feat(work): implement synchronous readback gate and cognitive true loopback engine

bin/super-cli.py

@@ -1493,6 +1493,8 @@ def cmd_work(args):
         box_work.cmd_menu(args)
     elif action == "cognitive":
         box_work.cmd_cognitive(args)
+    elif action == "loopback":
+        box_work.cmd_loopback(args)
     else:
         box_work.cmd_status(args)
 
@@ -6998,6 +7000,7 @@ def build_parser():
     p_w_start.add_argument("--goal", help="Optional detailed goal description")
     p_w_start.add_argument("--force", action="store_true", help="Bypass pre-flight health gate")
     p_w_start.add_argument("--no-heal", action="store_true", help="Fail immediately without attempting auto-heal if pre-flight checks fail")
+    p_w_start.add_argument("--no-readback", action="store_true", help="Bypass synchronous 45s readback gate")
     p_w_assign = work_sub.add_parser("assign", parents=[common], help="Assign existing ticket to an agent")
     p_w_assign.add_argument("issue", type=int, help="Issue number (e.g. 215)")
     p_w_assign.add_argument("--to", dest="agent", required=True, help="Agent username")
@@ -7013,6 +7016,8 @@ def build_parser():
     p_w_menu.add_argument("tab", nargs="?", default="activity", choices=["activity", "upcoming", "approvals", "identity", "all"], help="Menu tab to view")
     p_w_cog = work_sub.add_parser("cognitive", parents=[common], help="Probe real-time cognitive sensor (thinking, generating, sidechats)")
     p_w_cog.add_argument("agents", nargs="*", help="Optional agent usernames")
+    p_w_loop = work_sub.add_parser("loopback", parents=[common], help="Sweep active tickets with 3-tier loopback escalation")
+    p_w_loop.add_argument("--dry-run", action="store_true", help="Inspect without modifying tickets or sending DMs")
 
     p_tasks = subparsers.add_parser("tasks", parents=[common], help="Agent work queue: pending/claimed/done files (distinct from scheduled jobs)")
     p_tasks.add_argument("--dir", default=None, help="Task queue dir (default: fleet/tasks)")