Documentation article
Create a custom generator
Map a stable slice of Shape IR into a target format incrementally.
Custom generators are the clearest example of why the toolkit is organized around Shape IR rather than format-to-format adapters.
You do not need to solve every advanced edge case on day one. The first milestone is a generator that can consume a stable subset of Shape IR and report the gaps honestly.
A generator accepts Shape IR and emits a target representation such as TypeScript, JSON Schema, or a framework-specific config format.
The goal is not to mirror parser internals. The goal is to translate stable IR nodes into output decisions while preserving enough diagnostic context to explain loss or unsupported semantics.
A scoped first release is easier to test and much easier to explain in docs. That is especially important while the generator layer is still expanding.
The implementation can stay small as long as it makes unsupported cases explicit. Silent fallbacks make round-trip behavior much harder to trust later.
Minimal generator shape
type GeneratorResult = {
code: string;
diagnostics: string[];
};
export function emitShape(shape: Shape): GeneratorResult {
switch (shape.kind) {
case "string":
return {code: "string", diagnostics: []};
case "number":
return {code: "number", diagnostics: []};
default:
return {
code: "unknown",
diagnostics: [`Unsupported shape: ${shape.kind}`],
};
}
}