Capability: heartbeat

A Temporal schedule wakes the harness every interval to poll the channels, snapshot the running tasks and sweep expired records.

What it is

The periodic pulse that makes waiting tasks progress without an inbound request and keeps the store within its retention policy. A schedule, not a process loop, so it survives the server process and is visible in Temporal. Lives in tiny_harness/service/heartbeat.py.

Current behaviour

• The worker SHALL ensure the agent's schedule exists, starting HeartbeatWorkflow every [heartbeat] interval (30 s by default) on the agent's task queue: tiny-harness-heartbeat for the default agent id, tiny-harness-heartbeat-<id> for any other. tiny-harness schedules delete removes the configured agent's schedule.
• monitor_snapshot SHALL scan only the running task workflows on the agent's own task queue.
• Each tick SHALL run poll_channels (forwarding pending channel items to their tasks), monitor_snapshot (the task states and pending items) and retention_sweep (deleting expired store rows per kind).
• WHEN a tick finds a task waiting on input whose reply has arrived THEN the task SHALL be resumed.
• The snapshot SHALL be exposed to the API layer at GET /_monitor and to hooks through the heartbeat.tick operation.
• WHEN a tick fails THEN the failure SHALL be logged and the next tick run on schedule.

Design

design.md § Heartbeat.

History

Work item	What changed	Links
issue-3	The heartbeat workflow, schedule and activities (Layer 5)	spec, PR #11
issue-36	One schedule per agent; the scan filtered to the agent's queue	spec, PR #37, decision-009