Add a reflection layer to any tool-using Claude agent. Enforces think-before-act, verify-after-act patterns that catch errors early in multi-step tool chains. Especially valuable for agents with write or delete access. Free template — copy and paste into ChatGPT, Claude, or Copilot.