2nd/10_Wiki/Topics/AI_and_ML/Skybound Protocol 코드리뷰.md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-skybound-protocol-코드리뷰	Skybound Protocol 코드리뷰	10_Wiki/Topics	verified	self	Skybound Code Review Skybound Protocol Review	none	A	0.9	applied	skybound code-review game-dev project		2026-05-10	pending	language framework typescript phaser/unity

Skybound Protocol 코드리뷰

매 한 줄

"매 Skybound Protocol 의 code review 의 systematic checklist 의 game-specific concern (frame budget, deterministic sim, asset pipeline)". 매 generic SE review (naming, SRP, test coverage) 위 의 game layer (16ms frame, GC pause, hot path allocation, save/replay determinism). 매 2026 의 AI-augmented review (Claude Opus 4.7, Cursor) 의 first pass automation.

매 핵심

매 Game-specific review axes

• Frame budget: 매 60fps → 16.67ms / frame. 매 hot path 의 allocation X.
• Determinism: 매 replay / netcode 의 same input → same output 보장.
• Asset pipeline: 매 hot reload, 매 import settings, 매 platform-specific compression.
• Memory: 매 pool / arena, 매 GC pause < 1ms.

매 Skybound 의 specific concerns

• Firepower system: 매 damage calc 의 numeric stability (overclock multipliers).
• State sync: 매 client-authoritative vs server-authoritative.
• Asset gen: 매 AI-generated asset 의 validation pipeline.
• Balance: 매 v1.5 overclock 의 PvP / PvE separation.

매 Review checklist

1. 매 hot path allocation 의 X (object pool / struct).
2. 매 random source 의 seeded (replay).
3. 매 numeric: float vs int (deterministic 의 fixed-point).
4. 매 save format versioning.
5. 매 telemetry hook (drop-out, balance metric).

💻 패턴

Hot path allocation 의 review

// BAD: 매 frame 의 allocation
function updateEnemies(enemies: Enemy[], dt: number) {
  for (const e of enemies) {
    const targets = enemies.filter(o => o.team !== e.team);  // 매 frame allocate
    const closest = targets.sort((a, b) => dist(e, a) - dist(e, b))[0];
    e.target = closest;
  }
}

// GOOD: 매 reusable buffer
const _targetBuf: Enemy[] = [];
function updateEnemies(enemies: Enemy[], dt: number) {
  for (const e of enemies) {
    _targetBuf.length = 0;
    let bestD = Infinity, best: Enemy | null = null;
    for (const o of enemies) {
      if (o.team === e.team) continue;
      const d = distSq(e, o);
      if (d < bestD) { bestD = d; best = o; }
    }
    e.target = best;
  }
}

Deterministic RNG (replay-safe)

class XorShift {
  constructor(public state: number) {}
  next(): number {
    let x = this.state;
    x ^= x << 13; x ^= x >>> 17; x ^= x << 5;
    this.state = x >>> 0;
    return this.state / 0xffffffff;
  }
}

// 매 sim 의 single seeded RNG — 매 Math.random() X
const rng = new XorShift(matchSeed);

Firepower overclock validation

function applyOverclock(base: WeaponStats, level: number): WeaponStats {
  const cap = OVERCLOCK_CAPS[base.tier];
  const clamped = Math.min(level, cap);
  return {
    ...base,
    damage: base.damage * (1 + 0.08 * clamped),
    fireRate: base.fireRate * (1 + 0.04 * clamped),
    heat: base.heat * (1 + 0.12 * clamped),  // 매 cost scale faster than benefit
  };
}
// 매 review: 매 cap 의 enforce X 의 → exploit 의 PvP balance break.

Save versioning

interface SaveV2 { version: 2; player: Player; inventory: Item[]; flags: number[]; }

function migrate(raw: any): SaveV2 {
  if (raw.version === 1) {
    return { ...raw, version: 2, flags: raw.flags ?? [] };
  }
  if (raw.version === 2) return raw;
  throw new Error(`unknown save version ${raw.version}`);
}

AI-augmented review (2026)

# 매 Claude Code / Cursor 의 first pass
claude review --diff main..HEAD --rules .claude/skybound-rules.md

# .claude/skybound-rules.md
# - Flag any allocation in update / draw loops
# - Flag Math.random() in sim code (must use seeded RNG)
# - Flag hardcoded balance numbers (must be in DataConfig)

Telemetry hook

function fireWeapon(w: Weapon, target: Entity) {
  const dmg = computeDamage(w, target);
  target.hp -= dmg;
  Telemetry.emit("weapon.fire", {
    weaponId: w.id, dmg, overclock: w.overclock,
    matchId: GameState.matchId, t: GameState.tick,
  });
}

매 결정 기준

상황	Review focus
Sim / netcode	매 determinism + seeded RNG
Update loop	매 allocation + cache miss
Balance code	매 cap enforcement + telemetry
Save / persist	매 version migration
AI-gen asset	매 validation gate

기본값: AI first pass + human review on hot path / balance / determinism.

🔗 Graph

• 부모: Code-Review · Game-Development
• 변형: Skybound_Firepower_Overclock_v1.5 · Skybound_Asset_Generation_Roadmap
• 응용: Game-Performance · Deterministic-Simulation
• Adjacent: AI-Augmented-Review · Telemetry

🤖 LLM 활용

언제: PR diff 의 first pass, pattern violation 의 detect, balance number 의 sanity check. 언제 X: subjective game feel, designer intent — human only.

❌ 안티패턴

• Math.random() 의 sim: 매 replay 의 break.
• Frame allocation: 매 GC spike → frame drop.
• Hardcoded balance: 매 designer 의 iterate 의 X.
• Save without version: 매 migration impossible.

🧪 검증 / 중복

• Verified (Skybound internal review log, GDC perf talks).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — Skybound code review checklist + patterns