Rlm Heartbeat

RLM Heartbeat

RLM heartbeats are internal recurring prompts for the current agent session. They are separate from the user's visible /heartbeat: this skill cannot read, replace, pause, resume, or clear that user-level heartbeat.

Call directly from IPython:

await rlm_heartbeat.create("check test progress", interval="5m", label="tests")
await rlm_heartbeat.create("watch build", delivery_mode="follow_up")
await rlm_heartbeat.list()
await rlm_heartbeat.update("job-id", status="pause")
await rlm_heartbeat.delete("job-id")

API

• await rlm_heartbeat.list(include_inactive=False) — list this session's internal RLM heartbeats. By default this includes active and paused entries.
• await rlm_heartbeat.create(instruction, interval=None, label=None, delivery_mode=None) — create a recurring heartbeat for this session. The default interval is every 5 minutes. Multiple RLM heartbeats may run at once; use labels to distinguish them. delivery_mode is "steer" (default) or "follow_up".
• await rlm_heartbeat.update(id, instruction=None, interval=None, label=None, status=None, delivery_mode=None) — update one RLM heartbeat by id. status may be "pause" or "resume"; delivery_mode may be "steer" or "follow_up".
• await rlm_heartbeat.delete(id) — cancel one RLM heartbeat by id.

Delivery mode

Each heartbeat has a delivery mode controlling how the scheduled prompt reaches the session when it is busy:

• steer (default): interrupt the current turn so the heartbeat runs promptly.
• follow_up: wait for the current turn to finish before running the heartbeat.

Rules

• Use this when the user asks you to start, create, schedule, or manage your own heartbeat without explicitly referring to /heartbeat.
• Use this only for agent-internal recurring checks and long-running task coordination.
• Do not use this skill to satisfy a user's request to configure /heartbeat; that is a separate user-level surface.
• Keep heartbeat instructions specific and actionable so each recurring turn knows exactly what to inspect or continue.