Launch profiles: save reusable launch setups — model, reasoning effort, permission mode, CLI flags, environment, and an optional agent — then apply one per pane from the launcher, the full editor, or the atrium CLI, with a quick-switch popover and a session-footer chip.
Fixed
Agents now reliably receive their assigned agent and skills at session start, instead of silently dropping them when the injected context was large.
A running workspace command's sidebar button now reverts from stop when the command exits on its own.