import { type AppDocument, type ScreenAssembler, type ScreenOutcome, type ScreenRequest } from "../../contract/index.js";
import { type AppsRuntime } from "../runtime/runtime.js";
/**
 * What a scripted assembly run answers: the `app.tsx` it saves, or one of the
 * two non-assembling outcomes verbatim.
 */
export type AssemblerAnswer = string | ScreenOutcome;
/**
 * A screen assembler that really assembles.
 *
 * There is one engine, so a test that exercises create or edit needs something
 * in the `screen` slot. This is not a stub of the apps side: the screen goes
 * through the real checks floor — the five-stage component gauntlet, whose own
 * `ok` upserts the row and stores the source — so the row, the history entry,
 * the guard decision and the query execution are all the real ones. The only
 * thing standing in for a live agent is the choice of screen.
 *
 * `answer` is handed the request the runtime made and the app as it stands
 * (`null` on a create), which is what makes an EDIT expressible: the shipped
 * screen agent opens the screen, rewrites it and saves the whole thing, and a
 * fixture that reads `current` does the same.
 */
export declare const scriptedAssembler: (
/** A getter, because the slot is filled at compose time and the runtime it
 *  writes through is what composing RETURNS — the same knot `packages/vendo`
 *  ties. */
runtime: () => AppsRuntime, answer: (request: ScreenRequest, current: AppDocument | null) => AssemblerAnswer | Promise<AssemblerAnswer>) => ScreenAssembler;
/** The one-screen case: every ask, create or edit, saves this `app.tsx`. */
export declare const authoringAssembler: (runtime: () => AppsRuntime, screen: string) => ScreenAssembler;
/** What a scripted screen run answers: the `app.tsx` it saves, or one of the two
 *  non-assembling outcomes verbatim. */
export type ScreenAnswer = string | ScreenOutcome;
/**
 * A COMPONENT screen assembler that really assembles.
 *
 * There is one engine, so a test that exercises create or edit needs something
 * in the `screen` slot. This is not a stub of the apps side: `answer` returns the
 * `app.tsx` a screen agent would have written and it lands through
 * `authoredScreen` — the door the shipped floor's paint half calls — so the row,
 * the version and the CAS bracket are all the real ones. The only thing standing
 * in for a live agent is the choice of source.
 *
 * `answer` is handed the request the runtime made and the app as it stands
 * (`null` on a create), which is what makes an EDIT expressible: the shipped
 * screen agent opens the document, rewrites it and saves the whole thing.
 */
export declare const scriptedScreenAssembler: (
/** A getter, because the slot is filled at compose time and the runtime it
 *  writes through is what composing RETURNS — the same knot `packages/vendo`
 *  ties. */
runtime: () => AppsRuntime, answer: (request: ScreenRequest, current: AppDocument | null) => ScreenAnswer | Promise<ScreenAnswer>) => ScreenAssembler;
//# sourceMappingURL=screen-assembler.d.ts.map