้ฆ–้กต โ€บ concepts โ€บ claude code game studio architecture
agentarchitectureworkflowcoordinationmulti-agent type: concept ๅˆ›ๅปบ: 2026-04-08 ๆ›ดๆ–ฐ: 2026-04-08

Claude Code Game Studio Architecture

Overview

A tiered agent hierarchy that mirrors a real indie game studio's organizational structure. 49 agents organized across leadership, department heads, leads, and specialists โ€” coordinated through structured delegation, consultation, and conflict resolution rules.

Agent Hierarchy

[creative-director]     โ† Creative authority (Opus)
      โ†‘
[art-director] [audio-director] [narrative-director] [game-designer] [lead-programmer] [producer] [qa-lead] [community-manager]
      โ†‘                       โ†‘                    โ†‘
[technical-artist] [ux-designer] [writer] [sound-designer] [world-builder] [economy-designer] [level-designer] [systems-designer] [live-ops-designer]
      โ†‘
[gameplay-programmer] [engine-programmer] [ai-programmer] [network-programmer] [ui-programmer] [tools-programmer]
[specialists per engine: godot-*, unity-*, ue-* specialists]

Model Tier Assignment

Skills and agents are assigned to model tiers based on task complexity:

Tier Model When to use
Haiku claude-haiku-4-5-20251001 Read-only status checks, formatting, simple lookups โ€” no creative judgment needed
Sonnet claude-sonnet-4-6 Implementation, design authoring, analysis of individual systems โ€” default for most work
Opus claude-opus-4-6 Multi-document synthesis, high-stakes phase gate verdicts, cross-system holistic review

Skills by Tier

Haiku skills:

  • /help, /sprint-status, /story-readiness, /scope-check, /project-stage-detect, /changelog, /patch-notes, /onboard

Opus skills:

  • /review-all-gdds, /architecture-review, /gate-check

All other skills default to Sonnet. When creating new skills: assign Haiku if only reads/formatting; assign Opus if synthesizing 5+ documents with high-stakes output; otherwise leave unset (Sonnet).

Coordination Rules

Five Core Rules

  1. Vertical Delegation: Leadership agents delegate to department leads, who delegate to specialists. Never skip a tier for complex decisions.
  2. Horizontal Consultation: Agents at the same tier may consult each other but must not make binding decisions outside their domain.
  3. Conflict Resolution: When two agents disagree, escalate to shared parent. If no shared parent, escalate to creative-director (design) or technical-director (technical).
  4. Change Propagation: When a design change affects multiple domains, the producer agent coordinates propagation.
  5. No Unilateral Cross-Domain Changes: An agent must never modify files outside its designated directories without explicit delegation.

Conflict Escalation Map

game-designer โ†” narrative-director โ†’ creative-director (ludonarrative alignment)
art-director โ†” audio-director โ†’ creative-director (aesthetic coherence)
any "changes identity of the game" โ†’ creative-director
technical conflicts โ†’ technical-director
scheduling conflicts โ†’ producer
resource contention โ†’ producer

Subagents vs Agent Teams

Subagents (current, always active)

Spawned via Task within a single Claude Code session. Used by all team-* skills and orchestration skills.

  • Subagents share the session's permission context
  • Run sequentially or in parallel within the session
  • Return results to the parent agent

When to spawn in parallel: If two subagents' inputs are independent (neither needs the other's output to begin), spawn both Task calls simultaneously rather than waiting.

Example: /review-all-gdds Phase 1 (consistency) and Phase 2 (design theory) are independent โ†’ spawn both at the same time.

Agent Teams (experimental โ€” opt-in)

Multiple independent Claude Code sessions running simultaneously, coordinated via a shared task list. Each session has its own context window and token budget.

Use agent teams when:

  • Work spans multiple subsystems that will not touch the same files
  • Each workstream would take >30 minutes and benefits from true parallelism
  • A senior agent (technical-director, producer) needs to coordinate 3+ specialist sessions working on different epics simultaneously

Do not use agent teams when:

  • One session's output is required as input for another (use sequential subagents)
  • The task fits in a single session's context (use subagents instead)
  • Cost is a concern โ€” each team member burns tokens independently

Current status: Not yet used by default. Document usage when first adopted.

Parallel Task Protocol

When an orchestration skill spawns multiple independent agents:

  1. Issue all independent Task calls before waiting for any result
  2. Collect all results before proceeding to dependent phases
  3. If any agent is BLOCKED, surface it immediately โ€” do not silently skip
  4. Always produce a partial report if some agents complete and others block

Team Skills (72 total)

Team Composition Skills

  • team-audio โ€” Audio design team
  • team-combat โ€” Combat design team
  • team-level โ€” Level design team
  • team-live-ops โ€” Live operations team
  • team-narrative โ€” Narrative team
  • team-polish โ€” Polish and feel team
  • team-qa โ€” QA team
  • team-release โ€” Release management team
  • team-ui โ€” UI/UX team

Workflow Skills

  • adopt โ€” Adopt a skill from library
  • architecture-decision โ€” Create ADR through guided flow
  • architecture-review โ€” Validate ADRs
  • brainstorm โ€” Professional ideation with MDA/SDT analysis
  • bug-report, bug-triage โ€” Issue management
  • changelog โ€” Generate changelog
  • code-review โ€” Review code changes
  • consistency-check โ€” Cross-document consistency
  • create-architecture โ€” Create architecture document
  • create-epics, create-stories โ€” Story decomposition
  • design-review โ€” Validate design documents
  • dev-story โ€” Implement from story
  • estimate โ€” Effort estimation
  • gate-check โ€” Phase gate validation
  • help โ€” Built-in help
  • hotfix โ€” Hotfix workflow
  • localize โ€” Localization workflow
  • milestone-review โ€” Milestone validation
  • onboard โ€” Project onboarding
  • patch-notes, playtest-report, qa-plan
  • scope-check, skill-improve, skill-test
  • smoke-check, soak-test โ€” Stability testing
  • sprint-plan, sprint-status, story-done, story-readiness
  • tech-debt โ€” Technical debt tracking
  • test-evidence-review, test-flakiness, test-helpers, test-setup
  • ux-design, ux-review โ€” UX design and review

Gate Verdict System

Agents emit structured verdicts at phase gates:

[GATE-ID]: APPROVE   โ† passes gate
[GATE-ID]: CONCERNS  โ† passes with concerns noted
[GATE-ID]: REJECT    โ† blocks progression
[GATE-ID]: REALISTIC / UNREALISTIC (producer gates)

Calling skills read the first line for the verdict token.

Related