/**
 * Evaluation Harness, external task-suite adapter (Terminal-Bench-style).
 *
 * A Terminal-Bench task suite is a DIRECTORY of tasks, each its own
 * subdirectory holding a task definition and a verification script:
 *
 *   <suiteDir>/
 *     <taskId>/
 *       task.json     { id, instruction, verify, timeoutMs? }
 *       verify.sh     exits 0 = pass, non-zero = fail
 *       ...           any fixtures the task needs
 *
 * This adapter runs each task through a REAL session, supplied by the caller as
 * an injectable `TaskSessionExecutor` (the seam a consumer fills with an actual
 * model-driven session operating in the task's working directory), and then
 * runs the task's verification script to decide pass/fail. It reports a
 * per-task result and an aggregate.
 *
 * SCOPE (honest). This is the adapter CONTRACT plus a default script verifier,
 * not a benchmark import: it does not bundle Terminal-Bench's task corpus, a
 * container runtime, or a specific model harness. The session executor and (if
 * desired) the verifier are injected, so the same contract drives the bundled
 * example suite in tests and a real suite + real session in a consumer.
 */
/** One task's on-disk definition (task.json). */
export interface TaskDefinition {
    /** Stable task id (used as the result key). Defaults to the task directory name when omitted. */
    readonly id: string;
    /** The instruction handed to the session. */
    readonly instruction: string;
    /** Path (relative to the task directory) to the verification script. Defaults to `verify.sh`. */
    readonly verify: string;
    /** Optional per-task wall-clock cap for the session executor, in ms. */
    readonly timeoutMs?: number;
}
/** What the session executor is given for one task. */
export interface TaskExecutionContext {
    readonly taskId: string;
    /** Absolute path to the task directory (definition + fixtures + verify script). */
    readonly taskDir: string;
    /** The instruction from the task definition. */
    readonly instruction: string;
    readonly timeoutMs?: number | undefined;
}
/** The outcome the session executor reports (whether the session ran to completion). */
export interface TaskSessionOutcome {
    readonly completed: boolean;
    readonly summary?: string;
}
/** The injectable "real session" seam: run one task's instruction in its directory. */
export type TaskSessionExecutor = (ctx: TaskExecutionContext) => Promise<TaskSessionOutcome>;
/** What the verifier is given. */
export interface TaskVerifyContext {
    readonly taskId: string;
    readonly taskDir: string;
    /** Absolute path to the verification script. */
    readonly verifyScript: string;
    readonly timeoutMs?: number | undefined;
}
/** The verifier's verdict. */
export interface TaskVerifyResult {
    readonly passed: boolean;
    readonly output: string;
}
/** Runs a task's verification script and decides pass/fail. Injectable; defaults to a shell runner. */
export type TaskVerifier = (ctx: TaskVerifyContext) => Promise<TaskVerifyResult>;
/** One task's result. */
export interface TaskResult {
    readonly taskId: string;
    readonly passed: boolean;
    readonly sessionCompleted: boolean;
    readonly verifierPassed: boolean;
    readonly verifierOutput: string;
    readonly reason: string;
    readonly durationMs: number;
}
/** The aggregate for a whole suite directory. */
export interface TaskSuiteResult {
    readonly suiteDir: string;
    readonly results: readonly TaskResult[];
    readonly passed: boolean;
    readonly total: number;
    readonly passedCount: number;
}
export interface RunTaskSuiteOptions {
    /** The real-session seam. Required, there is no faked default session. */
    readonly executor: TaskSessionExecutor;
    /** Verification runner. Defaults to {@link defaultShellVerifier} (runs the script, exit 0 = pass). */
    readonly verifier?: TaskVerifier;
}
/** Load and validate a task.json from a task directory. Throws a named error on a malformed definition. */
export declare function loadTaskDefinition(taskDir: string): TaskDefinition;
/** A discovered task: its parsed definition and the absolute directory it lives in. */
export interface DiscoveredTask {
    readonly definition: TaskDefinition;
    readonly taskDir: string;
}
/** Discover the task subdirectories (each containing a task.json) of a suite directory, sorted by id. */
export declare function discoverTasks(suiteDir: string): DiscoveredTask[];
/** Default verifier: run the verification script through `bash` and treat exit 0 as pass. */
export declare const defaultShellVerifier: TaskVerifier;
/**
 * Run every task in `suiteDir` through the injected session executor, then its
 * verification script, and report pass/fail per task. A task passes only when
 * the session completed AND its verifier passed.
 */
export declare function runTaskSuite(suiteDir: string, options: RunTaskSuiteOptions): Promise<TaskSuiteResult>;
/** Format a task-suite result as a per-task, one-line-each report. */
export declare function formatTaskSuiteResult(result: TaskSuiteResult): string;
//# sourceMappingURL=task-suite.d.ts.map