Void Breaker — “Premium Roguelite” Overhaul¶

Date: 2026-06-28 Status: Phase 1 approved for implementation Owner: ka1kqi

Context & North Star¶

Void Breaker is a TypeScript arcade roguelite shooter (~9,800 lines) in the RMH Studios games monorepo. It has a dual 2D-canvas / 3D-three.js renderer with bloom + post-processing, procedural WebAudio, 40 waves across 3 maps, 11 enemy types, 8 bosses, 15 roguelite upgrades, 4 playable characters, 5 run modifiers, meta-progression (Void Cores / Void Forge), and a scripted 40-wave story. A full run is ~10–20 minutes.

The stated goal is to make the game “rival GTA6” with “10 hours of content,” “insane graphics and animations,” and “Destiny 2 / Bungie-level writing.”

Honest scope boundary¶

A browser-based TypeScript arcade shooter cannot literally rival GTA6 (a ~$1–2B, 1000+ person, ~7-year AAA production). We do not pretend otherwise. Instead we target the realistic and genuinely excellent north star for this medium:

Hades / Returnal / Dead Cells tier — enormous replay depth, exceptional game-feel, and genuinely great writing. Premium feel in a web game; we compete on craft and depth, not budget.

Non-goals: photoreal rendering, open world, full voice acting / mocap, literal AAA parity.

The Roadmap (3 phases)¶

Each phase gets its own spec → plan → autonomous-harness build (with parallel agents on independent units) → balance-sim + tests + review → checkpoint → next phase.

Phase	Theme	Headline scope
1. Feel	Game-feel & animation polish	Hitstop, trauma-based shake, enemy spawn/death animations, weapon-specific feedback, impact VFX layering, juicier UI/menus, signature “wow” moments + synced audio. Builds on the existing renderer — no rebuild.
2. Content	Depth for 10+ hours	The big “10 hours” lever: many more enemies, new bosses w/ unique mechanics, new biomes/maps, a much larger upgrade pool with real build archetypes/synergies, weapon variety, deeper meta-progression, daily/challenge modes. Will be decomposed into sub-phases (2a enemies/bosses, 2b biomes, 2c builds/economy).
3. Story	Bungie-tier writing	A real campaign: world bible, character arcs, branching/reactive dialogue, cinematic between-wave beats, and a Grimoire/lore-collectible system. Lands on top of a deep, polished game.

Order: Feel → Content → Story (confirmed). Rationale: each phase stands on a more premium base than the last. Graphics are already in a good spot, so Phase 1 is targeted polish that multiplies what exists, not a renderer rebuild.

Phase 1: Game-Feel & Animation Polish¶

Goal¶

Take combat from “fun” to “premium-feeling” without adding gameplay mechanics or content. Every impactful action should feel weighty, readable, and satisfying.

Core Architecture: the “feel layer”¶

Add a thin, renderer-agnostic feel layer in the engine. The engine decides what impactful thing happened and how hard (a weighted intensity); the renderers decide how to show it. This keeps the 2D and 3D renderers in visual sync, keeps the simulation deterministic and headless-testable, and isolates “feel” from “rules.”

Concretely:

A small amount of new engine state (e.g. hitstopTimer, a trauma accumulator, per-entity animation phase/lifecycle fields, queued “impact” descriptors) surfaced on GameState / entity records.
VoidBreakerEngine.update(dt, input) consumes hitstop by scaling the simulation dt; trauma decays each frame; spawn/death animation lifecycles advance.
Both renderer.ts (2D) and renderer3d.ts (3D) read this state and visualize it. Neither renderer owns timing — they are pure visualizers of engine state.
Audio (audio.ts) impact events are emitted by the engine in sync with the same beats.

Headless-safety: hitstop and animation timing must be a no-op (or fully fast-forwarded) on the sim path so scripts/void-breaker-balance-sim.ts still completes and balance numbers are unaffected. The feel layer changes presentation timing, never gameplay outcomes.

The seven workstreams (parallel-agent units)¶

These are intentionally independent so they can be fanned out to parallel agents inside the harness. Shared engine state changes (workstream 1 & 2) land first as a foundation; 3–7 build on it.

1. Hitstop / impact freeze (engine)¶

Micro time-freeze (scaled dt, typically a few frames) on high-impact events: crits, kills, detonations, boss hits, and player-taking-damage. Duration is trauma-weighted (bigger hit = longer freeze, capped). Drives a brief pause that both renderers and audio respect. No-op on the headless sim path.

2. Trauma-based screen shake (engine + both renderers)¶

Replace flat shake with a directional trauma model: trauma is a 0–1 accumulator that decays over time; shake magnitude = trauma² (or similar) so small events barely register and big ones slam. Events add weighted trauma (boss stomp ≫ pistol shot). Renderers translate trauma into camera/canvas offset + rotation. Respects the existing Reduced Effects accessibility option.

3. Enemy spawn & death animations (engine + both renderers)¶

Spawn: a brief warp-in telegraph (scale/fade/distort) before the enemy becomes active, so enemies don’t pop in.
Death: a shatter / dissolve / implode sequence instead of instant despawn, with per-enemy-type flavor (e.g. tank crumbles, splitter bursts, hive collapses). Death state is a lifecycle on the enemy record so the object pool only frees the slot after the animation ends.

4. Weapon & projectile feel (engine + renderers)¶

Recoil / kickback on fire, muzzle variety, tracer + impact-spark polish, and making distinct builds look distinct (multishot fan spread, piercing as a beam-like trail, high-caliber as heavy slow rounds). Driven by PlayerStats so visuals track the player’s current build.

5. Impact VFX layering (renderers)¶

Directional hit sparks, void-splatter on impact, kill bursts, escalating combo visuals, and damage-number polish (crit emphasis, color tiers, motion). Pure presentation on top of existing particle/popup systems.

7. Signature moments + audio sync (engine + renderers + audio)¶

Boss-death slow-mo + screen flash, surge-mode visual takeover, big-detonation cinematic, biome entrance flourish. Layered/synced procedural audio (audio.ts) so every beat above has matching sound. Reuses the hitstop + trauma primitives from workstreams 1–2.

Quality gates¶

Existing Vitest suite (lib/__tests__/void-breaker-*.test.ts) stays green.
New engine-level tests for the deterministic parts of the feel layer: hitstop dt-scaling, trauma accumulation/decay, death-animation lifecycle (spawn → active → dying → freed).
scripts/void-breaker-balance-sim.ts still completes and produces materially unchanged balance numbers (proves the feel layer is presentation-only).
Manual browser playtest checkpoint at the end of the phase (both 2D and 3D renderers; with and without Reduced Effects).

Scope discipline (YAGNI)¶

No new gameplay mechanics, no new content (enemies, bosses, upgrades, maps) — that is Phase 2.
No renderer rewrite; build on the existing 2D + 3D renderers.
If a “feel” idea turns into a mechanic (changes outcomes, not just presentation), it moves to Phase 2 instead of expanding Phase 1.

Working mode¶

Autonomous Plan → Sprint → Evaluate harness, with parallel agents fanning out the independent workstreams (3–7) once the shared foundation (1–2) is in place. Review checkpoint with the user before moving to Phase 2.