<@ULVA73B9P> why does my marvin agent ignore the t...
# ask-marvinp
PyHannes
10/02/2025, 1:14 PM@Marvin why does my marvin agent ignore the type hints of my tool?
This is my tool:
Copy code
def start_flow(name: Annotated[str, "The name of the flow to start."],
parameters: Annotated[dict[str, Any], "A dictionary of parameters to pass to the flow."]) -> str:
"""
Start a new flow run with the given parameters.
:param name: The name of the flow to start. Must be provided by the user.
:param parameters: A dictionary of parameters to pass to the flow. Must be provided by the user.
:return: The ID of the started flow run.
"""
log = get_run_logger()
<http://log.info|log.info>(f"Starting flow '{name}' with parameters: {parameters}")
return "abc" Copy code
gpt_agent = GptAgent.load("...").get_agent(
prompt="You are a helpful customer support assistant."
"If you're lacking information on tool calls or missing parameter values, always request it from the user!"
)
response = marvin.run(
instructions="start a flow run with arguments {'name': 'foobar', 'parameters': {'temp': '123', 'check': '445'}}",
agents=[gpt_agent],
tools=[start_flow],
raise_on_failure=True,
)m
Marvin
10/02/2025, 1:14 PM✅ thought for 424.8 seconds
Marvin
10/02/2025, 1:21 PMShort answer: the combo of a very open type (dict[str, Any]) and the parameter name "parameters" is tripping up the tool-call schema and the model. In practice, models often omit loosely-typed or ambiguously-named fields, and "parameters" is especially unlucky because it’s also the name of the JSON Schema field used to define a tool. The result is that the tool is being called with just the first arg.
What usually fixes this:
- Rename the argument away from "parameters" (e.g., "flow_parameters").
- Make the type concrete (TypedDict or a Pydantic model) instead of dict[str, Any].
Example with a TypedDict:
Copy code
from typing import Annotated, TypedDict
from prefect import get_run_logger
class FlowParameters(TypedDict):
temp: str
check: str
def start_flow(
name: Annotated[str, "The name of the flow to start."],
flow_parameters: Annotated[FlowParameters, "Parameters for the flow."]
) -> str:
"""
Start a new flow run with the given parameters.
"""
log = get_run_logger()
<http://log.info|log.info>(f"Starting flow '{name}' with parameters: {flow_parameters}")
return "abc" Copy code
response = marvin.run(
instructions="start a flow run with arguments {'name': 'foobar', 'flow_parameters': {'temp': '123', 'check': '445'}}",
agents=[gpt_agent],
tools=[start_flow],
raise_on_failure=True,
) Copy code
from typing import Annotated
from pydantic import BaseModel, Field
from prefect import get_run_logger
class FlowParameters(BaseModel):
temp: str = Field(..., description="Temperature")
check: str = Field(..., description="Check code")
def start_flow(
name: Annotated[str, "The name of the flow to start."],
flow_parameters: Annotated[FlowParameters, "Parameters for the flow."],
) -> str:
log = get_run_logger()
<http://log.info|log.info>(f"Starting flow '{name}' with parameters: {flow_parameters.model_dump()}")
return "abc"