multi-agent-autonomous-loop-pattern

What problem does it solve?

This Skill eliminates the risk and operational overhead of running long-running Claude Code autonomous work by coordinating parallel agents, persisting state, and safely triggering the next round.

Core Features & Use Cases

• 5-agent parallel autonomous rounds: Distributes work across meta, feature addition, verification, documentation, and options to keep progress moving without manual babysitting.
• ScheduleWakeup-driven long execution: Uses timed wakeups (40–50 minutes) with a dynamic loop prompt to sustain multi-hour operations.
• Atomic commit hook to prevent races: Blocks inconsistent “partial registry/skill” commits when multiple agents update install.sh and new SKILL.md concurrently, preventing data corruption and removing the need for force pushes.
• Safety guardrails for operations: Treats cutover/production changes as “write-only” and routes actual application to explicit user approval.

Quick Start

Run multi-agent autonomous loop mode by requesting: “Start the full autonomous loop for this long-running project in background, using ScheduleWakeup and the atomic commit hook safety rules.”