This is the orchestration engine for multi-step technical implementations. It reads plans from thoughts/shared/plans, decides whether to implement directly or spawn sub-agents per task, and creates handoffs that survive context compaction. The formal spec is dense but the practical bit is straightforward: it handles verification gates (automated then manual), stops on mismatches to ask you how to proceed, and maintains a ledger so work continues after compaction. You'd reach for this when you have a 4+ task plan and need agents to hand off context to each other rather than doing everything in one shot. The justification logic chains and temporal constraints are probably overkill for most cases, but the handoff persistence is legitimately useful.
npx skills add https://github.com/parcadei/continuous-claude-v3 --skill implement_plan_micro