0.9.9 (Extended Lite core + per-server latency)
• New third core: sing-box (extended lite).
A light build of the extended core for routers with little free flash -
~10 MB (UPX) or ~15 MB (pure ELF) instead of ~105 MB. All the features
NetShift gates on are kept: VLESS + Reality, VLESS Encryption, XHTTP,
VMess, Hysteria2/TUIC, Clash API, TPROXY, FakeIP. Install it from the
Components manager; the right variant (ELF or compressed) is picked
automatically based on free space. Cores installed manually with community
scripts are now detected correctly and no longer overwritten by the full
extended package by accident.
• Dashboard: per-server latency and skeletons.
The test button now measures every server in a section separately and shows
a skeleton on each card while measuring, so you can see the exact latency
of each node instead of one number for the whole group.
• Subscription groups on the dashboard got per-feed blocks with their own
"Fastest" (from 0.9.8 series), duplicate-link dedup, and cache cleanup for
removed feed URLs.
• VLESS Encryption (post-quantum ML-KEM keys) from links and Xray-JSON
subscriptions is now passed through to sing-box - with the value checked
against what the core itself accepts, so a broken key skips that one link
instead of failing the whole config.
• Fixed: VMess and XHTTP links on the stock core could break the whole
config (a "dependency not found" startup failure or an outbound that can
never connect). Such links are now skipped cleanly with a log message.
• Cache self-healing: a corrupted subscription cache is re-downloaded
automatically instead of leaving you without proxies.
Restart NetShift after updating (/etc/init.d/netshift restart).
If free flash is tight, try the new lite core - Components → sing-box
(extended lite) → Switch to lite.