SafeAct-Env: An OpenEnv environment for training agents to identify and avoid irreversible actions across 5 real-world operational domains

Ground-first reflex for Claude Code: the agent judges complexity from its own training, grounds before acting, and is hard-blocked from the irreversible step until it has.