Summary
LLMs create spaghetti architectures by violating layer boundaries—routes accessing databases directly, domain logic mixed with infrastructure. ESLint boundary rules automatically enforce layered architecture (presentation → application → domain → infrastructure), catching violations at lint-time and guiding LLMs to respect architectural patterns through immediate feedback.
The Problem
Without enforced boundaries, LLMs leak code between architectural layers—presentation accessing infrastructure directly, domain logic depending on external APIs, business rules scattered across layers. This creates unmaintainable spaghetti code that’s hard to test, debug, and refactor. Traditional code reviews catch this too late; LLMs need real-time feedback during generation.
The Solution
Use ESLint boundary rules (eslint-plugin-boundaries) to automatically enforce layered architecture constraints. Define clear layers (presentation, application, domain, infrastructure) with explicit dependency rules (presentation can use application/domain, application can use domain, domain depends on nothing). Linting fails immediately when LLM violates boundaries, forcing correction and teaching proper architecture through rapid feedback loops.
Related Concepts
- DDD Bounded Contexts for LLMs
- Type-Driven Development
- Sub-Agent Architecture
- Making Invalid States Impossible
- Invariants in Programming and LLM Generation
- Hierarchical Context Patterns
- Test Custom Infrastructure
- Quality Gates as Information Filters
- Custom ESLint Rules for Determinism
