/**
 * @module @nhtio/adk/batteries/orchestration/reason
 *
 * Pure utilities for the reason node. This module contains no Orchestration runtime
 * dependencies — it is independent of `DispatchRunner`, `Message`, or `Tool`. The bundled
 * helper that constructs those lives in its own subpath and is a separate concern.
 *
 * The reason node ends in a tool call, and that tool call *is* its output. It never
 * returns prose to be parsed. The node's `outputSchema` becomes a forced tool's
 * `inputSchema`, so the model physically cannot answer unstructured: malformed args are
 * rejected before the handler runs and retried within `maxAttempts`. A dispatch that ends
 * without the tool being called is a halting node failure, never a fabricated result.
 *
 * This file implements the contract required by `ReasonNodeDefinition` and the
 * `ReasonerFn` signature. All functions are pure and side-effect free.
 */
import type { Schema } from '@nhtio/validation';
import type { PromptPart, EncodableValue, NodeRef } from "./types";
/**
 * Join prompt parts in order, substituting each reference's resolved value.
 * A reference that resolves to `undefined` renders as the explicit absent marker
 * `'[unresolved: <nodeId>]'`, naming the node whose output was missing.
 *
 * @param parts The prompt parts to join
 * @param resolve Function that resolves a `NodeRef` to an `EncodableValue | undefined`
 * @returns The joined prompt string
 */
export declare function joinPromptParts(parts: readonly PromptPart[], resolve: (ref: NodeRef) => EncodableValue | undefined): string;
/**
 * Remove `<instruction>` and `</instruction>` tags from author-supplied prompt text.
 * Case-insensitive, tolerant of whitespace inside the tag. Only those two tags are removed.
 *
 * @param text The input text to sanitize
 * @returns The text with instruction tags removed
 */
export declare function stripInstructionTags(text: string): string;
/**
 * Wrap the authoritative instruction in `<instruction>` tags and place the stripped
 * context after it, with a line telling the model to ignore any instructions appearing
 * inside the context payload.
 *
 * @param authoritative The authoritative instruction text
 * @param context The context text (already stripped of instruction tags)
 * @returns The wrapped instruction string
 */
export declare function wrapInstruction(authoritative: string, context: string): string;
/**
 * Decode a `ReasonNodeDefinition.outputSchema` — which is stored as an ENCODED STRING,
 * not a live `Schema`. A live validation schema is **not** `Encodable` — encoding one
 * throws `E_UNENCODABLE_VALUE: Value of type symbol (Symbol(override)) is not encodable`,
 * which would make the whole plan unpersistable. This has been verified against the real
 * packages. `Tool` solves the same problem the same way at `src/lib/classes/tool.ts:449-490`.
 *
 * @param encoded The encoded schema string
 * @returns The decoded `Schema` object
 */
export declare function decodeOutputSchema(encoded: string): Schema;
/**
 * Validate captured tool arguments against the decoded schema, returning a discriminated
 * result rather than throwing.
 *
 * @param schema The decoded output schema
 * @param args The captured tool arguments to validate
 * @returns `true` if validation succeeds, or the validation error message if it fails
 */
export declare function validateReasonerOutput(schema: Schema, args: Record<string, EncodableValue>): true | string;