The Safety Layers

Layered permission architecture:

• Command validation — Every command is validated for correct structure and parameters before execution
• Permission checking — Each command is checked against your configured tool permissions before it reaches a handler
• Handler isolation — Commands are routed through category-specific handlers that enforce their own constraints
• File system sandboxing — File operations are restricted to your Unity project directory by default, with protected paths (.git, Library, Temp, Logs) that are always blocked regardless of configuration
• Dry-run default — Destructive file operations require explicit confirmation; simulation mode is the default
• Session isolation — Each AI client session operates independently with its own permission snapshot

Intentional gating philosophy:

These layers exist because we believe safety should not depend on a single point of control. If one layer is misconfigured, the layers beneath it still protect your project.

We designed GameCatalyst so that you can trust it to respect your boundaries, even when the AI client on the other end does not fully understand them.