/**
 * CommandVerifyGate is the real VerifyGate (ADR 0001): it runs the project's
 * configured deterministic gate commands (e.g. "bun run typecheck", "bun test")
 * in the issue's worktree via the injected ProcessRunner, and returns each
 * command's exit code. run-step fuses these MEASURED exit codes with the
 * verifier LLM's verdict — a non-zero exit fails the stage regardless of what
 * the LLM claimed.
 *
 * All commands are run (no short-circuit) so the worker-retry feedback can
 * report every failing gate, not just the first. Each command string is split
 * into argv on whitespace — sufficient for the simple "binary subcommand flags"
 * shape these gate commands take; commands needing shell features should be
 * wrapped in a script the user points the gate at.
 *
 * The ProcessRunner seam keeps this unit-testable with a fake; the live binding
 * (NodeProcessRunner) shells out for real.
 */
import type { ProcessRunner } from "../ports/agent.ts";
import type { VerifyGate } from "../app/run-step.ts";
import type { GateOutcome } from "../domain/measured-verdict.ts";
export declare class CommandVerifyGate implements VerifyGate {
    private readonly commands;
    private readonly runner;
    constructor(commands: readonly string[], runner: ProcessRunner);
    run(worktree: string): Promise<GateOutcome[]>;
}