Skip to main content

Overview

Core Principle: AI generates behavior. The engine governs structure.

AI coding agents generate behavior effectively but are unreliable at structural decisions — they invent folder structures, improvise namespaces, and silently violate dependency rules. This engine separates structural governance from AI code generation: the AI declares intent (e.g., "I need a Repository for Orders"), and the engine enforces structure (correct layer, namespace, file path, type kind, accessibility, and interface requirements).

What the Engine Does​

  1. Defines architecture as declarative YAML rule packs (layers with selectors, forbid and allow-only rules, component roles, naming conventions)
  2. Loads and composes multiple rule packs with deterministic merge semantics
  3. Generates code using Roslyn that conforms to the resolved rules
  4. Validates existing solutions against architecture rules with the compiler's semantic model (layer dependencies, forbid and allow-only rules, type checking, interface requirements)
  5. Syncs rule packs from git repositories with semver-aware version resolution
  6. Exposes all capabilities via CLI and an MCP server (stdio and HTTP) for AI agent integration

What the Engine Does Not Do​

  • It does not generate method bodies, business logic, or behavior
  • It does not enforce runtime rules — all checks are static (compile-time / design-time)
  • It does not replace a build system, linter, or test runner — it complements them