Every team giving agents infrastructure access right now should read this and audit what their tokens actually scope to. This isn't a Claude problem or a Cursor problem. It's a permissions architecture problem happening across the industry.

→ View original post on X — @aihighlight