/**
 * ContentSanitizer — fail-closed wrapper around damcrawler's `sanitize`
 * (spec-20260717-seo-improver-builder, Sprint 6; ADR-11).
 *
 * ADR-11's decision is to sanitize untrusted crawled/scraped free-text
 * INSIDE the damcrawler adapter, at the network→in-process boundary,
 * BEFORE it ever reaches `SeoAnalyzer` (which `JSON.stringify`s rows
 * verbatim into its LLM system prompt — the prompt-injection surface this
 * sprint closes). This class does no importing/loading of damcrawler
 * itself — the caller (`DamcrawlerCrawlEngine`, already holding the loaded
 * module behind the `site-crawl` axis) injects the narrow `SanitizeFn`.
 *
 * IMPORTANT SDK correction (see sprint briefing §0/§9.1): damcrawler's
 * `sanitizeWithReport(raw, sourceUrl?)` returns a bare `string` — it does
 * NOT yield `hadThreats`. The only damcrawler export exposing `hadThreats`
 * is `sanitize(raw, options?): { content; threats; hadThreats }`
 * (`damcrawler/src/lib/sanitize.ts:68-79,89`). This wrapper's `SanitizeFn`
 * type matches THAT signature — callers must inject `dam.sanitize`, not
 * `dam.sanitizeWithReport`.
 */
import { type Logger } from "../utils/logger.js";
/**
 * The narrow sanitize surface this module depends on — a SUBSET of
 * damcrawler's `SanitizeResult` (`sanitize.ts:68-79`, `threats` omitted,
 * this module never needs the detailed threat breakdown). Injected so
 * tests never require the real `damcrawler` dependency.
 */
export type SanitizeFn = (raw: string, options?: {
    sourceUrl?: string;
}) => {
    content: string;
    hadThreats: boolean;
};
/**
 * Fail-closed content sanitizer. `clean()` never throws: a successful
 * sanitize call logs+strips on `hadThreats`; a sanitize-function error
 * drops the text entirely (`content: ""`, `hadThreats: true`) rather than
 * letting raw, unsanitized attacker-controlled text pass through.
 */
export declare class ContentSanitizer {
    private readonly sanitizeFn;
    private readonly log;
    constructor(sanitizeFn: SanitizeFn, log?: Pick<Logger, "warn">);
    /**
     * Sanitizes `text` sourced from `url`. `hadThreats` is logged (warn, for
     * audit) and the STRIPPED `content` is returned either way — callers
     * must never fall back to the raw input. A thrown sanitize error is
     * treated as fail-closed: the page's free-text is dropped rather than
     * passed through unsanitized.
     */
    clean(text: string, url: string): {
        content: string;
        hadThreats: boolean;
    };
}
//# sourceMappingURL=content-sanitizer.d.ts.map