้ฆ–้กต โ€บ entities โ€บ claude code game studios
projectworkflowagentmulti-agentclaude-code type: entity ๅˆ›ๅปบ: 2026-04-08 ๆ›ดๆ–ฐ: 2026-04-08

Claude Code Game Studios

Overview

An open-source game development workflow system built on Claude Code. Turns a single Claude Code session into a full indie game studio via 49 coordinated agents, 72 skills, and a structured hierarchy mirroring real studio organization.

Repository: https://github.com/Donchitos/Claude-Code-Game-Studios License: MIT

Key Facts

  • 49 agents โ€” each owns a specific domain (programming, art, design, audio, etc.)
  • 72 skills โ€” reusable workflows (sprint planning, code review, brainstorm, architecture-decision, etc.)
  • 12 hooks โ€” event-driven automation
  • 11 rules โ€” enforced constraints per layer
  • Built for: Claude Code (Anthropic's CLI agent)
  • Configurable engines: Godot 4 / Unity / Unreal Engine 5

Architecture Pattern

User-driven collaboration, NOT autonomous AI generation. Every task follows:

Question โ†’ Options โ†’ Decision โ†’ Draft โ†’ Approval

Agents must ask "May I write this to [filepath]?" before writing files. Multi-file changes require explicit approval for the full changeset.

See claude-code-game-studio-collaboration-protocol for full protocol and examples.

Directory Structure

/                    # Root
โ”œโ”€โ”€ CLAUDE.md        # Master configuration
โ”œโ”€โ”€ .claude/         # Agent definitions, skills, hooks, rules, docs
โ”œโ”€โ”€ src/             # Game source code (core, gameplay, ai, networking, ui, tools)
โ”œโ”€โ”€ assets/          # Game assets (art, audio, vfx, shaders, data)
โ”œโ”€โ”€ design/          # Game design documents (gdd, narrative, levels, balance)
โ”œโ”€โ”€ docs/            # Technical documentation
โ”‚   โ””โ”€โ”€ engine-reference/  # Version-pinned engine API snapshots
โ”œโ”€โ”€ tests/           # Test suites (unit, integration, performance, playtest)
โ”œโ”€โ”€ tools/           # Build and pipeline tools
โ”œโ”€โ”€ prototypes/      # Throwaway prototypes (isolated from src/)
โ””โ”€โ”€ production/      # Sprint/milestone/release management
    โ”œโ”€โ”€ session-state/     # Ephemeral session state (gitignored)
    โ””โ”€โ”€ session-logs/      # Session audit trail (gitignored)

See claude-code-game-studio-directory-structure for details.

Agent Hierarchy

Leadership Tier (Opus model)

  • creative-director โ€” Creative authority: vision, pillars, aesthetic direction, scope arbitration
  • technical-director โ€” Technical authority: architecture, engine choices, code quality gates
  • producer โ€” Production: sprint planning, milestone tracking, risk, cross-department coordination

Department Heads (Opus/Sonnet)

  • Art Director โ€” Visual execution of creative direction
  • Audio Director โ€” Sonic execution of creative direction
  • Narrative Director โ€” Story execution of creative direction
  • QA Lead โ€” Testing strategy and quality gates
  • Community Manager โ€” Player community, social media, communications

Leads (Sonnet model)

  • game-designer โ€” Core loops, systems, progression, combat mechanics
  • lead-programmer โ€” Code architecture, coding standards, code review
  • Level Designer, Systems Designer, Economy Designer, Live Ops Designer, Localization Lead

Specialists ( Sonnet/Haiku)

  • Engine specialists: godot-specialist, godot-gdscript-specialist, godot-csharp-specialist, godot-gdextension-specialist, godot-shader-specialist
  • Engine specialists: unity-specialist, unity-ui-specialist, unity-shader-specialist, unity-dots-specialist, unity-addressables-specialist
  • Engine specialists: unreal-specialist, ue-blueprint-specialist, ue-gas-specialist, ue-replication-specialist, ue-umg-specialist
  • Domain specialists: engine-programmer, gameplay-programmer, ai-programmer, network-programmer, ui-programmer, tools-programmer
  • security-engineer, devops-engineer, performance-analyst, analytics-engineer, technical-artist
  • prototyper, ux-designer, accessibility-specialist, qa-tester, release-manager, sound-designer, writer, world-builder

Model Tier Assignment

Tier Model When to use
Haiku claude-haiku-4-5 Read-only status checks, formatting, simple lookups
Sonnet claude-sonnet-4-6 Implementation, design authoring, analysis โ€” default
Opus claude-opus-4-6 Multi-document synthesis, high-stakes phase gates, cross-system review

Haiku skills: /help, /sprint-status, /story-readiness, /scope-check, /changelog Opus skills: /review-all-gdds, /architecture-review, /gate-check

Coordination Rules

  1. Vertical Delegation: Leadership โ†’ Department Heads โ†’ Specialists. Never skip a tier for complex decisions.
  2. Horizontal Consultation: Same-tier agents may consult but cannot make binding decisions outside their domain.
  3. Conflict Resolution: Escalate to shared parent. If no shared parent, escalate to creative-director (design) or technical-director (technical).
  4. Change Propagation: Design changes affecting multiple domains โ†’ producer coordinates.
  5. No Unilateral Cross-Domain Changes: Agents cannot modify files outside their designated directories.

See claude-code-game-studio-architecture for full coordination details.

Coding Standards

  • All game code must include doc comments on public APIs
  • Every system must have an ADR in docs/architecture/
  • Gameplay values must be data-driven (external config), never hardcoded
  • All public methods must be unit-testable (dependency injection over singletons)
  • Commits must reference the relevant design document or task ID
  • Verification-driven development: Write tests first when adding gameplay systems

Design Document Standard (GDD)

Every mechanic document must contain 8 required sections:

  1. Overview โ€” one-paragraph summary
  2. Player Fantasy โ€” intended feeling and experience
  3. Detailed Rules โ€” unambiguous mechanics
  4. Formulas โ€” all math defined with variables
  5. Edge Cases โ€” unusual situations handled
  6. Dependencies โ€” other systems listed
  7. Tuning Knobs โ€” configurable values identified
  8. Acceptance Criteria โ€” testable success conditions

Subagents vs Agent Teams

  • Subagents (current, always active): Spawned via Task within a single Claude Code session. Share session permission context. Default pattern.
  • Agent Teams (experimental, opt-in): Multiple independent Claude Code sessions coordinated via shared task list. Requires CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1.

Use agent teams when: work spans multiple subsystems that won't touch the same files, each workstream would take >30 minutes and benefits from true parallelism, a senior agent needs to coordinate 3+ specialist sessions on different epics.

Related Concepts