jd-opensource/joyagent-jdgenie · error · AgentExecutionError

{error_msg}

Error message

{error_msg}

What it means

smolagents-style CodeAgent raises AgentExecutionError in _step_stream when executing the generated code in the sandbox fails (runtime exception in the model-written Python, tool error, or forbidden import). The sandbox error message is passed through verbatim.

Solutions

1. Read the wrapped error_msg to see the actual sandbox exception and fix the model prompt or the code
2. Pass additional_authorized_imports=[...] to CodeAgent when the model legitimately needs imports like requests or numpy
3. Add the failing packages to the execution environment the sandbox uses
4. Increase max_iterations / improve the system prompt so the agent can self-correct on execution failures

Example fix

# before
agent = CodeAgent(tools=[python_interpreter])
# model code 'import requests' -> AgentExecutionError: Import of requests is not allowed
# after
agent = CodeAgent(tools=[python_interpreter], additional_authorized_imports=["requests"])

authorized = {"os", "math"} | set(additional_authorized_imports)
imports = set(re.findall(r'^\s*(?:import|from)\s+(\w+)', generated_code, re.M))
missing = imports - authorized  # pass missing into additional_authorized_imports first

from smolagents import AgentExecutionError
try:
    result = agent.run(task)
except AgentExecutionError as e:
    if 'is not allowed' in str(e):
        agent.authorized_imports += [extracted_module]
    logger.warning('execution failed: %s', e)

When it happens

Trigger: Any exception raised while running the model-generated code via the python interpreter (sandboxed_executor/execute_python_code), including unauthorized imports detected by the sandbox ("Import of X is not allowed").

Common situations: Model writes code calling libraries not in additional_authorized_imports; generated code hits NameError/ZeroDivisionError; sandbox lacks packages the model assumed (requests, pandas); model calls undefined functions.

Related errors

• Empty or invalid response from LLM
• Invalid tool_choice: " + toolChoice
• Invalid or empty response from LLM
• Error in generating model output
• Error in code parsing

AI-assisted analysis of jd-opensource/joyagent-jdgenie@2417e0b8b6 (2026-09-08). Data as JSON: /api/errors/0a28eba470b63bfa. Report an issue: GitHub.

Appendix: source

Thrown at genie-tool/genie_tool/tool/ci_agent.py:197

            ):
                execution_logs = str(self.python_executor.state["_print_outputs"])
                if len(execution_logs) > 0:
                    execution_outputs_console = [
                        Text("Execution logs:", style="bold"),
                        Text(execution_logs),
                    ]
                    memory_step.observations = "Execution logs:\n" + execution_logs
                    self.logger.log(
                        Group(*execution_outputs_console), level=LogLevel.INFO
                    )
            error_msg = str(e)

            if "Import of " in error_msg and " is not allowed" in error_msg:
                self.logger.log(
                    "[bold red]Warning to user: Code execution failed due to an unauthorized import - Consider passing said import under `additional_authorized_imports` when initializing your CodeAgent.",
                    level=LogLevel.INFO,
                )
            raise AgentExecutionError(error_msg, self.logger)

        memory_step.observations = observation

        finalObj = FinalAnswerCheck(
            input_messages=self.input_messages,
            execution_logs=execution_logs,
            model=self.model,
            task=self.task,
            prompt_temps=self.prompt_templates,
            memory_step=memory_step,
            grammar=self.grammar,
            request_id=f"{model_request_id}-final",
        )
        finalFlag, exeLog = finalObj.check_is_final_answer()
        self.logger.log(Group(*execution_outputs_console), level=LogLevel.INFO)
        # self.logger.log(f"check finalanswer 已完成 {finalFlag}  {str(exeLog)}")
        memory_step.action_output = exeLog

View on GitHub (pinned to 2417e0b8b6)