Getting started is clearer. Choose Build, Review PRs or Verify, and follow the next step for your project. Starting a new project and connecting an existing repository are easier to find. Decision cards now put Treft's independent test result first, with agent summaries clearly labeled and supporting evidence close at hand. Setup errors offer a retry while keeping your existing cards available. We also improved layouts for larger text sizes.

You can now steer a run while it is working. Stop it, tell it what to do differently, and it carries on with everything it had already done. Dependency bumps and other bot pull requests fold into one row, so the work you started stays in front of you. The decision card now says the test result once, in plain words, and keeps the detail one tap away.

• History: review the runs you've already made and every decision you've settled — approved, rejected, shipped or stopped. • A quieter decision card. Each fact is stated once, so there's less to read before you decide. • The live activity stays honest: a run isn't done until its card is ready to judge, and it says so. • Signing out now asks first. • Craft throughout: every control answers a press, the diff clears the home indicator, and sign-in scrolls when the keyboard is up.

Treft re-runs the tests itself, so you never have to take an agent's word for it. - Every pull request is re-checked in a clean room: a fresh checkout, the exact patch, the network cut. - The card shows graded evidence rather than a claim - what was checked, what passed, and what it does not establish. - Review tiers put your attention on the changes that earned it. - Approve or reject from your phone. Nothing merges without you.

Sharper, quicker, and more careful about what it claims. • Your own runs stay at the top of the queue, newest first. A card no longer sinks the moment it is graded. • Cards open faster. The change is fetched and read once, not once per file. • Evidence shows the pages your change actually named. A page it never touched is no longer shown beside one it did. • Clearer wording on what was checked, what was not, and why — including a plain next step when the tests are too thin to say much. • Shipping something unverified now takes one honest hold instead of two, and it never reads as cleared. • Fixes across the review deck, notifications, the run view and account screens.