This is part of a multi-agent orchestration system where tasks get spawned off to fresh Claude instances that work independently and hand off context when done. Each agent gets a slice of a larger plan, implements it using strict TDD (tests must fail first, no exceptions), then writes a structured markdown handoff documenting what changed, decisions made, and patterns for the next agent. The interesting bit is how it handles context boundaries: agents read the previous handoff, do their work, save their own handoff, and exit. It enforces the red-green-refactor cycle religiously and includes quality checks with qlty. You'd use this when you want to break down a big implementation into isolated chunks that different agent instances execute sequentially, each one starting fresh but informed by the last.
npx skills add https://github.com/parcadei/continuous-claude-v3 --skill implement_task