Back to Home

DEMIURGOS: unified rules for AI agents in development

The article describes the DEMIURGOS system for creating unified rules for AI agents in development. It covers architecture, application examples in Godot, Python, and frontend, as well as benefits for teams.

DEMIURGOS: Make AI agents work by your rules
Advertisement 728x90

DEMIURGOS: Unified Rules for AI Coding Agents in Development

AI agents for code generation often produce inconsistent results, ignoring project specifics. The solution is DEMIURGOS, a system that automatically creates and maintains a single set of rules for all coding agents on your team.

DEMIURGOS Architecture: Three Layers of Control

DEMIURGOS is built on a three-layer architecture that ensures flexibility without redundancy.

  • Core (.rules/): Centralized source of project rules. It defines the tech stack, coding standards, architectural patterns, and constraints.
  • Native Adapters: Lightweight projections of the core tailored to specific AI agents. For example, .cursor/rules/*.mdc for Cursor, CLAUDE.md for Claude Code, .github/copilot-instructions.md for GitHub Copilot. Rules aren't duplicated—adapters reference the core.
  • Extensions: Optional layer for complex scenarios, like multi-agent systems or MCP (Model Context Protocol) integrations. Created only when truly needed.

Real-World Use: Examples Across Tech Stacks

The system shines in various tech contexts, spotting rule gaps via debug mode.

Google AdInline article slot

Game Development with Godot 4.3 and GDScript

When building a health component, the agent follows rules to generate typed code using node composition and signals. Debug mode (/debug full) uncovers gaps:

  • Missing conventions for input action names (e.g., player_dash).
  • Lack of rules for linking abilities to UI via cooldown_started signals.
  • Suggestion to replace SceneTreeTimer with Timer nodes for pause-support.

Python Backend with Async SQLAlchemy

For an order API, the agent applies three-layer architecture rules, custom exceptions, and batch queries. Debug reveals:

  • No exception-to-HTTP-status mapping in the router.
  • Missing rules for race conditions using SELECT FOR UPDATE.
  • Unclear separation of flush() in services vs. commit() in routers.

Frontend with Astro and SolidJS

Building a theme toggle, the agent sticks to Islands Architecture and UnoCSS rules. Gaps include:

Google AdInline article slot
  • No FOUC prevention strategy via inline <head> scripts.
  • Missing universal try/catch for Browser APIs.
  • Uncertainty between createSignal and createStore for state.

Key Benefits of Implementing DEMIURGOS

Adopting this rules system delivers measurable wins for dev teams.

  • Code Consistency: All AI agents follow the same standards, cutting review and refactoring time.
  • Faster Onboarding: New devs pull rules via git, and their agents instantly generate team-aligned code.
  • Iterative Improvement: Debug mode (/debug) spots rule gaps after 10–20 prompts, prioritizing fixes. Common areas:

- Game design conventions in game projects.

- Business logic and transactions in backends.

Google AdInline article slot

- Browser API edge cases and accessibility in frontends.

  • Supports 22+ Tools: One core rule set projects to Cursor, Windsurf, Claude Code, GitHub Copilot, Kiro, JetBrains AI Assistant, and more via adapters.
  • Minimal Overhead: Rules live as repo files—no extra infra or complex setup needed.

Key Takeaways

  • DEMIURGOS fixes AI agents' lack of project context with centralized rules.
  • Three-layer architecture keeps things flexible without code duplication.
  • Debug mode uncovers rule gaps for ongoing refinement.
  • Supports 22+ dev tools for team-wide consistency.
  • Implementation cuts agent management time and boosts code quality.

— Editorial Team

Advertisement 728x90

Read Next