Watchdog Heartbeat

Provide observability and recovery awareness for a resident OpenClaw system. Verify process aliveness, heartbeat freshness, and workflow integrity.

Input

Required:

• service_list — list of monitored services and their expected health states
• health_endpoints — map of service → health check endpoint or method
• heartbeat_records — recent heartbeat timestamps per agent/skill
• workflow_status_records — current status of all active workflows
• restart_records — history of service restarts and recovery events

Output Schema

service_health_summary: {
  service: string
  status: "healthy" | "degraded" | "down" | "unknown"
  last_check: string      # ISO-8601
  latency_ms: number | null
  error: string | null
}[]

expired_heartbeat_list: {
  agent_or_skill: string
  last_heartbeat: string  # ISO-8601
  seconds_expired: number
  severity: "warning" | "critical"
}[]

stuck_workflow_list: {
  workflow_id: string
  workflow_name: string
  stuck_since: string     # ISO-8601
  stuck_duration_min: number
  last_progress: string | null
  severity: "warning" | "critical"
}[]

recovery_recommendation: {
  action: "restart" | "notify" | "escalate" | "no_action" | "degraded_mode"
  target: string
  reason: string
}[]

degraded_mode_recommendation: {
  affected_services: string[]
  degraded_features: string[]
  estimated_recovery_time: string | null
  user_impact: string
}

watchdog_log: {
  check_id: string
  check_time: string     # ISO-8601
  services_checked: number
  heartbeats_checked: number
  workflows_checked: number
  issues_found: number
  observability_gap: string[] | null
}

Rules

1. Process alive ≠ healthy. Check recent success, not just process existence.
2. Expired heartbeat triggers attention. Do not ignore stale heartbeats.
3. Stuck workflows must be explicitly surfaced. Don't let them disappear into silence.
4. Silent failure is unacceptable. If something fails and no one is notified, that's a system failure.
5. Distinguish warning from critical. Warning = may self-recover. Critical = requires intervention.

Heartbeat Expiry Thresholds

Workflow Stuck Thresholds

Recovery Actions

• no_action — within normal parameters
• notify — alert human, no automatic restart
• restart — attempt automatic restart
• escalate — human intervention required
• degraded_mode — reduce functionality, maintain partial service

Failure Handling

If monitoring data is incomplete:

• Set observability_gap with missing field names
• Report status = "unknown" for affected services
• Do not fabricate health states
• Recommend escalate if critical services have observability gaps