GUARDRAIL PROTOCOL
A Unity platformer where real AI-security failures are the level mechanics
What it is
Guardrail Protocol is a game I am building in Unity 6 that turns AI-security concepts into a playable facility. Prompts are physical data capsules, tokens ride conveyors, and retrieval systems look like automated archives. The goal is a platformer that is genuinely fun first, while the security ideas are real rather than decorative.
Technically it is a Unity 6 project on URP with a full overhead 3D camera, the new Input System, and AI Navigation for enemies. The code is split across several assembly definitions (character, enemies, camera, security, and the level slices) so systems stay isolated and testable, backed by edit-mode and play-mode test assemblies. All the 3D assets come from a modular Blender pipeline that bakes reusable factory kits rather than one-off props.
The first level, Context Foundry, is modeled directly on an OWASP prompt-injection scenario: six input sources feeding one request, with inspection and approval gates you have to navigate.
Features
- Stylized 3D facility where prompts, tokens and RAG are physical machines
- Level 1 Context Foundry models an OWASP prompt-injection failure chain
- Full overhead 3D movement with Input System and AI Navigation
- Modular Blender asset pipeline feeding reusable factory kits
- Character selector with multiple playable mascots and animation rigs
- Edit-mode and play-mode test assemblies for security and slice logic
Stack
~29,000 lines of C# · 231 runtime scripts · 70 test files