Validation

ToolDefinition.validate() performs thin argument checking against the generated parameters schema.

This is not full Pydantic validation — use Pydantic for domain models.

Basic usage

from toolschema import schema, ValidationSuccess, ValidationFailure

definition = schema(my_tool)

result = definition.validate({"name": "Ada", "limit": 5})

if isinstance(result, ValidationSuccess):
    my_tool(**result.value)  # defaults applied
else:
    for issue in result.issues:
        print(issue.kind, issue.path, issue.message)

What gets validated

Check	Example
Required fields	Missing name when required
JSON types	"abc" for integer param
enum / Literal	Value not in allowed set
minLength / maxLength	String too short
pattern	Regex mismatch
minimum / maximum	Number out of range
additionalProperties: false	Extra keys rejected
Nested objects / arrays	Recursive validation
anyOf / optional types	At least one branch matches
Defaults	Missing optional fields filled from schema

ValidationSuccess

ValidationSuccess(
    value={"name": "Ada", "limit": 10},  # defaults merged in
    issues=None,
)

Use result.value as kwargs for your function.

ValidationFailure

ValidationFailure(
    value=None,
    issues=(
        ValidationIssue(
            message="Missing required property 'name'",
            path=("name",),
            kind=ValidationIssueKind.REQUIRED,
        ),
    ),
)

Issue kinds

Kind	Meaning
REQUIRED	Missing required property
TYPE	Wrong JSON type
ENUM	Not in enum
CONSTRAINT	minLength, pattern, min/max, etc.
ADDITIONAL_PROPERTY	Extra key when not allowed

Standard Schema integration

validate is also exposed via the Standard Schema protocol:

standard = definition.standard["~standard"]
result = standard["validate"]({"name": "Ada"})

Limitations (v1.0)

• No $ref resolution at validate time (schemas are usually inline)
• No format keyword (email, uri, etc.)
• No oneOf discrimination beyond trying each branch
• Not a replacement for Pydantic model_validate

For production argument parsing, consider: validate() → my_fn(**result.value) or Pydantic models as parameter types.