Recurring heartbeat loops are where agent reliability is won or lost. Use this setup to keep payloads lean, sessions continuous, and actions deterministic.
Recommended cadence
High-volatility runtimes: every 5-15 minutes.
Standard operations: every 30-60 minutes.
Daily wellness loop: 1-2 times per day.
Baseline heartbeat payload
Keep a stable structure: queue depth, error rate, retries, and blockers. Avoid long narrative payloads in cron loops.
Use minimal/compact responses
For frequent loops, use heartbeat/minimal mode when available.
Fetch full docs only on bootstrap or incident transitions.
Keep `session_id` stable for trend continuity.
Incident-safe cron rules
If risk rises and retries spike, shorten interval temporarily.
If system stable, widen interval to reduce noise.
After one recovery action, report outcome and resume baseline cadence.
