/**
 * ExternalDataGuard
 *
 * Validates ALL external data before it enters LLM context.
 * Covers API responses, tool outputs, RAG results, file contents,
 * webhook payloads, and any other untrusted data source.
 *
 * This is an ARCHITECTURAL guard — it enforces boundaries on what
 * external data can reach the LLM, regardless of whether the LLM
 * itself has been compromised. Defense-in-depth at the data boundary.
 *
 * Threat model:
 * - Indirect prompt injection via API responses or RAG documents
 * - Context stuffing via oversized payloads
 * - Data exfiltration via embedded URLs in external content
 * - Secret/credential leakage through external data
 * - Poisoned data from compromised or unknown sources
 */
import { GuardLogger } from "../types";
export interface ExternalDataGuardConfig {
    /** Allowlist of trusted data sources (exact match or prefix) */
    allowedSources?: string[];
    /** Blocklist of known-bad data sources */
    blockedSources?: string[];
    /** Max characters of external content allowed (default: 50000) */
    maxContentLength?: number;
    /** Scan content for prompt injection patterns (default: true) */
    scanForInjection?: boolean;
    /** Detect leaked secrets, API keys, credentials (default: true) */
    scanForSecrets?: boolean;
    /** Detect data exfiltration URLs in content (default: true) */
    scanForExfiltration?: boolean;
    /** Require provenance metadata for all data (default: false) */
    requireProvenance?: boolean;
    /** Optional logger */
    logger?: GuardLogger;
}
export interface ExternalDataGuardResult {
    allowed: boolean;
    reason?: string;
    violations: string[];
    /** Identified data source */
    source?: string;
    /** Length of the content inspected */
    contentLength: number;
    /** Specific threat categories detected */
    threats: string[];
}
export interface DataProvenance {
    /** Where the data came from (URL, service name, file path, etc.) */
    source: string;
    /** Content type hint (e.g. "application/json", "text/html") */
    contentType?: string;
    /** When the data was retrieved (ISO string or epoch ms) */
    retrievedAt?: string | number;
    /** Max acceptable age in seconds before data is considered stale */
    maxAgeSec?: number;
}
export declare class ExternalDataGuard {
    private config;
    constructor(config?: ExternalDataGuardConfig);
    /**
     * Validate external data before it enters LLM context.
     *
     * @param content  - The raw external content (string or object)
     * @param provenance - Optional metadata about the data source
     */
    validate(content: string | Record<string, unknown>, provenance?: DataProvenance): ExternalDataGuardResult;
    /**
     * Validate a batch of external data items (e.g. multiple RAG chunks).
     * Returns individual results and a combined summary.
     */
    validateBatch(items: Array<{
        content: string | Record<string, unknown>;
        provenance?: DataProvenance;
    }>): {
        results: ExternalDataGuardResult[];
        allAllowed: boolean;
        totalThreats: number;
    };
    private isBlockedSource;
    private isAllowedSource;
    private log;
    private safeStringify;
}