swarm-monitor-loop - Claude MCP Skill

# Swarm Monitor Loop Agent

## Goal

Continuously monitor the distributed agent swarm and report status every 2-3 minutes without stopping.

## Process

1. Check Worker Heartbeats

- Search for swarm-heartbeat

- memories

- Verify timestamps are recent within 3 minutes

- Flag dead workers

2. Check Task Status

- Search for all swarm-task

- memories

- Count: pending, claimed, in-progress, completed, failed

- Identify stuck tasks no progress in 10+ minutes

3. Check for Activity

- Search for swarm-claim

- memories new claims

- Search for swarm-progress

- memories progress updates

- Search for swarm-result

- memories completed work

4. Generate Status Report

- Worker health summary

- Task completion metrics

- Recent activity log

- Issues or alerts

5. Take Action

- Collect and validate completed results

- Reassign failed/stuck tasks

- Create new tasks if queue is low

- Update orchestrator session log

6. Continue Loop

- Immediately start next monitoring cycle

- Dont stop, dont pause, keep monitoring

## Continuous Operation

This agent runs in a loop. After each report, it IMMEDIATELY begins the next check. No pausing, no waiting for instructions.

## Commands to Execute

bash

# Worker statusmcp__DollhouseMCP__search_portfolio --query swarm-heartbeat --type memories

# Task status  mcp__DollhouseMCP__search_portfolio --query swarm-task --type memories

# Claimsmcp__DollhouseMCP__search_portfolio --query swarm-claim --type memories

# Progressmcp__DollhouseMCP__search_portfolio --query swarm-progress --type memories

# Resultsmcp__DollhouseMCP__search_portfolio --query swarm-result --type memories

Execute these commands every 2 minutes and report findings.