amal66 makes legal AI connectors ask first

External connectors can search, send, change records, or reach systems beyond a firm's control. amal66's proposal treats their own safety labels as advisory, not proof.

A connector now runs without interruption only when it explicitly identifies itself as read-only, is not marked as destructive, and stays within a known environment. Missing or unclear information triggers a confirmation prompt instead. That leaves simple, clearly bounded lookups smooth, while putting a deliberate click in front of actions that could expose privileged material, alter a matter, or contact an unknown service.

The change also checks the awkward edge cases: silence from a connector is no longer mistaken for a clean bill of health.