/**
 * cleanWebPageSourceForAI
 *
 * Strips non-UI noise from raw HTML so an AI agent can read page structure
 * and generate XPaths without drowning in scripts, styles, hidden subtrees,
 * inline base64, or analytics chatter.
 *
 * Aggressive trims (NEW-3):
 *   1. Removes non-UI tags entirely (script, style, link, meta, noscript,
 *      template, svg, canvas).
 *   2. Removes HTML comments.
 *   3. Removes hidden subtrees outright (was: marked with special-remark).
 *   4. Strips `on*` event-handler attributes.
 *   5. Strips inline `style` attributes after hidden detection.
 *   6. Strips analytics `data-*` attributes (gtm, ga, segment, track,
 *      analytics).
 *   7. Replaces base64 data URIs in `src`/`href` with `[base64 N bytes]`.
 *   8. Truncates text nodes longer than 200 chars.
 *   9. Recurses into iframe `srcdoc` content (cleans embedded HTML).
 *
 * Preserved (intentional, per design):
 *   - Everything outside `<body>` (`<head>`, `<title>`, `<html>` attrs).
 *   - All `class` attributes — sometimes the only stable selector.
 *
 * Hidden detection covers: `hidden` attr, `aria-hidden="true"`,
 * `type="hidden"`, inline `display:none`/`visibility:hidden`/`opacity:0`,
 * and `<dialog>` without an `open` attribute.
 *
 * Returns raw HTML unchanged on any parse/serialise error (fail-safe).
 *
 * Known limits (out of scope for this cleaner):
 *   - CSS-class-based hiding (external stylesheets, media queries) is
 *     undetectable from raw HTML. Closed dropdowns / modals styled via
 *     `.is-hidden { display:none }` survive cleaning.
 *   - Shadow DOM is not in `getPageSource()` output by default —
 *     Web Components' shadow trees never reach this function.
 *   - Native Android (UIAutomator) / iOS (XCUITest) XML is bypassed
 *     entirely via `isWebPageSource()`. A separate cleaner is needed
 *     to trim deep native layout trees.
 *   - `<svg>` roots are preserved when they carry locator-relevant
 *     attributes (`aria-label` / `role` / `id` / `data-test*`); inner
 *     geometry (`<path>` / `<circle>` / `<g>` / `<title>` etc.) is
 *     stripped either way. Decorative svg with no such attrs is
 *     removed entirely. SVGs with a `<title>` child but no locator
 *     attrs on the root will lose the title — wrap in `<button>` or
 *     hoist the title text to an `aria-label` if you need that surface.
 *   - iframe `srcdoc` recursion is verified synthetically; production
 *     pages using `srcdoc` (sandboxed previews, ad units, Storybook)
 *     are uncommon in our sample. Behavior on real-world srcdoc
 *     content is asserted-by-construction, not measured.
 */
export interface CleanStats {
    removedNonUI: number;
    removedComments: number;
    removedHiddenSubtrees: number;
    strippedEventHandlers: number;
    strippedInlineStyles: number;
    strippedAnalyticsAttrs: number;
    base64Replaced: number;
    truncatedTextNodes: number;
    iframesRecursed: number;
    svgsRemoved: number;
    svgRootsPreserved: number;
    originalLength: number;
    cleanedLength: number;
}
export interface CleanResult {
    html: string;
    stats: CleanStats;
}
export declare function cleanWebPageSourceForAI(html: string): CleanResult;
/**
 * Returns true if the page source looks like web HTML rather than
 * native Android/iOS XML.
 */
export declare function isWebPageSource(source: string): boolean;
/**
 * Pretty-print HTML/XML markup for GREPPABILITY: one element per line, indented by
 * nesting depth, text nodes trimmed onto their own line. Cleaned DOM is often written
 * as one enormous line (6000+ chars), which makes `grep -A/-B` flood and bounded reads
 * useless. This is NOT a spec-perfect formatter — it is a robust-enough pass so that
 * `grep -n "Section"` + a bounded read shows real structure the model can reason over
 * (e.g. which input sits under which section). Void / self-closing tags don't change
 * depth; comments and declarations print as-is. Returns the input unchanged when it
 * isn't markup.
 */
export declare function prettyPrintHtml(source: string): string;
export interface NativeCleanStats {
    originalLength: number;
    cleanedLength: number;
    /** Default-valued boolean attributes removed (focused="false" etc.). */
    attrsStripped: number;
    /** Repeated Android package="…" attributes removed (root keeps the first). */
    packageAttrsStripped: number;
}
export interface NativeCleanResult {
    xml: string;
    stats: NativeCleanStats;
}
export declare function cleanNativeXmlForAI(xml: string): NativeCleanResult;