/** * reply Command - read an agent session's latest reply (Issue #3039) * * commandmate reply [--instance ] [--agent ] * [--since ] [--json] * * ## Why this exists next to ask / capture * * `ask` reads the reply to the turn IT sent. A supervisor that nudged a worker * with a plain `send` ("stop and report") had no way to read the answer short * of opening the tool's transcript file itself — which works for Claude, whose * path is computable from the cwd, and not for Command Code, whose project slug * is not. `capture` is no substitute: the pane also holds the nudge's own echo, * and a pane line cannot be attributed to a turn. * * The server already has the reply: the transcript readers write each turn into * the chat ledger keyed `-turn:`, and `ask` reads it from there. This * command is that read on its own, through the same shared rules * (`utils/reply-ledger`), so the two can never disagree about what a reply is. * * ## What it deliberately does not do * * - No pane fallback. A row off the screen is exactly what cannot be attributed * to a turn; returning it would re-introduce the problem this command solves. * Only transcript-reader rows count, for every tool. * - No wait. It is one read; "no reply yet" is an answer (exit 0, empty stdout, * `reply: null`), not an error, so a caller can poll or give up on its terms. */ import { Command } from 'commander'; /** The `--json` payload. `reply` is null when there is no reply yet. */ export interface ReplyJsonPayload { worktreeId: string; instanceId: string | null; cliToolId: string | null; reply: string | null; requestId: string | null; at: string | null; } export declare function createReplyCommand(): Command; //# sourceMappingURL=reply.d.ts.map