Creating an API token used to mean ticking scopes one by one in a flat list. The dialog now shows one row per resource (organizations, projects, secrets, issues, sites, agents, tokens) and you pick a single level on each: none, read, write, and where it exists rotate, run or admin.

Levels are cumulative. Picking write on secrets issues secrets:read and secrets:write, so a token never ends up with a write scope and no matching read. Three presets fill every row at once: read only, read and write (no admin-class scope), and full. The exact scope list is shown under the rows before you create the token.

The owner choice is now two cards instead of a dropdown: a token that belongs to you stops working when your access does, one that belongs to the organization keeps working and is the one to use for a pipeline. Each card shows who that is: your avatar and name, or the organization's.

Tokens created through the API are unaffected, and existing tokens keep their scopes.