Post-remediation polish: the database Web Studio is now served through the panel itself (fixing the broken embedded iframe under HTTPS), the update banner is honest about WHY it could not check for updates and recovers from transient failures in minutes instead of hours, and the installer verifies that what it downloaded is actually the release it asked for.
Added
- The database Web Studio is served through the panel origin. A
same-origin reverse proxy (
/v1/databases/:id/studio-proxy/) fronts the loopback-bound Adminer / Redis Commander container: the embedded iframe finally works under HTTPS and CSP, the studio port stays off the network entirely, and access rides an HMAC-signed, path-scoped, 8-hour operator-only cookie instead of nothing at all. Upstream cookie paths are rewritten into the proxy scope so parallel studios never share sessions. - The About page tells you WHY an update check failed (checks switched off vs feed unreachable, with the feed's own status line) and offers a Check again button instead of waiting out the cache.
- Installer provenance checks: the release tarball's content must match the requested tag (version binding, no symlinks, no setuid bits) and the docker-mode compose file must parse with the compose engine, reference exactly one image — the panel's own — before it is deployed.
Fixed
- Update-check failures are cached 10 minutes, not 6 hours. A single boot-time blip (network not up yet, one API rate-limit 403) used to pin "unavailable" on the dashboard for six hours.
- The failed-check result now carries a machine-readable reason, so future UIs can distinguish disabled from unreachable.