Does output_pydantic support Enums?

[emremrah]

I have following task and agent, but the task output gives an error. Seems like it doesn't respect the enum output format:

pydantic_core._pydantic_core.ValidationError: 1 validation error for TroubleshootingTaskOutput
fixable
  Input should be 'fixable', 'not_fixable' or 'unknown' [type=enum, input_value='TroubleshootingEval', input_type=str]
    For further information visit https://errors.pydantic.dev/2.11/v/enum

During handling of the above exception, another exception occurred:

class TroubleshootingEval(str, Enum):
    fixable = "fixable"
    not_fixable = "not_fixable"
    unknown = "unknown"

class TroubleshootingTaskOutput(BaseModel):
    report: str = Field(..., description="A report about the issue")
    fixable: TroubleshootingEval = Field(..., description="Evaluation of the issue fixability. Should be one of the defined enum values")


troubleshooting_agent = Agent(
    llm=master_llm,
    role="Troubleshooting Agent",
    goal="""Identify and resolve the customer's issue: {customer_message}""",
    backstory="""You're a seasoned technician with home appliance products and their troubleshooting. You are here to help customers resolve their issues.""",
    cache=False,
    verbose=True,
    max_rpm=30,
    max_iter=5,
    tools=[read_troubleshooting_guide],
)

troubleshooting_task = Task(
    description="""Troubleshoot the customer's issue.

<customer_info>
{customer_info}
</customer_info>

<customer_message>
{customer_message}
</customer_message>""",
    agent=troubleshooting_agent,
    expected_output="A report about the issue",
    human_input=False,
    output_pydantic=models.TroubleshootingTaskOutput
)

[lucasgomide]

is expected to work. Could you run with verbose=True to get more logs?

[emremrah]

I did but it has a long log. Do you need any specific part?

[emremrah]

This is the crew manager's final answer:

# Agent: Crew Manager
## Final Answer: 
{
  "report": "# Report on Coffee Machine Not Heating Up\n\n## Possible Causes:\n1. **Power Supply Issues:**\n   - The coffee machine may not be receiving power due to a faulty power outlet or a disconnected power cord.\n\n2. **Thermostat Malfunction:**\n   - The thermostat, which regulates the heating element, may be faulty or damaged.\n\n3. **Heating Element Failure:**\n   - The heating element itself may be burned out or damaged, preventing the machine from heating.\n\n4. **Internal Wiring Issues:**\n   - There could be a problem with the internal wiring or connections within the coffee machine.\n\n5. **Control Board Failure:**\n   - The control board that manages the heating function may be defective.\n\n## Troubleshooting Steps:\n1. **Check Power Supply:**\n   - Ensure that the coffee machine is plugged into a working power outlet.\n   - Try using a different outlet to rule out power supply issues.\n\n2. **Inspect the Thermostat:**\n   - If accessible, check the thermostat for any visible signs of damage or wear.\n   - Consider replacing the thermostat if it appears faulty.\n\n3. **Examine the Heating Element:**\n   - If possible, inspect the heating element for any visible damage or burn marks.\n   - Replace the heating element if it is found to be defective.\n\n4. **Check Internal Wiring:**\n   - Inspect the internal wiring for any loose connections or damaged wires.\n   - Secure any loose connections and replace damaged wires as necessary.\n\n5. **Evaluate the Control Board:**\n   - If other components are functioning correctly, the control board may need to be tested or replaced by a professional.\n\n## Conclusion:\nIf the above troubleshooting steps do not resolve the issue, it is recommended to contact a professional technician for further diagnosis and repair. The issue may be fixable depending on the specific component that is malfunctioning.",
  "fixable": "TroubleshootingEval"
}

[repo-darklinux]

Trying it:

You need to help the agent fill it out correctly.

Improve the expected_output (more guidance)

Instead of just putting "A report about the issue", you can detail it like this

expected_output="""
You must respond with a JSON containing:

- 'report': a detailed report about the issue.
- 'fixable': one of the following exact values: 'fixable', 'not_fixable', or 'unknown'.

Do not invent new fields. 
Only use exactly one of the three values for 'fixable'.
"""

Do not invent new fields.

Only use exactly one of the three values ​​for 'fixable'.
"""

This way, LLM will know how to structure it correctly.

let me know if it worked after you run it....

[emremrah]

Thanks for the reply. That's actually working. I just assumed that we wouldn't need it because the LLM "reads" the description of the pydantic schema; I thought, with this way, the process will be automatized without dealing with the expected output if we change something in the output model.