Security
- Tracker credentials stay on the tracker's own host. A definition path not
starting with a single "/" is refused, since it could send the API key or
cookie elsewhere; API keys no longer follow redirects to another host; and
responses from trackers and Prowlarr/Jackett/QUI are capped at 8 MiB. - A saved key or cookie no longer follows a URL to another host. Changing
a tracker's URL, or the qui, Prowlarr or Jackett address, to a different
host, port or scheme now clears the saved key, cookie or password unless a
new one is entered with it, and a tracker Test against an unsaved address
uses only what is typed. Path and trailing-slash edits keep them as before.
Changed
- Faster page loads. Responses are now gzip-compressed (the history chart
data shrinks about fivefold), and the pathways pages no longer re-read every
route's requirements on each view. - Build toolchain refreshed. The image builds the frontend on Node 24 LTS
(Node 20 is end-of-life) and runs on Alpine 3.24; vite and puppeteer-core
are updated to releases that clear all opennpm auditadvisories.
Fixed
- Login protection is harder to get round. Every password prompt now
counts towards the five-attempt lockout, parallel guesses can no longer
slip past it, a spoofedX-Forwarded-Forno longer resets it, and removing
the account asks for your 2FA code when 2FA is on. Turning 2FA on signs out
every other session. - Tracker-supplied group colours and icons are validated before they reach
the page, a few more tracker-derived values are escaped, and responses now
carry a basic Content-Security-Policy. - One tracker can no longer crash Yata: a profile page with certain non-ASCII text panicked the scraper, and any panic in a background refresh now fails only that tracker; the log file also recovers after a failed rotation, and the server times out slow request headers.
- The weekly digest could repeat. An out-of-range day or hour (from a
hand-edited or imported config) sent it every five minutes, and a digest
sent while Alerts were being saved, or alongside Send now, could go out
twice. Out-of-range schedules are now refused or reset to Monday 09:00. - A tracker's text could ping a whole Discord server or lose a Telegram
alert. Discord messages no longer allow mentions, so an event named
@everyonestays text; Telegram messages are escaped, so a stray_or*
no longer gets them rejected. - Growth rates measured across UTC midnight were too low (a tracker on its
second day could read 6.25 GiB/day instead of 100), inflating target ETAs
and raising false "behind pace" and seed-size drop alerts. Rates now use
when each value was actually recorded, and are much cheaper to compute. - After signing out and back in, the alerts bell wouldn't open and a
Pathways step or pin button undid itself in the same click. - The "7d" change on the Uploaded, Downloaded and Buffer cards counted a
newly added tracker's whole total as this week's change. - A negative buffer of a tebibyte or more showed in GiB ("-2048.00 GiB"
rather than "-2.00 TiB"). - Searching the log for "amp", "lt" or "quot" mangled lines containing
&,<or". - The dashboard came up blank when the browser blocks site storage.
- A manual refresh during a background one could be overwritten by the
older reply, and a slow qui piled up overlapping requests. - One promotion could alert twice, and a tracker that was down was
re-contacted by every open tab's poll: refreshes of the same tracker now
wait for each other, and a failed fetch counts towards the refresh interval
like a successful one (the refresh button still always fetches). - A target added already met could fire "target met" after all of a
tracker's targets had been removed and one re-added. - Pinned paths slowed every refresh cycle with many trackers.
/api/summarynever reportedwarnings.- Importing an older config could leave alerts with nowhere to record and
a renamed tracker type not collecting until the next restart.