Skip to content

Changelog

All notable changes to LeafFuzz are documented in this file.

[Unreleased]

[0.1.1] - 2026-11-11

Added

  • Headless Pass 1 zero-server static batch and single-submission triage CLI (LeafFuzzBatchTriageRunner: java -jar LeafFuzz-0.1.1.jar triage (--input-dir <dir> | --jar <file.jar>) [--max-pass 200] [--threads <n>] [--out-json <f>] [--out-csv <f>] [--out-md <f>] [--fail-on-reject], alias scan) with < 0.5% empirical false-positive rate, composite delivery scoring, Folia/Fabric Jar-in-Jar causal tracing, constant-pool companion dependency inference (Vault, PlaceholderAPI, WorldGuard, LuckPerms), and deterministic CONFIRMED_FOLIA_CLAIM_VIOLATION / CONFIRMED_MALICIOUS_BACKDOOR_SINK rejection gates (REQ-FUZZ-TRIAGE-001..004).
  • In-memory per-slice BIT verification harness (SliceBitHarness, LeafFuzzRuntime#auditSliceBit, FUZZ-CBIT-004) executing isolated PBIT/CBIT/IBIT checks per causal slice via EphemeralSliceClassLoader and InMemorySliceBitSession without mutating .jar files on disk (7/7 PASS offline, 10/10 PASS live).
  • :config-bit-api FuzzDiagnosticSnapshotSource diagnostic telemetry implementation across LeafFuzzRuntime, LeafFuzzScorecardReport, and PaperLeafFuzzPlugin alongside the self-contained LeafConfig Plugin UML workspace tab and per-key configKeyBadges export (LeafFuzzRuntime#buildLeafConfigWorkspaceExtension, registered automatically in PaperLeafFuzzPlugin when LeafConfig is installed).
  • Dedicated :bytecode-api static 18-rule CWE, causal Folia runtime exception (FOLIA_CAUSAL_EXCEPTION), causal force-OP/backdoor and stage-2 classloader dropper (SEC-FORCE-OP-BACKDOOR / CWE-269, SEC-COVERT-CLASSLOADER-DROPPER / CWE-494), and cross-platform parity analyzers (BytecodeCweAnalyzers, BytecodeParityAnalyzers, BytecodeCausalGraphScanner, SchemaInferrer, ServerConfigAtlas) detecting shared-state concurrency hazards (ClassSharedStateHazard), synchronous I/O hops (SyncIoHop), and directed fuzz targets (DirectedFuzzTarget).
  • Reversible Stage 2 synthetic bot world-mutation micro-skills (PersonaPolicyBotActor, LeafFuzzPaperEpisodeRunner, NmsEmbeddedPlayerSession, bench-swarm.js) covering exploration (rtp_or_anchor), mining (harvest_nearest), and building (place_wall_ring) actions with automatic block-state restoration and fault oracle recording.
  • Bundled default config.yml and messages.yml resources in leaffuzz-common so fresh installs unpack complete commented defaults on first boot.

Changed

  • Renamed Stage 5 from "Claim-Delivery Audit" (stage=claims, auditClaimDelivery, stage5ClaimDeliveryVerdict) to "Bytecode & Config Causal Audit" (stage=causal, auditCausalDelivery, stage5CausalAuditVerdict, BytecodeCausalGraphScanner.CausalFinding) across /fuzz check stage=causal, /fuzz report (LeafFuzzMenuBuilder), and exported scorecards (LeafFuzzScorecardReport), while retaining stage=claims as a backward-compatible alias in LeafFuzzCommand.
  • Enforced the check.requireDisposableForAll (default true) safety gate in LeafFuzzRuntime and LeafFuzzCommand across all platforms (Paper, Spigot, Folia, Fabric, NeoForge), blocking target=all / * whole-server fuzz runs unless an explicit .bench-disposable opt-in marker file is present in the LeafFuzz data directory or server root, and expanded the default synthetic dispatch denylist to block save-all, save-off, timings, and spark alongside check.extraDenylist patterns.
  • Shortened built-in synthetic persona prefixes (LF_Expl, LF_Cmd, LF_Gui, LF_Red, LF_Sort, LF_Hop, LF_perm, LF_Idle) and centralized indexed bot name formatting in PersonaPolicyBotActor#actorName(int) so multi-bot swarm names (_1.._500) never exceed Minecraft's 16-character username limit.
  • Aligned leaffuzz-fabric (fabric.mod.json), leaffuzz-neoforge (neoforge.mods.toml), and storefront documentation to the MC 26.1+ official mod-loader baseline and compiled leaffuzz-paper to the Java 17 bytecode baseline (classfile 61, ADR-113).

Fixed

  • Moved /fuzz reset (including .jar.sleuth-disabled campaign archive restoration) and /fuzz reload (config.yml, messages.yml, personas.jsonl, and JAR ownership re-indexing) onto accessor.scheduler().runAsync(...) so disk I/O never blocks the server tick thread (S-005).
  • Fixed LeafFuzzConfig#validateDirectory rejecting partial config.yml files that omit defaultPlayers during /fuzz reload instead of falling back to the default 1 (matching LeafFuzzConfig#parse).
  • Isolated io.papermc.paper.command.brigadier.CommandSourceStack lookups inside PaperCommandTreeInspector so Stage 1 command-tree inspection degrades cleanly on plain Spigot servers without throwing NoClassDefFoundError.
  • Added a badParameterInFlight guard in LeafFuzzCommand so invalid name=value parameters reject the command cleanly instead of falling through to default execution.
  • Synchronized NodeBandit UCB1 arm updates and PersonaPolicyBotActor state updates (volatile counters and synchronized recordOutcome) for concurrent multi-player swarm runs.

[0.1.0] - 2026-10-30

Added

  • Standalone multi-loader community QC, boundary fuzzer, post-GC memory-leak screener, and bytecode causal auditor deliverable (LeafFuzz-0.1.0.jar) supporting Paper, Spigot, Folia, Fabric, and NeoForge from a single archive (ADR-125, ADR-126, leaffuzz-ADR-001).
  • Stage 1 Command & Least-Privilege Surface Auditor (CommandSurfaceAuditor, BrigadierDispatcherInspector, PaperCommandTreeInspector) inspecting live Brigadier and Bukkit command trees across default and op tiers, augmenting bare command roots with white-box bytecode subcommand and inline permission schemas (BytecodeCausalGraphScanner.BytecodeCommandSchema), folding namespace:cmd aliases, enforcing DEFAULT_DENYLIST (stop, restart, op, deop, ban, kick, reload), and generating ready-to-paste commands.yml and LuckPerms mitigations (REQ-SLEUTH-FUZZ-002).
  • Stage 2 In-JVM Synthetic Persona Bot, GUI Menu & Swarm Boundary Fuzzer (PersonaPolicyBotActor, SleuthServerFuzzer, LeafFuzzPaperEpisodeRunner, NmsEmbeddedPlayerSession, NodeBandit, FuzzActionShrinker) evaluating the deterministic 256-byte Int8[16x16] sleuth_persona_v1 policy genome (selectPolicyActionInt with Q10 EXP2_FRAC_LUT, 16 POLICY_MICRO_SKILLS, and 8 built-in archetypes plus optional <rootFolder>/personas.jsonl profiles) to drive player-only commands, salient GUI menu clicking (hazard:gui-item-leak icon extraction auditing), reversible world block micro-skills, concurrent player rosters dispersed across an Archimedean spiral (players=auto / players=1..500), and follow-up behaviors (quit_lt1s, quit_delay, move, chunk_border_thrash, window_click, repeat, concurrent) with Zeller's ddmin + binary numeric boundary bisection (REQ-SLEUTH-FUZZ-004).
  • Stage 3 Main-Thread Stall, Synchronous Hazard & Exception Oracle (LogAndStallSignatureOracle, JarClassOwnershipIndex) attributing uncaught exceptions (exc:), slow-tick stalls (stall:), sub-100ms synchronous main-thread hazards (hazard:sync-chunk:, hazard:sync-io:, hazard:blocking-wait:), and grey-box thread coverage (cov:) outside GC pauses to the deepest plugin or mod .jar stack frame (REQ-SLEUTH-FUZZ-001, REQ-SLEUTH-FUZZ-003).
  • Stage 4 Tick-Paced Post-GC Class Retention Screener (ClassRetentionScreener) capturing DiagnosticCommandMBean#gcClassHistogram snapshots off-thread (S-005) with an idle-control baseline (qc-idle-control) and candidate-scoped Bonferroni OLS regression across player departures and command executions.
  • Stage 5 Zero-Classloading Bytecode & Config Causal Auditor (BytecodeCausalGraphScanner, LlmCausalChainAuditor) verifying first-order (0-hop registration and YAML wiring, including REGISTERED_CONFIG_GATED vs UNREGISTERED listeners) and second-order ($k$-hop causal chains to terminal effect sinks, $k \le 4$) invariants (<DeclaredSurface> -> <CausalDefect>), corroborated with live Stage 2 telemetry and optional AST-grounded Jminor LLM fault explanations (corpus-atlas-ADR-001, corpus-atlas-ADR-002).
  • Dual-purpose standalone external TCP swarm CLI (LeafFuzzStandaloneCli, LeafFuzzTcpSwarmClient) inside LeafFuzz.jar (java -jar LeafFuzz.jar swarm --players auto and java -jar LeafFuzz.jar extract-mineflayer).
  • Canonical CommandsAPI command tree (/fuzz, /leaffuzz, /qc) with name=value parameter grammar (commands-api-ADR-001), multi-target selector support (target=PluginA,PluginB or target=all), stage selector (stage=all|commands|fuzz|faults|leaks|causal|snapshot), /fuzz reset [scope=all|faults|leaks|campaign], interactive WCAG 2.1 parchment Book scorecard (/fuzz report via LeafFuzzMenuBuilder and BookContrastValidator), status-leading Markdown/YAML/JSON report export (reports/<target>-qc-report.md, reports/<target>-qc-data.yml, reports/<target>-qc-data.json), and [FUZZ bit] PBIT/CBIT/IBIT/LIVE self-tests (/fuzz test: FUZZ-PBIT-001..002, FUZZ-CBIT-001..003, FUZZ-IBIT-001, FUZZ-LIVE-001..003).