CAPS is the safety pillar of Foundation — the halt switches, autonomy ceilings, confidence budgets, and rollback paths every AI agent operates under.