Image: GitHub
Why it mattersAn indie dev or small team can take on Unity, Unreal or Godot work with a role-based agent team. The template makes the AI session act like a studio instead of one assistant.
Building a game with Claude Code gives you one eager assistant and no one asking the awkward design questions. Donchitos packaged a template that turns the same session into a 49-role game studio. The repository is now 29,000 stars and sits 25th on Trendshift's daily list.
The project is called Claude Code Game Studios. It is MIT-licensed and under one person, Donchitos, who has run the public repository since February 2026. The pitch in the README is direct. A single Claude Code chat has no structure. Nothing stops you from writing spaghetti code, skipping design docs, or ignoring QA. The template adds the structure of a real studio.
What is in the template
The counts come from the README. Forty-nine agents sit in three tiers. Three directors cover creative, technical and production. Eleven department leads cover design, programming, art, audio, narrative, QA, release and localisation. Thirty-five specialists cover the rest, from gameplay programming to accessibility.
Seventy-four skills act as slash commands. Named ones cover design brainstorms, architecture, story breakdowns, dev handoff, QA plans and release checklists. Nine /team-* commands run several agents at once on one feature. Fifteen hooks sit on commits, pushes, asset changes, and session events. Fourteen rules enforce coding standards per path. Sixty-two document templates cover design docs, UX specs and sprint plans.
Three game engines have their own agent sets. Godot 4, Unity, and Unreal Engine 5 each get a lead agent and three or four sub-specialists. The three directors ask for Claude Opus. The lead programmer and fifteen specialists ask for Sonnet. The other thirty inherit whatever model the session is on.
The honesty about documents
The README includes a small measurement. The author built four games from one brief at different rigor levels. For each one he counted the documents the agents wrote before the first line of game code. At minimal, one document, the brief. At standard, 30 documents and 58 minutes. All four produced a running, tested game. Two blind reviewers and a human playtest ranked the standard build last. The author's own conclusion: the 29 extra documents bought traceability on the project while the game itself did not improve.
That is a rare caveat in a project this size, and it is worth repeating. More ceremony inside an AI session does not always produce better output. The README's own advice to an indie dev is to pick a lower rigor mode until the project needs more.
Where the model and the game engine stay
Everything runs inside Claude Code on the user's own account. There is no model or inference engine in this repository. The template is prompts, hooks and workflow glue. Art and audio are the user's own job. The three major engines have agent sets, but you still open Unity or Unreal yourself. No sample game is included on purpose. The author says somebody else's game in your repository would be clutter.
An indie studio paying for Claude Code already has the AI budget to try this. Clone the repository, run /start, and see whether the structured session answers better questions than an unstructured one. If 30 documents feels like too much, the minimal mode is in the same template.
Source
Donchitos/Claude-Code-Game-Studios, on GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
