Manage a solo game project with a simulated studio hierarchy
When directing AI for a solo game, it is easy to hardcode magic numbers or skip design documents. This repository places 49 specialized agents and 74 workflows within the AI session to replicate the hierarchy of a real game studio. You make all decisions while the team asks the right questions, catches errors early, and manages the project systematically from initial concept to release.
Studio Hierarchy and Role Division
The agents are organized into a three-tier hierarchy similar to a real studio. Each agent has a clear scope of responsibility, reporting structure, and quality gates to prevent role overlap. The repository includes agent sets specialized for Godot 4, Unity, and Unreal Engine 5, allowing you to select the engine that fits your project. Model usage is flexible: supervisors use Opus, leads and some specialists use Sonnet, and the rest inherit the current session's model.
Process Intensity and Automatic Safeguards
The core of this system is the modes.rigor setting in the project.yaml file. You can adjust the process intensity across three levels: minimal, standard, and full, with minimal as the default. minimal minimizes documentation and QA evidence for jam game speed, while full requires all GDD sections and thorough reviews. Additionally, 15 hooks run automatically on commit, push, and file read/write events to check for hardcoded values, JSON errors, and missing design documents. These hooks block or warn about dangerous tasks; if Python is not available, some validations are skipped, but data JSON commits are still blocked.
Decision Authority Remains with the User
This is not an autopilot system that completes games autonomously. All agents ask questions before proposing solutions, presenting 2 to 4 options with their pros and cons, and then wait for your decision. Nothing is recorded without final approval, and you always retain control. Stories containing changes that affect the player's view are not considered complete until the game runs and screenshots are saved to production/qa/evidence/. This measure prevents issues where the screen does not render correctly despite passing tests.
Pre-Application Considerations
This repository is a template, assuming you will clone it and then modify and extend it directly. Git, Claude Code, Python 3, and Bash are required; on Windows, use the Bash included in Git for Windows. The minimal mode is the default because heavier processes do not guarantee a better game and do not offer advantages when a new person takes over the project. As the project grows, you can increase the intensity using the /settings command, and if you want to manage specific systems with stricter criteria, you can use system_overrides. It is released under the MIT license and has been tested on Windows 10 (Git Bash), macOS, and Linux.
