Claude Code's '/goals' separates the agent that works from the one that decides it's done

Anthropic's Claude Code /goals adds a native evaluator model that checks task completion after every agent step, ending ...