Cursor CLI's --trust flag works in interactive sessions

The July 20, 2026 Cursor CLI release lifts a restriction on --trust: it previously only worked in headless mode. Passed in an interactive session now, it trusts the workspace up front and skips the trust dialog, recording the same saved trust decision the dialog writes when you accept it manually.

The model catalog stays fresh in long sessions

The CLI now refetches the model catalog every 10 minutes in the background. Previously, a model released after your session started wouldn't show up in /model or the model slash shortcuts until you restarted. If a refresh fails, the CLI keeps the current catalog and retries on the next interval — a stale list rather than a broken one.

A CPU-spin fix for the branch watcher

The watcher that keeps the prompt footer's git branch current used to retry in a tight loop when it couldn't attach — most commonly after exhausting the inotify watch limit on Linux — pegging a CPU core. It now backs off exponentially on failed attaches, while branch switches still update the footer immediately when the watcher is working.

What to check

If you script Cursor CLI setup with --trust for CI or onboarding, the interactive case is now covered too — you no longer need --print/headless mode purely to get past the trust dialog non-interactively. See the Cursor CLI sheet for the current flag and config surface.