Skip to content
Run, stop, steer, and queue messages
On this page

Run, stop, steer, and queue messages

Use the composer and queue to control when a message runs. Sending a follow-up during active work is not always the same as steering the current provider turn.

Start, wait, and stop

  1. Check the folder, model, and permission mode before sending a new task.
  2. Send the message and follow the transcript’s activity state.
  3. If the agent requests approval or a decision, answer its card.
  4. Use Stop when you intend to interrupt active work, then wait for the resulting state before starting a replacement request.

Stopping is not an undo operation. Commands or file changes that completed before interruption remain and should be inspected. A cancelled, parked, waiting, or error state explains different conditions; use the displayed message rather than assuming every non-running task completed successfully.

Queue a follow-up

  1. While work is active, submit a follow-up and inspect whether Rudr queued it or sent it through a supported native steering path.
  2. Open Queued prompts and select the intended card. Use Expand for a list when several prompts are queued.
  3. Choose Edit, then Save or Cancel, to change a queued request. Recall pulls it back into the composer; Remove discards it.
  4. Leave ordinary queued work to run when the current turn finishes, or choose Interrupt & send to stop the running turn and send that prompt now.
  5. When no turn is running, the action reads Send now. Confirm the prompt and destination before using it.

Interrupt & send is an interruption followed by another request, not native mid-turn steering. Native steering is available only for supported harness/version combinations; a queue entry does not prove the current provider turn received the instruction immediately.

Recovered drafts and queued work may require confirmation after interruption or provider continuation. Verify their text, attachments, folder, and model before explicitly sending or confirming them. Closing the queue panel only changes its visibility; use Remove when you intend to discard a queued request.

Recover without duplicating work

If the connection fails near submission, inspect the latest transcript and session state before retrying. An uncertain acceptance can mean the provider started work even though Rudr did not receive a clean acknowledgement. Avoid sending the same mutating instruction repeatedly until you know whether it ran.

For the next step, use activity details, continuation recovery, or cost warnings.

Instructions have been reviewed against the application source; native setup and account walkthroughs remain to be verified. Platform and provider limitations are described above.

Run, stop, steer, and queue messages — Rudr