shepherd

Supervise coordinator runs across context exhaustion with respawns and handoffs.

5|Updated Mar 10, 2026

Search Tags:#automation #resilience #tmux #handoff #coordinator #supervisor #keepalive

Ready to run

npx skills add https://github.com/humblemuzzu/ghosttyyy --skill shepherd-humblemuzzu

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: shepherd
Source: https://github.com/humblemuzzu/ghosttyyy/tree/main/pi-setup/config-skills/shepherd
Command: npx skills add https://github.com/humblemuzzu/ghosttyyy --skill shepherd-humblemuzzu

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

watchdog-style supervise across context exhaustion and agent deaths to keep long-running coordinator runs alive, enabling seamless handoffs and respawns.

Core Features & Use Cases

Keeps coordinators alive by monitoring liveness, pinging state, and respawning on death.
Orchestrates handoffs to successors when context limits are reached, preserving continuity.
Guards against premature "done" claims by validating state and handling edge cases.

Quick Start

Spawn a shepherd to supervise a coordinator thread and keep it alive across context exhaustion.

Frequently Asked Questions about shepherd

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I keep autonomous coordinator runs alive across context exhaustion?▼

A keepalive supervisor monitors liveness, pings state, and respawns agents on death to keep autonomous coordinator runs alive across context exhaustion. This ensures long-running tasks span multiple panes or sessions seamlessly.

What is the best way to orchestrate handoffs when an agent reaches its context limit?▼

The best way to orchestrate handoffs when an agent reaches its context limit is using a watchdog supervisor that coordinates transitions to successors with persistent state tracking. This preserves continuity across context resets and agent deaths.

How does a tmux supervisor validate state and prevent premature task completion?▼

A tmux supervisor validates state by pinging the target pane and checking conditions before accepting completion. It guards against premature done claims by handling edge cases and triggering respawns when state validation fails.

Can I use a keepalive supervisor to manage long-running coordination tasks across multiple sessions?▼

Yes, you can use a keepalive supervisor to manage long-running coordination tasks across multiple sessions. It supervises target panes, tracks persistent state, and coordinates handoffs to keep tasks running despite context resets.

Why do my autonomous coordinator runs stop when experiencing agent deaths and context resets?▼

Autonomous coordinator runs stop during agent deaths and context resets because they lack a watchdog-style supervisor to respawn them. Without persistent state tracking and handoff orchestration, the process cannot recover automatically.