Editor-only class swap via the existing scalar dev-RPC family (new DebugOp.SetClass): F1/F2 keybind (ClassSwitchHotkeySystem), DebugOverlay '- Class -' buttons, and DebugCommandSendSystem.SetWarrior/SetRanger/SetClass statics. Server (DebugCommandReceiveSystem) swaps class in place on the spawned player: strips+re-seeds the ClassTraits StatModifier seeds, swaps the AbilityRef Fire slot, resets the ability cooldown, and heals a LIVING player to the new max (dead players skip the heal so respawn isn't raced). Server-authoritative + prediction-correct (same buffer-mutation path as GrantUpgrade); wire type unchanged so the RpcCollection hash is unaffected.
ClassTraits gains a shared Seeds core (spawn + swap can't drift), ClassSeedCount, IsClassSeed, a DynamicBuffer AppendSeeds overload, and Reapply. +3 EditMode tests (exact-count round-trip, value-equality fold, boundary/foreign-mod preservation); 351/351 green; Warrior<->Ranger round-trip Play-validated (server+client agree).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Realigns the skill with how sessions actually run now — it predated the
Workflow tool and the ultracode two-review practice. Driven by an analysis
of the full session history + the 26-file memory corpus + skill-authoring
research, adversarially reviewed.
Key changes:
- Workflow-first orchestration: drop the dead manual "swarm (<=N agents)"
model and the impossible 3-5-agent impl swarm; implementation is serial
orchestrator MCP edits. New references/workflow-patterns.md (ground
fan-out + design-review lens/critic) replaces agent-briefs.md.
- Pre-code design-review + post-impl diff-review are now first-class gated
phases (the spine that catches what green tests + a clean Play miss).
- Size by blast-radius / netcode-heaviness, not time estimates.
- Lean SKILL.md (222 -> 131 lines, -25% KB): leans on CLAUDE.md instead of
duplicating its MCP cheat-sheet / anti-patterns / error-recovery.
- ctx7 CLI / find-docs mechanism (the MCP verbs are gone); live-verified
6.5-era library pins; kill the dead NetCodeTestWorld test path.
- Encode operator gates (no-time-estimates, present-forks, never-defer,
tuning-autonomy) + the highest-recurrence MCP-edit / Workflow gotchas.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Slice 2 complete: Warrior/Ranger, DRG-asymmetric, aim-directed cone, menu picker,
class carrier via GoInGameRequest. Two VFX-polish items deferred (review-sanctioned).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
MainMenuController gains a 2-class picker that sets WorldLauncher.SelectedClass;
WorldLauncher seeds a ClassSelection singleton into the client world at session start,
which GoInGameClientSystem carries on the spawn RPC. Default Warrior. Completes the
Slice 2 loop: pick a class in the menu -> spawn with its kit. Editor-default boot stays
Warrior (the menu path drives the choice). 348/348.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The per-player class travels on GoInGameRequest.ClassId (client reads a ClassSelection
singleton); GoInGameServerSystem seeds the class at spawn via ClassTraits (AbilityRef +
permanent trait StatModifiers on a reserved ClassSourceId; CharacterStatsRef stays Default
so the DRG-asymmetry deltas ride the replicated OwnerSendType.All buffer). AbilityFireSystem
gains the aim-directed Cone archetype: cooldown predicted both worlds, server-only cone
damage to living enemies (same-tick, SourceTick-stamped, like the melee cleave). 345/345.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Foundation for Two Classes (DR-037). New ids (CharacterId.Warrior/Ranger,
AbilityId.WarriorCone, StatTarget.MeleeDamage/MeleeRange); CharacterStatsRef.Id ->
[GhostField] so the owning client folds the right class stats; MeleeComboSystem
folds per-player MeleeDamage/MeleeRange off the replicated StatModifier buffer
(HasBuffer-guarded -> identity without class seeds, so behavior-preserving).
345/345 EditMode. Slice 2 design review + locked forks logged in the session note.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
DR-038 records Slice 1 (combat readability + HUD declutter) with the two reusable
netcode patterns: bake-client-safe for client-needed/server-owned/never-changes
data (EnemyTelegraph), and a [GhostEnabledBit] derived once/tick (IsLunging). Open
item: the operator visual fun-gate. Session log captures the design redirect +
Slice 1 build/validation.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Four playtest do-now wins:
- Enemy health bars: pooled world-space Canvas, on-damage-sticky + fade,
always-on <25% HP (CombatFeedbackSystem; no new replication).
- Telegraph fix: new baked client-safe EnemyTelegraph sizes the danger-cone ramp
per enemy (0->1 ending at impact, fixes the Charger plateau); windup 18->22;
a windup scale-pulse.
- Build-mode toggle: BuildPaletteState.PaletteOpen hides the palette by default,
Tab / gamepad-Y toggles, with a discovery chip (HudSystem/BuildSendSystem).
- Charger committed-lunge tell: [GhostEnabledBit] IsLunging derived once/tick from
LungeState (the Dead idiom); the danger cone persists through the lunge.
345/345 EditMode (+3 IsLunging derive tests); Play-validated: ghost-hash change
did not break the handshake, bake correct (telegraph on all enemies, IsLunging
baked-disabled on the Charger, replicated to client), no runtime errors.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Playtest of the single-arena slice (Scratch Notes 6152026) found it stale fast.
Answer the post-END-2 Decision Gate as continue/expand: re-scope off the fixed
June-30 demo to a co-op roguelite-ARPG — base = persistent buildable hub (never
resets), the procedural Expedition region = the required combat spine, two classes
(Warrior/Ranger), persistent meta (SaveData v6 later). Mission-as-Sortie model:
fulfils locked pillar #4, preserves the never-a-run-reset pillar; reverses DR-031's
expedition pause. Backed by an 8-agent design+feasibility pass (feasibility GREEN:
server-spawns-ghosts removes seed replication). Slice doc archived/superseded.
See DR-037.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Keep the local screenshot scratch dir and the personal vault "User Sessions"
notes on disk but out of git, so the worktree stays clean.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Uncommitted residue from the SL-4 visual-cohesion pass, swept up while
cleaning the worktree (unrelated to the 6.5 upgrade):
- Mat_StructureOwned_Cyan: _Color nudged to match _BaseColor (0.11,0.22,0.26)
- Gameplay.unity: a prop transform moved (0,1,8) -> (5.1,2.46,20.7)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Validated the 6.5 upgrade: 342/342 EditMode tests pass, no compile/Burst/
source-gen errors, and a clean netcode Play boot (ServerWorld+ClientWorld
connect, NetworkId handshake, in-game, ghost replication synced 23==23,
player owned-ghost spawned) -> the 6.6.0a6 "invalid wrapped network
interface" transport bug does NOT affect 6.5-stable. URP 17.5 render clean
(magenta scan 0/1564). The MaterialLocation.External FBX-import warnings are
benign 6.5 deprecation noise (2737 importers), import-time only.
Updates the CLAUDE.md stack table to 6.5.0 (net -13 bytes, under budget) and
adds the full session log.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Owned structures (Turret/Wall/Fabricator/Pylon) now read cyan-owned via a dedicated Mat_StructureOwned_Cyan (dark desaturated-cyan body + a subtle cyan self-illum below the bloom gate) - replacing the shared PolygonFantasyKingdom atlas + M_Turret so there's no atlas bleed. Tuned dimmer than the Core so the Engine Core stays the single luminance peak. Completes the faction palette: dark ground / cyan owned (Core bright, structures muted) / amber Ore / orange Husks. Edit-mode-instantiated + screenshot-verified. Follow-up (deferred): the research's dark-when-unpowered dynamic (emissive only on active/powered structures) needs a small presentation system reading structure state.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The first pass crushed readability ('can't see anything'): dark ground x low ambient x dim sun x +12 contrast (clipped everything below midtone to black) x vignette x dark fog. Rebalanced toward legibility while keeping the dark-frontier mood: ColorAdjustments contrast 12->4 (the main fix), post-exposure 0.55->0.85, saturation 0->4; sun 0.9->1.5 cool-white; ambient intensity 0.62->1.0 with lifted cool ambient colors; vignette 0.32->0.20; ground material lifted ~18%->~28% value; fog pushed out (start 18->26, end 42->60); Core beacon range 9->13. Ground, player silhouette, Ore ring and shadows now all read; Core stays the bright cyan peak.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Research-grounded transform of the base arena from a bright meadow into a dark 'Aether Siege Outpost' reading as a concentric bullseye (Core -> Ore ring -> dark perimeter). Camera (Game.unity): pitch 45->58, FOV 55->44, dist 13->17, lead 1->0.5, follow 8->6, targetH 1.3 (telephoto so Core + arena read together, near-iso enemy spacing). Lighting: sun dimmed+cooled 1.6->0.9, ambient 1.0->0.62 dark-cool, fog Linear 18->42 dark; WorldAtmosphere base darkened. Post-FX (PostFX_Daylight): ACES kept, bloom gated (thr 1.0->1.2 + clamp 10), exposure/contrast up, saturation +6->0, added ShadowsMidtonesHighlights cool-shadow/warm-highlight split, vignette 0.15->0.32. Ground re-tinted dark teal-grey; ~390 meadow-cheer objects cut. Core staged as hero (crystal 14m->4.6m cyan glow + cyan beacon light). Ore ring pulled into the arena (23.5-27m -> 6-11.5m, count 10->12) and recoloured AMBER (new Mat_OreNode_Amber, emissive, no atlas bleed).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The terminal banner gains PLAY AGAIN + QUIT TO MENU buttons so a finished run has a clear action (no Esc-hunting). PLAY AGAIN restarts a fresh Single run via WorldLauncher.StartSession (the proven menu lifecycle, no save load); QUIT TO MENU reuses TeardownToMenu (autosave + menu); both self-guard on WorldLauncher.Busy. The button row picks (Position) under the Ignore banner root, matching the build-palette idiom. AimReticleSystem (the sole Cursor.visible writer) keeps the cursor visible while RunOutcome != InProgress so the buttons are clickable regardless of aim state. 342/342 EditMode green. Co-op retry-together stays a cut slice-limit (a client's Play Again starts a solo run). See DR-036.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The client derives the climactic final siege from the replicated Charge>=Target and marks it distinctly: 'FINAL SIEGE INCOMING - Ns' during the cap-reached arming window (vs the generic 'INCURSION'), 'HOLD THE ENGINE - FINAL SIEGE' + intense red during the wave, and a last-stand location line. Pure client presentation (no sim/replication change); 342/342 EditMode green. Serves the END-2 fun-gate (the Engine telegraphs the climax + prompts deliberate prep).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Path A spine COMPLETE (14/14): Backlog SL-3 blocker cleared + marked done; Path_to_Fun END-2 done + banner; Milestones END-2 row. CLAUDE.md gains the END-2 gotcha line (replicate the outcome, don't client-derive; SiegeTimeout off during the final), net-zero via EB-1/EB-2/END-1/M7/inventory/build-grid condensations (40,445 then 40,510 w/ history note, under the 40,960 limit). DR-036 + session log capture the design, the operator forks (halt+banner, Target=4, SaveData v5), and the pre-coding + post-impl adversarial reviews.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
At GoalProgress.Charge>=Target a new server-only GoalReachedSystem arms a larger final siege (x live FinalSiegeMultiplier) and flips RunPhase=FinalDefense; CyclePhaseSystem latches a REPLICATED RunOutcome (Victory on clear / Loss on Core breach) and halts the director. RunOutcome is a [GhostField] byte on the global CycleDirector ghost (the client banner observes it); RunPhase stays server-only. ThreatDirector/CoreRestore/CoreDamage halt once decided; SiegeTimeout is off during the final siege. SaveData v5 persists the outcome so a won/lost run loads finished. GoalProgress.Target 10->4. Completes Path A's spine. See DR-036.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
DR-034 records the losable-Core decisions: CoreIntegrity on the global ghost,
the soft-loss edge inside CyclePhaseSystem, the Core-as-fallback-target and
despawn-on-breach forks, the transient OverrunTick (vs END-2's latching
outcome), and SaveData v4. Session log captures the build + validation
(330/330 EditMode; Play-verified server==client drain->regen->replicate).
CLAUDE.md adds the END-1 bullet; EB-1/EB-2/inventory bullets condensed
net-neutral to stay under the size budget.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
CoreSystemsTests (new): a breaching Husk drains + is consumed; idles at 0;
regen fires once per interval in Calm only; no regen mid-Siege; caps at Max.
CyclePhaseSystemTests: the soft-loss overrun edge ends the siege, drains the
ledger, despawns husks, withholds the goal charge, and resolves once.
StorageMathTests: DrainFraction floors per row, drops zeroed rows, clamps.
SavePersistenceTests: CoreCurrent round-trips at v4; a pre-END-1 save with no
CoreCurrent defaults to 0 (-> born full); the v3->v4 version pin updated.
TuningConfig golden pin extended with the 3 Core defaults.
See DR-034.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds CoreIntegrity{[GhostField] Current,Max,OverrunTick} on the GLOBAL
CycleDirector ghost (no new ghost/relevancy). CoreDamageSystem (server,
after EnemyAISystem): a Husk within ~3u of PlotCenter drains + is consumed;
CoreRestoreSystem regenerates only in Calm. The SOFT-loss edge lives inside
CyclePhaseSystem (sole Phase writer): Current<=0 in Siege flips to Calm with
NO goal reward, StorageMath.DrainFraction drains the shared ledger, all Husks
despawn, and OverrunTick is stamped (a transient HUD-flash pulse, not a
latching outcome - the Victory latch is END-2's). EnemyAISystem treats the
Core as a FALLBACK target so an undefended base is overrun instead of idling.
SaveData -> v4 persists CoreCurrent (0 -> born full, the EB-1 HP sentinel);
3 live TuningConfig knobs + a red HUD Core bar. Soft-loss + targeting +
breach-resolution forks operator-locked.
See DR-034.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
DR-033 records the felt-spend design (shared Charge ammo, atomic soft-fail,
ledger-fed Fabricator, no-ordering-edge trade-off, global HUD cue, no SaveData
bump). CLAUDE.md adds the EB-2 ★ bullet net-zero (trimmed bullets archived to
the gotchas archive under a 2026-06-12 heading).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- TurretFireSystem: seed a Charge pool for existing tests; add soft-fail-when-dry,
consume-one-Charge-per-shot, two-turrets-share-a-finite-pool.
- FabricatorProductionSystem: ledger-fed withdraw/deposit, two machines split via the
live in-loop read, and a catch-up affordability-clamp regression pin.
See DR-033.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Mined Ore now has an ongoing sink: a ledger-fed Fabricator converts Ore->Charge
(1 Ore -> 3 Charge / 30t) and turrets spend Charge per shot, soft-failing (no
shot, no cooldown burn) when the shared pool runs dry.
- ResourceId.Charge=4 rides the existing [GhostField] StorageEntry ledger (no new wire).
- TurretFireSystem: single ledger resolve + atomic spend / soft-fail / partial-refund.
- Fabricator.InputFromLedger (byte, server-only) feeds input from the shared ledger,
read live in-loop so two machines split a finite pool; both modes deposit to ledger.
- HudSystem: violet Charge chip + global quiet-turret cue when siege && Charge==0.
- StorageMath.TotalOf backs the affordability read; catalog re-enables the Fabricator (4 entries).
See DR-033.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
10 new EditMode tests (312 total, all green): HealthApplyDamage destroys a Destructible at 0 + a wounded structure survives clamped; PickWeightedNearest x5 (player-only, structure-preferred-by-weight, player-in-the-way wins, raze undefended base, no targets); persistence (StructureSave.HP round-trip + writes v3, v2 in the load floor, SaveApply.ToPending maps the wounded HP - the staging-copy bug the pre-code review caught); + the StructureAggroWeight default pin. See DR-032.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Structures (Turret/Wall/Pylon) reuse the combat spine: authoring bakes Health(GhostField)+DamageEvent buffer+a Destructible tag (no HitRadius -> no friendly projectile fire; no EffectiveCharacterStats -> clamp-to-0). HealthApplyDamageSystem destroys a Destructible at 0 (occupancy auto-frees). EnemyAISystem fortress-targets the weighted-nearest of players+structures via the shared EnemyAIMath.PickWeightedNearest (StructureAggroWeight TuningConfig knob, <1 prefers structures, squared factor; snapshot above the early-return so an undefended base is razed). Persistence v3: per-structure HP threaded through 5 sites (SaveData/PendingStructure/scan-guarded/BaseRestore same-ECB born-correct/WorldLauncher via SaveApply.ToPending); SaveService floor-gate [2,3] loads old saves. Loss feedback: proximity-gated StructureFeedbackSystem; CombatFeedbackSystem suppressed for structures. Pre-code review caught the DamageEvent-buffer crash blocker + 8 majors; post-code review clean. See DR-032.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Session log + DR-031 (base-local mining, any-attack harvest, scheduled base sieges, Synty asset swap) capturing the diagnosis, locked operator forks, both adversarial reviews, and the tuning knobs. CLAUDE.md: base-local loop is now the model (BaseFieldSpawnSystem + harvest region-routing + ThreatDirector Schedule source); net-neutral condensation of M7/biome/HUD reference bullets.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replace every placeholder-cube ghost prefab with nested Synty (PolygonFantasyKingdom) models: ResourceNode->crystal, Turret->ballista, Wall->spike palisade, Pylon->crystal beacon, Storage->chest, UpgradePickup->gem, BlightClutter->rock chunk (nest as a Model child, strip colliders, reset the [GhostField] root Scale, atlas for bare FBXs). Gameplay subscene: place the BaseFieldSpawner, trim Harvester/Fabricator/Conveyor from the build palette. CycleDirector prefab bakes the schedule siege config. See DR-031.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
8 new EditMode tests (302 total, all green): BaseFieldSpawnSystem (target count + Base/Ore + cadence + top-up), ExpeditionFieldSystem teardown preserves the base field, ThreatDirector schedule arming, and melee harvest routing (base->ledger, expedition->personal inventory) which guards the cross-region leak the post-impl review caught. See DR-031.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Consolidate the divorced combat + economy halves into one base-local loop. BaseFieldSpawnSystem tops up RegionTag{Base} Ore nodes around the plot; harvest routes Base->shared ledger and Expedition/untagged->personal inventory for BOTH the projectile (ResourceHarvestSystem) and melee (MeleeComboSystem server-only block, writes Remaining back for VFX). Activate the reserved Schedule source in ThreatDirectorSystem so base sieges arm WITHOUT an expedition trip (the loop-closer: previously zero waves ever attacked a base-only player). Region-filter the ExpeditionFieldSystem teardown so it no longer wipes the permanent base field. HudSystem shows phase-aware loop copy. See DR-031.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Session log for the PrototypeCameraRig walk-shift fix: root cause, the separate-view-position fix, genre research (Diablo/PoE/Hades/Cinemachine), and the live-tunable knobs.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The movement look-ahead added its 2.5u lead raw onto the framed point and smoothed only the final camera position, so the lead target snapped with the instantaneous input -- start/stop panned the view and reversing direction swung it ~5u (the jarring walk-shift).
Fix with the genre-standard separate view-position technique: ease a dedicated _leadOffset toward the desired lead via a new gentle LeadSharpness knob, independent of FollowSharpness. Cut the lead magnitude 2.5->1.0 (AimLeadDistance, 0 = fully centred like Diablo/PoE) and pull the default zoom 16->13 (~19% closer). Code defaults + the live Game.unity Main Camera values both updated. See 2026-06-11_Camera_Feel_LookAhead_Zoom.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
An attack clip that keys ONLY the Root bone makes Rukhanka collapse every un-keyed bone (Hips/Spine/legs) to identity for the duration of the state -> the body folds halfway into the floor (player MeleeSwing + enemy Attack; writeDefaultValues does NOT prevent it -- confirmed since even a pure yaw, which is height-preserving, still sank). Fix: build the attack clips FROM the full idle pose (every bone keyed) + a Root YAW twist on top, so nothing is un-keyed. Applied to both runtime clips (PlayerMeleeSwing/EnemyAttackWindup) and both editor recipes (PlayerRigTools/EnemyRigTools) so rebuilds stay correct. Operator-verified in Play.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
These are editor crash-recovery scene dumps (not project content) that had been accidentally committed. Removed from the repo + disk and added to .gitignore so future recovery dumps stay out of git.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
CombatFeedbackSystem.UpdateEnemyDanger paints a red ground danger cone at each enemy while its AttackWindup counts down -- oriented along the enemy facing, sized to EnemyStats.AttackRange, brightening + scaling as the strike nears (intensity = 1 - ticksRemaining/22) so the player reads WHERE + WHEN to dodge. Client-only observe-only; one pooled mesh per winding-up enemy, pruned each frame. Play-verified (14-enemy wave, 8 telegraphs at once, zero errors).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
kronic
Luis Gonzalez