Detection Modes

AIM supports three enforcement detection modes, each suited for different types of governance rules.

Static Detection

Deterministic, tool-based enforcement using established static analysis tools.

enforcement: static
detect:
  type: tool
  command: "tsc --noEmit"

Best for: Code quality, type safety, security patterns, formatting

Tools: TypeScript (tsc), ESLint, Semgrep, Prettier, custom scripts

Pattern Detection

Regex-based pattern matching for quick, lightweight checks.

enforcement: static
detect:
  type: pattern
  match: "console\\.log\\("

Best for: Simple prohibited patterns, naming conventions, comment requirements

Semantic Detection

LLM-as-judge evaluation for nuanced, context-aware enforcement.

enforcement: semantic
detect:
  type: semantic
  criteria: "Code follows clean architecture principles with proper separation of concerns"
  model: claude-sonnet
  threshold: 0.8
  few_shot:
    - input: "Direct database call in controller"
      expected: false
    - input: "Repository pattern with dependency injection"
      expected: true

Best for: Architecture decisions, code quality beyond syntax, business logic validation, tone/style guidelines

Choosing a Detection Mode