huggingface/smolagents · error · InterpreterError

Code execution failed at line '{ast.get_source_segment(code,

Error message

Code execution failed at line '{ast.get_source_segment(code, node)}' due to: {type(e).__name__}: {e}

What it means

This is the executor's generic runtime-failure wrapper: any exception raised while executing a node that is not FinalAnswerException is re-raised as InterpreterError with the source segment of the failing line and the original exception type and message. The original exception is not chained by `from e`, so the message text is the primary diagnostic. It marks where in the agent's code action the failure occurred.

Source

Thrown at src/smolagents/local_python_executor.py:1659

        try:
            for node in expression.body:
                result = evaluate_ast(node, state, static_tools, custom_tools, authorized_imports)
            state["_print_outputs"].value = truncate_content(
                str(state["_print_outputs"]), max_length=max_print_outputs_length
            )
            is_final_answer = False
            return result, is_final_answer
        except FinalAnswerException as e:
            state["_print_outputs"].value = truncate_content(
                str(state["_print_outputs"]), max_length=max_print_outputs_length
            )
            is_final_answer = True
            return e.value, is_final_answer
        except Exception as e:
            state["_print_outputs"].value = truncate_content(
                str(state["_print_outputs"]), max_length=max_print_outputs_length
            )
            raise InterpreterError(
                f"Code execution failed at line '{ast.get_source_segment(code, node)}' due to: {type(e).__name__}: {e}"
            )

    # Apply timeout if specified
    if timeout_seconds is not None:
        _execute_code = timeout(timeout_seconds)(_execute_code)

    return _execute_code()


@dataclass
class CodeOutput:
    output: Any
    logs: str
    is_final_answer: bool


class PythonExecutor(ABC):

View on GitHub (pinned to 30bb116109)

Solutions

1. Read the inner `TypeName: message` to identify the real cause and fix the offending line quoted in the error
2. For agent workflows, pass this error message back to the LLM for self-correction (step_number/output trimming is standard)
3. Add defensive checks (None guards, isinstance checks) around tool outputs used in subsequent code
4. If the inner error is itself InterpreterError about unsupported syntax, address that root cause instead

Example fix

# before
code = "result = alphabet_split('hello')[0]"  # tool returns str, not list

# after
code = "res = alphabet_split('hello')\nresult = res[0] if isinstance(res, list) else res"

from smolagents.local_python_executor import InterpreterError
try:
    output, is_final = evaluate_python_code(code, state=state)
except InterpreterError as e:
    # parse 'due to: <Type>: <msg>' from the message
    inner = str(e).split('due to: ', 1)[-1]
    if inner.startswith(('NameError', 'TypeError')):
        code = fix_with_llm(code, str(e))  # self-correction loop

When it happens

Trigger: Any runtime error inside executed code: NameError for undefined variables, TypeError on bad tool arguments, ZeroDivisionError, KeyError on subscripting, errors raised by called tools, or nested InterpreterErrors from unsupported nodes.

Common situations: LLM hallucinating variable or tool names; wrong argument types passed to tools; iterating over None returned by a previous step; errors inside imported authorized modules.

Related errors

• Error executing tool '{tool_name}' with arguments {str(argum
• Code parsing failed on line {e.lineno} due to: {type(e).__na
• Tool call needs to have a key '{tool_name_key}'. Got keys: {
• Error during jinja template rendering: {type(e).__name__}: {
• Cannot specify both 'messages' and 'steps' parameters. Use '

AI-assisted analysis of huggingface/smolagents@30bb116109 (2026-08-28). Data as JSON: /api/errors/3df09408e258cf92. Report an issue: GitHub.