huggingface/smolagents · error · TypeError

Attribute output_schema should have type dict, got {type(out

Error message

Attribute output_schema should have type dict, got {type(output_schema)} instead.

What it means

The optional Tool.output_schema attribute, when set, must be a dict (JSON-schema style). validate_arguments raises TypeError if output_schema is present but is any other type. Leaving it unset (None) is fine; the check only applies when you provide it.

Source

Thrown at src/smolagents/tools.py:164

            "description": str,
            "name": str,
            "inputs": dict,
            "output_type": str,
        }
        # Validate class attributes
        for attr, expected_type in required_attributes.items():
            attr_value = getattr(self, attr, None)
            if attr_value is None:
                raise TypeError(f"You must set an attribute {attr}.")
            if not isinstance(attr_value, expected_type):
                raise TypeError(
                    f"Attribute {attr} should have type {expected_type.__name__}, got {type(attr_value)} instead."
                )

        # Validate optional output_schema attribute
        output_schema = getattr(self, "output_schema", None)
        if output_schema is not None and not isinstance(output_schema, dict):
            raise TypeError(f"Attribute output_schema should have type dict, got {type(output_schema)} instead.")

        # - Validate name
        if not is_valid_name(self.name):
            raise Exception(
                f"Invalid Tool name '{self.name}': must be a valid Python identifier and not a reserved keyword"
            )
        # Validate inputs
        for input_name, input_content in self.inputs.items():
            assert isinstance(input_content, dict), f"Input '{input_name}' should be a dictionary."
            assert "type" in input_content and "description" in input_content, (
                f"Input '{input_name}' should have keys 'type' and 'description', has only {list(input_content.keys())}."
            )
            # Get input_types as a list, whether from a string or list
            if isinstance(input_content["type"], str):
                input_types = [input_content["type"]]
            elif isinstance(input_content["type"], list):
                input_types = input_content["type"]
                # Check if all elements are strings

View on GitHub (pinned to 30bb116109)

Solutions

1. Convert the value to a dict, e.g. json.loads(schema_string) or use the already-parsed schema object
2. If using pydantic, pass MyModel.model_json_schema() (which returns a dict), not the model class
3. Remove output_schema entirely if the tool returns plain text

Example fix

# before
class MyTool(Tool):
    output_schema = '{"type": "object", ...}'  # str -> TypeError

# after
class MyTool(Tool):
    output_schema = {"type": "object", "properties": {"result": {"type": "string"}}}

schema = getattr(MyTool, "output_schema", None)
assert schema is None or isinstance(schema, dict), "output_schema must be a dict"

def output_schema_ok(cls) -> bool:
    s = getattr(cls, "output_schema", None)
    return s is None or isinstance(s, dict)

try:
    MyTool()
except TypeError as e:
    if "output_schema" in str(e):
        MyTool.output_schema = dict(MyTool.output_schema)  # or parse JSON string
        MyTool()

When it happens

Trigger: Setting output_schema to a JSON string, a pydantic model, a list, or any non-dict value on a Tool subclass; error fires at instantiation via new_init → validate_arguments.

Common situations: Copy-pasting a JSON schema from docs as a string instead of parsing it into a dict; assigning a pydantic BaseModel class or .model_json_schema() result's string form; refactoring a tool from dict-based to string-based config.

Related errors

• Attribute {attr} should have type {expected_type.__name__},
• Input '{input_name}': when type is a list, all elements must
• {e}
• Object is not iterable
• Cannot add non-list value {value_to_add} to a list.

AI-assisted analysis of huggingface/smolagents@30bb116109 (2026-08-28). Data as JSON: /api/errors/9c4dc4390f5d25ef. Report an issue: GitHub.