Claude Code skill that enforces task scope boundaries during agent execution, preventing drift beyond the approved plan.