Monitor and nudge stalled agents

Monitor polecat health, detect stalls, send nudges, escalate to mayor

Passive observation of worker agent health where escalation to a supervisor (mayor) is needed but the observer never modifies work or resolves conflicts.

• · rig name (agent pool to monitor)
• · patrol interval (seconds between health checks)
• · stall threshold (seconds of inactivity)
• · nudge count threshold (when to escalate)

• · nudge messages to stalled agents
• · stall escalations to mayor (durable mail)
• · health report summaries (RigHealthReport)

• · StateManager for agent list and hook queries
• · AgentMessage mail protocol
• · timestamp comparison (stall detection)
• · durable file storage for escalations

• · StateManager initialized
• · Multiple polecats active in rig
• · Hook timestamps updated regularly by agents
• · Mayor listening for escalation mail

• · Hook timestamps not updated: false positive stalls (nudge anyway)
• · Agent responds to nudge but doesn't reset timestamp: counted as stalled again
• · Mayor unreachable: escalation mail queues but never processed
• · Patrol interval too short: observer becomes bottleneck
• · Patrol interval too long: stalls detected late

• · Strictly read-only: observer never modifies agent work or state
• · Graduated escalation: warning→alert→critical with configurable thresholds
• · Patrol loop design avoids tight polling (default 5min interval)
• · Health report aggregation provides rig-wide visibility