Feedback round 1: readability, projects in the URL, backup locks
Feedback round 1 after the UI redesign: larger type and stronger contrast, projects in the URL, a full-screen code reader in Kaisen, and a fix for backups failing their retention cleanup on a stale repository lock.
Yesterday we shipped the UI redesign. Chas then tested it by hand and sent a list. This is what changed in response, plus a backup fix that went out with it.
Readability and layout
- Larger type: 14px body text, 15px on phones.
- Stronger contrast, bolder gauges and sparklines, and a bolder divider on the dev banner.
- Danger zones are outlined in red, and table edges are padded.
- One "Legal" link instead of three, and the logo is back beside the name.
- Settings pages use the full width.
- Templates no longer show the port. "Serve sensitive paths" moved under Advanced options, and Health's "Lifetime statistics" moved to the bottom.
Projects in the URL
Apps and Kaisen pages now live under /projects/:id/... and name their project. Before, some pages depended silently on the last project you had clicked. In the left tree, each project now has Apps, then Kaisen (Builds, Settings). Apps built by Kaisen carry a "Kaisen" tag on the Apps page. Old links redirect.
Kaisen
There is a full-screen code reader with a folder tree, a filter, line numbers, wrap and keyboard navigation. Diffs open in it. "Uploads" and "Project files" are now named apart.
Backups
Platform backups were failing their retention cleanup. The cause was a restic repository lock, typically left behind by a killed job. Three changes:
- A failed cleanup now queues a clean-up run that clears stale locks.
- The backup image runs
restic unlock(stale locks only) before cleanup. - Failures report one reason from a fixed set: repository locked, access denied, connection failed, repository not found, out of space, or other. Restic's raw text is never shown, so nothing internal can leak.
Per-app backups had the same flaw and got the same fix.
Honest notes
A pentest of the round found low-severity issues, and all were fixed before release. Two CI runs failed on a race in the test harness for the new backup-script tests. The deploy gate held both, so nothing broken reached production. We have not yet confirmed a scheduled backup running on the new image.