ai-agents-tool-argument-valida-when-using-hierarchical-process-with-crewai-the-ma-be2dbe08

THEN To mitigate, ensure that task and context are provided as plain strings. Consider updating the tool's input schema to accept both string and dict using Union[str, dict] with a custom validator to coerce dict values into string. Alternatively, use a more capable model like gpt-4o that is less prone to this behavior. Also verify that task descriptions in YAML configuration are simple strings, not objects.

from typing import Union
from pydantic import BaseModel, validator
class DelegateWorkToolSchema(BaseModel):
    task: Union[str, dict]
    context: Union[str, dict]
    coworker: str

    @validator('task', 'context', pre=True)
    def coerce_dict_to_string(cls, v):
        if isinstance(v, dict):
            return v.get('description', str(v))
        return v