Anthropic added the /goal command to Claude Code in v2.1.139. You set a completion condition; the agent keeps working across turns; a second model reads the transcript and decides whether the condition was met. It is the built-in version of the keep-going loops people have been hand-rolling for
