validateArgs

function validateArgs(schema: ToolSchema, args: unknown): ValidationResult

import { validateArgs } from '@driftengine/ai';

Parameters

In depth

Never throws. A model producing bad arguments is ordinary rather than exceptional — it is the thing schemas exist for — and putting the ordinary case on the exception path means every call site needs a try it will eventually forget.

An unknown field is a failure, not something to drop. A model inventing a field is a signal: it means the prompt and the schema disagree about the tool's shape, and silently discarding it is how that disagreement survives a release. What it costs: a provider that helpfully adds metadata to a tool call breaks until the schema names it. What would make it wrong: if a mainstream provider does that unavoidably, the answer is a declared passthrough field rather than a blanket tolerance.