Recover or re-enter an existing Codex-owned current-thread run after the foreground thread was interrupted.
Use this skill when a foreground current-thread run already exists and the operator needs to recover the same-thread worker after an interruption.
Do not use this skill as the canonical next step for a fresh 루프 시작; $loop-control owns the normal start and resume chain.
Do not use this skill as a proxy for the app-server transport.
Recover an existing Codex-owned current-thread run without spawning nested codex exec sessions.
codex exec or codex exec resume.transport_mode = current-thread.runtime/live-state.json, runtime/round-phase.json, runtime/operator-surface.json, and committed round_summary.json files instead of guessing from chat history.runtime/current-thread-protocol.md when it exists and treat it as the operator checklist for the active run.npm run loop:continue -- --run-dir <run> --json as the machine contract for deciding whether the run is at a Codex checkpoint, a human stop, an external block, or a terminal state.attention_required = codex as an internal same-thread checkpoint, not a user-facing pause.attention_required is human or external, when the run is terminal, or when same-thread recovery cannot safely continue.npm run loop:continue -- --run-dir <run> --json.terminal, summarize the result and stop.human_stop, ask the blocking question and stop.external_stop, explain the environment block and stop.codex_checkpoint, read active_prompt_path, complete the checkpoint on the same thread, write active_response_path with the matching checkpoint_id, then call loop:continue again.codex_checkpoint.summary.json, runtime/live-state.json, and runtime/round-phase.json.round-contract.json, patch-request.json, and runtime/operator-surface.json before acting.loop:continue --json returns codex_checkpoint, complete the active checkpoint immediately instead of stopping at a phase boundary.