/**
 * `useToolAction` — the write, in code-land (blueprint §5.4).
 *
 * The same door as the read (`POST /apps/:appId/call`), the same guard-bound
 * caller, the same `ToolOutcome`. Every mount site in this repo already shapes
 * an action as one line —
 * `onAction={({action, payload}) => client.apps.call(appId, action, payload ?? {})}`
 * — and this is that line for code.
 *
 * Two laws it carries so the app author never has to:
 *   - a non-ok outcome is a CONTAINED notice, never a crash (the renderer's
 *     `runAction`, ui/tree/renderer.tsx);
 *   - a SUCCESSFUL action refreshes the screen's queries (§6.3 law 2).
 */
import type { Json, ToolOutcome } from "../../core/index.js";
export interface ToolAction {
    /** Run it. Resolves with the outcome; never throws. */
    run(args?: Json): Promise<ToolOutcome>;
    /** A run is in flight (its query refresh included). */
    pending: boolean;
    /** The last NON-OK outcome, for the contained notice — `blocked` says why,
     *  `pending-approval` and `connect-required` carry their own affordance. A
     *  success clears it. */
    outcome: ToolOutcome | undefined;
}
export declare function useToolAction(ref: string): ToolAction;