Donchitos/Claude-Code-Game-Studios 如何把單一 Claude Code 工作階段變成配備 49 個智能體的遊戲工作室

A solo game project in a single Claude Code chat tends to drift. Nothing stops you from hardcoding magic numbers, skipping the design doc, or shipping spaghetti, because no one plays QA or asks whether a feature fits the game's vision.

Donchitos/Claude-Code-Game-Studios gives one Claude Code session the structure of a studio. The README lists 49 agents, 74 skills (slash commands), 15 hooks, 14 path-scoped rules and 62 document templates. You still make every decision, but a team of specialist agents asks questions and flags mistakes along the way.

How the studio is organized

Agents sit in three tiers, like a real studio:

Directors: creative-director, technical-director, producer.
Department leads: game-designer, lead-programmer, art-director, audio-director, narrative-director, qa-lead, release-manager, localization-lead.
Specialists: gameplay-programmer, engine-programmer, ui-programmer, level-designer, sound-designer, qa-tester and more.

Each agent declares its model in its frontmatter. The directors ask for Opus, so expect heavier usage on the big decisions.

Prerequisites

From the README: Git, Claude Code, Python 3 and Bash. The README recommends jq as well. Python matters more than it looks, because it reads project.yaml for every skill and hook. Without it every setting silently falls back to its default. On macOS and Linux Bash is built in, and on Windows it comes with Git for Windows.

Claude Code itself can be installed with the native installer, or with this command on Node.js 22 or newer:

npm install -g @anthropic-ai/claude-code

Setup

Clone the repo as your new game folder:

git clone https://github.com/Donchitos/Claude-Code-Game-Studios.git my-game
cd my-game

Then start Claude Code from the project root, the folder that holds CLAUDE.md and project.yaml:

claude

Do not start it in a subfolder such as src/. The README warns that this skips the project's hooks.

Your first session

The skill list names the workflow phases: /start, /design-system, /create-epics, /create-stories, /dev-story and /story-done. A sensible first pass:

Run /start and answer the questions about the game you want to make.
Use /design-system to write a design document for one mechanic before any code exists.
Break the work down with /create-epics and /create-stories.
Implement one story with /dev-story, then close it with /story-done.

The path-scoped rules apply while you edit gameplay, engine, AI, UI and network code, so standards are enforced where they matter. Hooks validate commits, pushes and asset changes.

Does it produce games, or documents?

The README has a section with this exact question, and it is worth reading before you commit a weekend to it. The templates cover GDDs, ADRs, sprint plans and HUD design, so a large part of what you get is structure and documentation around the code you and Claude write together.

Where it fits

You are building a game alone and want review gates without hiring anyone.
You already use Claude Code and want conventions it follows every session.
You want a tested hierarchy to copy for non-game projects, by editing the agents and rules.

Our full write-up with more detail is on the Claude Code Game Studios use case page. If you are choosing among agent kits, the use cases for Claude Code itself are a good baseline.

Caveats

Fifteen hooks run bash on your machine. Read them in .claude/ before you trust them.
The hierarchy can burn tokens quickly, since many agents may be invoked per task.
This walkthrough follows the README and has not been run end to end by us.

Source repo: Donchitos/Claude-Code-Game-Studios.

Want the setup commands for 50 AI repos people actually ship with? Get the free AI Repos That Actually Ship pack: https://dedupedai.com/ship-pack/?utm_source=devto&utm_medium=article&utm_campaign=shippack

Get one email every Monday with the week's hottest AI repos and exactly how to use them: subscribe free at dedupedai.com