/**
 * Ingest-depth step extensions: the unified `extract.text` format dispatch (PDF text layer,
 * DOCX via mammoth, XLSX via ExcelJS, ODT/ODS via ODF XML, PPTX via slide XML, images via the
 * OCR engine) and format-aware `extract.metadata`.
 *
 * @remarks
 * Internal sibling of the `@nhtio/adk/batteries/media` entry. Ported from the source server's
 * four extractors, collapsed into one verb that routes on the resolved media's format family
 * (the single biggest order-out-of-chaos win — the model says `extract text` and doesn't care
 * what the container is). OCR runs through the configured `ocr` engine when the input is an
 * image, or when a PDF has no text layer with `ocr=auto`, or always with `ocr=force`.
 */
import type { StepImpl } from "../runtime";
/**
 * `extract.text` — full format dispatch. Routes by MIME, with OCR engine fallback for images
 * and force-OCR for any input via `ocr=force`.
 */
export declare const extractTextDeepStep: StepImpl;
/** `extract.metadata` — format-aware metadata (page counts, sheet/slide inventories). */
export declare const extractMetadataDeepStep: StepImpl;
/** The ingest step registry fragment (overrides the Phase 0 native-text-only versions). */
export declare const INGEST_STEPS: ReadonlyArray<[
    string,
    StepImpl
]>;