Releases: PotenFYR-Studios/AuthCore
Release list
AuthCore - Latest Development Build
Development build of AuthCore 1.0.0
Commit:a13c848- built 2026-09-12 17:22 UTC
One merged changelog for the whole 1.0.0 line - every feature, fix and hardening
pass that ever shipped under a1.0.0*label lives in this single section
(the separatealpha.1-alpha.5entries were folded in; nothing was lost).
Build, mixin & proxy-gate hardening (2026-09-06)
Build-time remap errors fixed (all 4 Cannot remap warnings gone)
startSleepinglobby restriction: theEither-returning overload never existed on any
supported version; unified to the void overload that exists on every version.- Elytra block:
startFallFlyingwas removed from the mappings long ago (the injection
silently never ran); it now injectsupdateFallFlying, the per-tick glide driver, which
also extends the elytra + jump lobby restrictions to the 1.16-1.18 range. - Mount block:
startRiding(Entity,Z)was replaced by a 3-arg overload in 1.21.9; the mixin
now injects the 1-arg final entrypoint (present on every version) plus the era-specific
force overload, so mount blocking actually runs on 1.19-1.21 instead of silently no-oping. - Deop tracking: the
GameProfile->NameAndIdAPI swap happened in 1.21.9, not 26.x;
the stonecutter cut point was corrected and the handler now uses@Coerceso the same
jar's bytecode never references the 1.21.9-onlyNameAndIdtype.
Runtime mixin crashes fixed (host matrix back to green)
ServerPlayNetworkHandlerMixinlegacy handlers andPlayerListOpMixinusedObject
handler parameters, which Mixin rejects at runtime (InvalidInjectionException) - the
1.16-1.18 fabric legs and all 26.x legs failed as a result. Handlers now declare the
exact packet types.
Startup banner fixed on log4j-era loaders
- On runtimes without slf4j (1.16-1.18 Fabric/Forge) the entire banner printed literal
{}placeholders instead of values. The fallback console logger now substitutes
slf4j-style placeholders, so version, Minecraft version, database type and every
security flag display correctly.
Proxy-gate bypass & spoofing hardening
- Velocity/BungeeCord proxy gate: a gate error previously failed OPEN (unauthenticated
players allowed through even withblock-unauthenticated=true); it now fails CLOSED
by default - deny unless the operator explicitly setsfail-closed=false. - Interop messages (
AUTH_CHANGED) are no longer accepted from player connections -
only backend-server senders are trusted, and messages are consumed so they can never
be forwarded to clients.
CI: rolling latest-build release
- Every
mainpush that passes build + host tests now republishes thelatest
GitHub release: same-version uploads REPLACE the jars and regenerate the changelog
notes; the stablev*tag release remains the "Latest" release.
Test suite grown to 180+ checks
- New suites: proxy spoof-guard hardening, trusted-proxy source + CIDR validation,
proxy-config parsing strictness, interop message parsing, proxy-side session cache. - Docker harness now verifies server-mode auto-detection and banner data correctness
on every leg.
Detection hardening & bypass resistance (2026-09-05)
7-layer defense-in-depth stack
- Layer 1 - Session Binding: per-server random 32-byte companion attestation key,
generated on first boot, persisted toconfig/authcore/attestation.key, rotated on
/authcore reload(invalidates all pending challenges). No hardcoded keys anywhere. - Layer 2 - Packet Sequence Validation: login packet state machine (HELLO → SETTINGS →
READY) tracks every connection; anomalous sequences are logged asPACKET_SEQUENCE_ANOMALY.
Stale entries are pruned on tick and on player leave. - Layer 3 - Behavioral Profiling: ClientGuard risk scoring now includes confusable-name
detection (O(1) normalized-name index), concurrent-connection fingerprinting
(CONCURRENT_FARM: ≥3 distinct usernames from same IP within 5s = +25 risk),
look-pattern bot detection (camera rotation delta variance = +20), and observation-window
entropy analysis (low timing variance = +15). - Layer 4 - Look-Pattern Analysis: tracks per-player look deltas (pitch/yaw changes
between packets); computes coefficient of variation - bots have zero or perfectly regular
patterns, humans have natural variation. - Layer 5 - Login Timing Distribution: IP-level login timestamp analysis (60s window,
≥3 samples); low coefficient of variation (< 0.15) flags bot-farm synchronized timers. - Layer 6 - Concurrent Connection Fingerprinting:
ClientGuard.checkConcurrentFarm
tracks distinct usernames per IP within a 5-second window and attachesCONCURRENT_FARM
(+25) to the joining player's profile. - Layer 7 - Login Intelligence: device fingerprint (SHA-256 of IP + country), new-IP /
new-country alerts (+15 risk), 2FA code rate limiting viaRateLimiter.tryMfa(5
attempts/minute per IP + account).
Fail-closed security defaults
- Proxy support is hard-disabled when
trusted-proxiesis empty (was a warning, now a
startup block viaSecurityConfiguration.enforce). - Redis-backed proxy auth gate defaults to fail-closed (
fail-closed = truein
ProxyConfig) - Redis unreachable blocks proxy-authenticated connections instead of
allowing them. - Session token TTL enforcement: tokens expire after 24 hours regardless of session timeout,
preventing indefinite session fixation.
Performance & robustness
- Batch GeoIP lookups: IPs are collected over a 500ms window and resolved in parallel,
bounded by a 4-permit semaphore - eliminates thundering herd under join floods. - Detection state integrity:
PACKET_SEQUENCESmap is pruned on tick (expired + disconnected
connections); all detection maps have cardinality bounds and self-pruning. - Web panel CORS support: OPTIONS preflight handler +
Access-Control-Allow-Originheaders
on every response; CORS headers also set on error paths.
Test coverage
- Security test suite expanded with 12 new test cases: attestation key generation/rotation,
concurrent-farm detection, look-pattern inner classes, login timing null-safety, packet
sequence tracking null-safety, and 2FA rate limiter integration.
Test-infra overhaul & harness fixes (2026-08-28)
- Temurin-based Docker verification: the host-test harness runs on the official
eclipse-temurin JRE images (17/21/25,
one per Minecraft version group) and boots the whole matrix IN PARALLEL - one
container per MC/loader combination. Newtest/docker/(Dockerfile + self-
provisioning entrypoint + parallel runner),test/build.sh(builds all 7 variants
with no local JDK needed) andtest/run-security-tests.sh(bash port of the
standalone security suite). Oldtools/host-tests,tools/security-tests,
scripts/and their workflows removed.
test/build.sh(builds all 7 variants inside the official eclipse-temurin JDK image with
no local JDK needed) andtest/run-security-tests.sh(bash port of the
standalone security suite). Oldtools/host-tests,tools/security-tests,
scripts/and their workflows removed. - Minecraft version detection fixed on Forge/NeoForge:
Compat.getGameVersion()
relied onSharedConstants.getCurrentVersion().getName()via reflection - method
names are SRG-renamed on Forge runtimes, so the startup banner printed
Minecraft : unknown(and falsely warned "not in the officially tested set").
Now resolves through FML's non-obfuscatedVersionInfo(Forge static accessor;
NeoForge'sFMLLoader.getCurrentOrNull().getVersionInfo()), with the Mojang
reflection kept as last resort. - Versioned documentation site: a common main page (
index.html: downloads,
quick start, FAQ) plus per-release doc sets underdocs/<version>/, cross-linked
through a shared version bar (docs/assets/nav.js+docs/versions.json) and
deployed by the recreated.github/workflows/pages.yml- the existing
authcore.potenfyr.in/docs/<version>/...URLs are preserved. - Parallel Docker host tests, never stuck: the whole (version × loader) smoke/
verify matrix boots in parallel containers; server state lives in named Docker
volumes (Windows 9p bind mounts stalled the Forge/NeoForge installers), the
installer runs under a hard timeout and every test has an outer kill deadline. - Premium-detection hardening on 1.18.2:
Compat.serverUsesAuthenticationcould
miss the mojmap method name and default to "online", routing cracked players through
the wrong auth flow. Now triesisOnlineMode/getOnlineMode/usesAuthentication
plus anonlineModefield fallback before defaulting. - IDEA: broken fabric-only run configurations replaced with Gradle-based
Build all variantsconfigs driving the newbuildAll/testAll/dockerTest
aggregate tasks (all loaders × all ranges in one pass).
Security-audit & CI pass (2026-08-26)
Critical fixes
- MySQL / PostgreSQL servers no longer brick at boot: the hard-coded SQLite DDL
(AUTOINCREMENT,TEXTprimary key) threw on both dialects, which set
migrationBlocked = trueand suspended login/register for the whole server. Table
creation is now dialect-aware (AUTO_INCREMENT / BIGSERIAL / VARCHAR(36) keys). - Session-resume premature kick fixed: resuming a session re-ran
login()without
stopping the previous session's timeout task - the stale timer fired at the ORIGINAL
deadline and kicked freshly-authenticated players. The old timer is now cancelled on
every (re)login. - Direct-client IP spoofing closed: proxy IP forwarding accepted a BARE handshake
address as a forwarded IP - but that field is client-controlled, so any modified client
could claim an arbitrary IP (defeating rate limits, GeoIP, login intelligence and IP
rules). Only the real NUL-separated proxy forwarding payload is trusted now. - ...
AuthCore v1.0.0
Hybrid mode, per-account login style & release hardening (latest)
Anti-float platform - mid-air logouts can never kick players in the limbo
- A player who logs out mid-air (or underwater) used to be kicked by vanilla's
"Flying is not enabled" floating check while standing in the limbo. AuthCore now
places an invisible BARRIER platform under the limbo player (lobby.anti-float-platform,
on by default); diving players are stood on a platform at the water/lava surface instead
of being left submerged. The original block is restored shortly after the authentication
flow completes (lobby.anti-float-platform-delay-ms, default 10 seconds) - never
overwriting a block another player placed in the meantime. - Crash recovery for the limbo (already present, verified): the pre-limbo snapshot
(exact position + dimension, inventory, effects, game mode, health/food/xp) is persisted
at lock time; if the server (or the player's session) crashes mid-limbo, the snapshot is
restored on the next join BEFORE the fresh lock, so after login the player always returns
to the exact spot they were at their last disconnect - regardless of the admin-configured
limbo location. - Movement lock hardened: with movement disabled, the per-tick limbo re-assert now
snaps the player back at a 0.25-block drift (was 1.0) - on runtimes without the
movement-cancel mixin (Fabric/Forge 1.16-1.21 without a mixin refmap) this is the only
server-side lock, and it now holds the player essentially in place. Known limitation:
on those runtimes the client can briefly ghost-walk up to the snap interval; the
restriction is fully packet-level on NeoForge and every 26.x jar.
Server mode is always taken from server.properties - the config override is gone
- The
session.server-modesetting is removed entirely: the mod always reads the real
online-modefrom the running Minecraft server (MinecraftServer#usesAuthentication).
Premium auto-login works the same on online AND offline-mode servers (on offline servers
AuthCore re-runs the vanilla encryption handshake and verifies with the server's own
Mojang session service - no external API calls; failures fall back to offline
register/login, players are never blocked or kicked). - Online-mode servers get a startup warning to keep
enable-secure-profile=falsein
server.properties, so clients without a secure chat profile (cracked/modded players)
can still join and chat.
Hybrid servers: offline-mode players can join BOTH server modes
- New
allow-offline-playersconfig (session.authentication, defaulttrue): with it on,
offline (cracked) players can join and register/login on online-mode servers too -
the login mixin intercepts offline-UUID clients on online-mode servers and runs them
through vanilla's own offline accept flow (no Mojang session check, offline UUID kept;
online-mode players still use their real UUID and the normal session verification).
Fail-safe: on versions where the offline accept flow cannot be driven, vanilla's normal
rejection takes over - nobody is stranded. - With
allow-offline-players = falsethe server is online-mode-only everywhere: offline
players are kicked with a clear message on arrival (join gate, before session resume /
register / login), and on online-mode servers they are disconnected at login.
Premium auto-login respects its config on BOTH modes; auto-login players keep a null password
- The
premium-auto-loginconfig (on by default) is now honored regardless of the server's
mode. Auto-login players are NEVER given an auto-generated password - their stored
password stays null (the oldpremium-auto-registerrandom-password path is removed). - If auto-login is turned off (or a player switches to password login), verified online-mode
players are treated like any standard account: because their password is null they are
asked to/registeron their next re-auth. Their online-mode status is preserved, so
auto-login simply resumes when the config (or their mode choice) is switched back.
Per-account login style: players and admins can switch online/offline mode
- New player command
/account set-mode online|offline(permissionauthcore.user.setmode,
level 0): switches the player's own account between automatic login and password login. - Switching to password login is smart about passwords: a player who ALREADY has a stored
password keeps it and simply logs in with it (/login); only auto-login accounts with a
null password are asked to/registera new one. The active session is destroyed either
way so the change takes effect on the next join. /authcore set-mode offline <player>no longer takes a password argument and follows the
same rule (existing password kept / register when null). Both admin mode commands destroy
the session too.- A player's own mode choice is honored on join: accounts explicitly set to password login
are never auto-logged-in, even on an online-mode server where the session was verified.
Terminology: "premium"/"cracked" replaced project-wide
- Non-user-facing text (console logs, debug output, admin commands, web panel labels, code
comments, docs) now uses online-mode players and offline-mode players instead of
"premium players" and "cracked players". Internal identifiers and thepremium-auto-login
config key stay unchanged for API/config compatibility. - Player-facing messages stay human: auto-login greets with "Welcome to the Server!", mode
switches talk about "automatic login" / "password login" - no technical jargon anywhere
in chat, titles or kick screens.
Limbo guard debug output
- The per-join limbo guard report is now debug-level and formatted like the startup banner:
one aligned row per guard (movement / chat / commands / block-break / block-use / item-use
/ item-drop / attacks, eachallowedorLOCKED) plus a live lobby-usage line
("35% used (7/20)", "unlimited" when no cap).
Scale, race-condition safety & docs
- Performance/scalability pass documented for 500k+ registered accounts and thousands of
concurrent players: O(1) user lookups on every hot path, lazy DB loading, bounded
self-cleaning caches, no resource spikes under join/login bursts (per-user throttles,
rate limits, fixed-size daemon IO pool), race-condition-free concurrency (canonical
single-User-per-account cache,synchronizedDB access,volatileshared state,
deduped join/leave hooks, atomic counters). - Docs re-skin: the hosted docs now use a black + red fortress-cyber theme with an enhanced
sidebar table of contents (topic count, in-TOC scroll progress, back-to-top, glowing
active states) and a wider, spread-out content layout; README and every docs page updated
for the current behavior.
Limbo, performance & configuration overhaul
Limbo quality pass (no more screen vibration, no bypasses)
- Anti-vibration movement correction: the client was snapped back on EVERY violating
movement packet (up to 40 position packets/s → rubber-banding). Corrections are now
distance- and throttle-based:lobby.movement-correction-radius(default 1.5 blocks) and
lobby.movement-correction-interval-ms(default 600ms), the classic AuthMe feel
(ghost-walk a little, one clean snap). Movement packets are still cancelled on every
packet, so the server entity never leaves the anchor (no bypass). - Vehicle-movement bypass closed:
handleMoveVehiclewas uncovered: lobby players on
boats/minecarts (or spoofing the packet) could move freely. Now cancelled + anchored like
player movement. - Inventory lock without touching chat: the inventory is fully inert in the limbo
(every slot click blocked and force-closed on interaction, including shift-clicks and
armor equipping), while the chat input is NEVER interrupted, so/registerand/login
always work. A periodic force-close packet was tried and removed again: the client closes
ANY screen (including chat) on a container-close packet, and there is no server-side
signal for "inventory open"; click-based blocking is the only safe approach. - Attack-callback fix (the "can't hit mobs" bug): the fabric
AttackEntityCallback
listener was registered under the wrong method name (attackinstead ofinteract),
so the reflective proxy returned null for every attack and the fabric event cancelled
ALL attacks for everyone, in and out of the lobby. One-line fix. - Server-side auth menu removed (chest menu + book input +
/menucommand): auth is
purely chat-driven with clickable buttons; the menu system, its mixins and its command
are deleted entirely. - Context-aware chat buttons: the login/register buttons build the EXACT command shape
the player needs (password confirmation, 2FA code, captcha code) and show it in the
action bar; no confusion about which auth factors apply. - Styling: clickable chat buttons are now underlined; the action bar gets the same
drop-shadow as titles/subtitles. - Crash-safe limbo verified: the pre-limbo snapshot is saved at lock, restored before
the fresh lock on rejoin after a crash, and deleted on clean unlock; the unlock lifts
restrictions before any restore step so a failed restore can never keep a player stuck.
Performance pass (constant per-packet cost, bounded memory)
- O(1) user lookups on every hot path: new
User.getUser(UUID)/User.getUser(player)
. a single map get, no string allocations, no scans, no DB, and all per-packet mixin guards
(movement, clicks, chat, ticks, entity events, commands) now use it. - Indexed username lookups: precomputed lowercase names + a
byLowerNameindex make
lookUpByUsernamemode O(1) too (previously a full-map scan with per-entry
toLowerCaseallocations). - Throttled cache touches: the last-access map put now happens at most once per minute
per user instead of on every pac...