Waiting workers always start. A task queued behind another (depends_on) could stay queued forever if brindle missed the moment its dependency merged. brindle now checks again whenever the supervisor's turn ends, and starts every queued task whose dependencies have all merged. Sessions already stuck this way pick up again after you upgrade.
Failed starts are reported. If a queued worker can't start, the supervisor is told why, instead of the task being skipped without a word.
One cloud sign-in per expired login. When a Google, AWS SSO or Azure login expired, every pane could open its own browser sign-in, and signing out or a network blip could trigger one too. Now only a real expiry opens a sign-in, just one at a time across the machine, and only when you're at the keyboard. The session shows PAUSED until you sign in again, then everything picks up where it stopped.