Takes plans from your thoughts/shared/plans directory and executes them phase by phase, with built-in verification at each step. The interesting bit is the dual mode approach: small plans (3 or fewer tasks) run directly in your conversation, while larger ones use agent orchestration where each task gets its own agent that writes a handoff file. This keeps you compaction-resistant since handoffs persist on disk even if context gets trimmed. It won't let you proceed until automated checks pass, then explicitly pauses for manual verification before moving to the next phase. Includes a premortem check that blocks on high-severity risks, and can resume plan-agent sessions if you need to clarify something mid-implementation.
npx skills add https://github.com/parcadei/continuous-claude-v3 --skill implement_plan