List or search persistent local Codex threads through thread/list, or read one thread through thread/read. This does not reconstruct live Bridge events.

Start a persistent Codex thread and turn, or resume an existing thread and start a turn. Prefer continuing the same native thread when its context remains useful, but a fresh thread is allowed; thread_id is not a permanent task identity. Returns as soon as turn/start is accepted; observe separately for events and completion.

Read bounded incremental sanitized Bridge runtime events, pending requests, and terminal output for a thread. Optional wait_ms performs one bounded event-driven wait only when the live turn is active and the current snapshot has nothing useful; it is not polling or stall detection. After Bridge process loss, falls back to persistent thread/read history and marks live state unreconstructable. A long interval with no new command or output can still mean Codex is actively reasoning; absence of new command activity alone is not evidence of a stall. When actively supervising an in-progress turn, use repeated bounded-wait observe calls until terminal unless the user explicitly pauses or stops; do not end supervision merely because one snapshot is inProgress. After every wake or deadline return, inspect the newly available events/state and decide whether steer, respond, or interruption is needed before starting the next bounded wait.

Append text to the same active Codex turn using turn/steer with an expected turn-id precondition. This does not create a new turn. Do not steer merely because reasoning is taking a long time or no new command has appeared; steer only for a semantic redirect or correction based on new evidence or changed user intent.

Answer one currently pending app-server request by its original raw JSON-RPC id and exact thread/method scope. Supports only command/file approval methods with concrete response contracts and item/tool/requestUserInput; unsupported methods remain pending and observable.

Directly request turn/interrupt for the specified active Codex thread and turn. It does not stop or restart the Bridge or Codex app-server processes.

Optional, bounded supervisor cognition memory keyed to one native Codex thread_id; the key is not a permanent task identity and does not require future work to remain on that thread. Use it to protect the original goal, constraints, and acceptance plus concise supervisor state during long or complex supervision when context dilution or goal drift makes an external anchor worthwhile. Initialization is not tied to crossing a ChatGPT window or round, starting another Codex turn, or switching native threads; initialize early when a task is already expected to be sufficiently long or complex for that protection. Do not use for one-shot work, and do not turn duration into a hard threshold: elapsed time, observe/poll count, token count, or mere silence are not automatic triggers. Later updates remain semantic-event driven and require a material change in understanding or root cause, constraint or scope interpretation, steering decision, user-authorized amendment or effective goal, or acceptance judgment or an explicit decision not to accept yet. Before final acceptance of a checkpointed task, read it once to re-anchor the original goal, constraints, acceptance, and current supervisor frame. This tool is optional and uncoupled from all other tools. Store concise supervisor summaries only; never prompts, transcripts, raw events, command output, final answers, or raw event streams. Updates preserve only immutable original plus bounded previous/current supervisor state.