Skill1.3k repo starsupdated today

patrol-monitoring

Patrol Monitoring implements continuous health supervision for multi-agent systems using a Deacon/Witness hierarchical pattern that detects stuck or unresponsive agents through heartbeat monitoring and executes automated recovery actions like restarts, reassignments, or escalations. Use this skill during active multi-agent workflows, particularly long-running convoy executions where agents may become unresponsive and require automated intervention without human oversight.

View source Repository: babysitter

Install in Claude Code

Copy

git clone --depth 1 https://github.com/a5c-ai/babysitter /tmp/patrol-monitoring && cp -r /tmp/patrol-monitoring/library/methodologies/gastown/skills/patrol-monitoring ~/.claude/skills/patrol-monitoring

Then start a new Claude Code session; the skill loads automatically.

Definition

SKILL.md

# Patrol Monitoring

## Overview

Continuous monitoring using Gas Town's Deacon/Witness pattern. The Deacon supervises overall health, the Witness manages per-rig agent lifecycle, and the Boot (Dog) watches the Deacon itself.

## When to Use

- During active convoy execution
- When agents may become stuck or unresponsive
- For long-running multi-agent workflows
- When automated recovery is desired

## Process

1. **Health check** all active agents and convoys
2. **Detect** stuck or unresponsive agents via heartbeats
3. **Recover** - restart, reassign, or escalate as needed
4. **Report** patrol findings with trend analysis

## Monitoring Roles

- **Deacon**: Daemon supervisor, monitors overall health
- **Witness**: Per-rig lifecycle manager for workers
- **Boot (Dog)**: Watches the Deacon every 5 minutes

## Recovery Modes

- **restart**: Restart the stuck agent session
- **reassign**: Move beads to a different agent
- **escalate**: Alert human for manual intervention

## Tool Use

Invoke via babysitter process: `methodologies/gastown/gastown-patrol`