Boundary Bench: test your coding-agent permissions against SWE-MISBEHAVIOR

CLI replaying boundary-violation scenarios against your Claude Code or Codex permission config. Teams pay for CI runs.