Agents. https://www.camel-ai.org
Python agentai-societiesartificial-intelligencecommunicative-aicooperative-aideep-learninglarge-language-modelsmulti-agent-systemsnatural-language-processing
Feat: Require confirmation for internal Python shell commands (#4142)
Carbene2a0ba901d3ce
diff --git a/camel/interpreters/internal_python_interpreter.py b/camel/interpreters/internal_python_interpreter.pyindex a5b39b144..badfcfd17 100644--- a/camel/interpreters/internal_python_interpreter.py+++ b/camel/interpreters/internal_python_interpreter.py@@ -78,6 +78,8 @@ class InternalPythonInterpreter(BaseInterpreter):unsafe_mode (bool, optional): If `True`, the interpreter runs the codeby `eval()` or `exec()` without any security check.(default: :obj:`False`)+ require_confirm (bool, optional): If True, prompt user before running+ shell commands on the local computer. (default: :obj:`True`)raise_error (bool, optional): Raise error if the interpreter fails.(default: :obj:`False`)allow_builtins (bool, optional): If `True`, safe built-in functions@@ -92,6 +94,7 @@ class InternalPythonInterpreter(BaseInterpreter):action_space: Optional[Dict[str, Any]] = None,import_white_list: Optional[List[str]] = None,unsafe_mode: bool = False,+ require_confirm: bool = True,raise_error: bool = False,allow_builtins: bool = True,) -> None:@@ -101,6 +104,7 @@ class InternalPythonInterpreter(BaseInterpreter):self.import_white_list = import_white_list or list()self.raise_error = raise_errorself.unsafe_mode = unsafe_mode+ self.require_confirm = require_confirm# Add safe built-in functions if allowedif allow_builtins:@@ -648,6 +652,9 @@ class InternalPythonInterpreter(BaseInterpreter):Returns:tuple: A tuple containing the stdout and stderr of the command."""+ if self.require_confirm:+ self._confirm_execution(command)+try:proc = subprocess.Popen(command,@@ -662,3 +669,24 @@ class InternalPythonInterpreter(BaseInterpreter):return stdout, stderrexcept Exception as e:raise InterpreterError(f"Error executing command: {e}")++ def _confirm_execution(self, command: str) -> None:+ r"""Prompt the user before executing a local shell command."""+ while True:+ choice = (+ input(+ "The following shell command will run on your "+ f"computer: {command}\nRunning command? [y/N]:"+ )+ .lower()+ .strip()+ )+ if choice in ["y", "yes", "ye"]:+ return+ if choice in ["no", "n", ""]:+ raise InterpreterError(+ "Execution halted: User opted not to run the "+ "command. This choice stops the current operation "+ "and any further command execution."+ )+ print("Please enter 'y' or 'n'.")diff --git a/camel/toolkits/code_execution.py b/camel/toolkits/code_execution.pyindex ccc2b8c17..021b10055 100644--- a/camel/toolkits/code_execution.py+++ b/camel/toolkits/code_execution.py@@ -43,8 +43,8 @@ class CodeExecutionToolkit(BaseToolkit):import_white_list (Optional[List[str]]): A list of allowed imports.(default: :obj:`None`)require_confirm (Optional[bool]): Whether to require confirmation- before executing code. If `None`, subprocess execution requires- confirmation by default. (default: :obj:`None`)+ before executing code. If `None`, local host execution sandboxes+ require confirmation by default. (default: :obj:`None`)timeout (Optional[float]): General timeout for toolkit operations.(default: :obj:`None`)microsandbox_config (Optional[dict]): Configuration for microsandbox@@ -75,8 +75,9 @@ class CodeExecutionToolkit(BaseToolkit):self.verbose = verboseself.unsafe_mode = unsafe_modeself.import_white_list = import_white_list or list()+ host_execution_sandboxes = {"internal_python", "subprocess"}resolved_require_confirm = (- sandbox == "subprocess"+ sandbox in host_execution_sandboxesif require_confirm is Noneelse require_confirm)@@ -95,6 +96,7 @@ class CodeExecutionToolkit(BaseToolkit):self.interpreter = InternalPythonInterpreter(unsafe_mode=self.unsafe_mode,import_white_list=self.import_white_list,+ require_confirm=resolved_require_confirm,)elif sandbox == "jupyter":self.interpreter = JupyterKernelInterpreter(diff --git a/test/toolkits/test_code_execution.py b/test/toolkits/test_code_execution.pyindex bb7178e58..ac2621b0b 100644--- a/test/toolkits/test_code_execution.py+++ b/test/toolkits/test_code_execution.py@@ -16,7 +16,11 @@ import socketimport pytest-from camel.interpreters import InterpreterError, SubprocessInterpreter+from camel.interpreters import (+ InternalPythonInterpreter,+ InterpreterError,+ SubprocessInterpreter,+)from camel.toolkits.code_execution import CodeExecutionToolkitfrom camel.utils import is_docker_running@@ -190,6 +194,13 @@ def test_subprocess_requires_confirmation_by_default():assert toolkit.interpreter.require_confirm is True+def test_internal_python_requires_confirmation_by_default():+ toolkit = CodeExecutionToolkit(sandbox="internal_python")++ assert isinstance(toolkit.interpreter, InternalPythonInterpreter)+ assert toolkit.interpreter.require_confirm is True++def test_subprocess_default_confirmation_blocks_code(monkeypatch):toolkit = CodeExecutionToolkit()monkeypatch.setattr('builtins.input', lambda _: 'n')@@ -210,6 +221,16 @@ def test_subprocess_default_confirmation_blocks_command(monkeypatch):assert "Execution halted" in str(exc_info.value)+def test_internal_python_default_confirmation_blocks_command(monkeypatch):+ toolkit = CodeExecutionToolkit(sandbox="internal_python")+ monkeypatch.setattr('builtins.input', lambda _: 'n')++ with pytest.raises(InterpreterError) as exc_info:+ toolkit.execute_command("echo blocked")++ assert "Execution halted" in str(exc_info.value)++def test_subprocess_default_confirmation_blocks_empty_code(monkeypatch):toolkit = CodeExecutionToolkit()monkeypatch.setattr('builtins.input', lambda _: '')@@ -237,6 +258,15 @@ def test_subprocess_confirmation_can_be_disabled():assert toolkit.interpreter.require_confirm is False+def test_internal_python_confirmation_can_be_disabled():+ toolkit = CodeExecutionToolkit(+ sandbox="internal_python", require_confirm=False+ )++ assert isinstance(toolkit.interpreter, InternalPythonInterpreter)+ assert toolkit.interpreter.require_confirm is False++def test_invalid_sandbox_type():with pytest.raises(RuntimeError) as exc_info:CodeExecutionToolkit(sandbox="invalid")