simplify-ai-architecture
Simplify AI Architecture
Core Problem This Skill Solves
Ritualistic layered architectures (Clean Architecture, Hexagonal, DDD Tactical patterns applied verbatim) were designed for human navigability. They impose a hidden tax in AI-assisted codebases:
- Token inflation: A single-field addition touches 10–18 files across layers.
- Context dilution: The LLM opens 7–13 files to reconstruct what a single vertical slice would show in 1–3.
- Precision loss: When dependencies are hidden behind interfaces resolved only at runtime, LLMs correctly identify needed files ~76% of the time (Navigation Paradox, 2026).
- Abstraction hallucination: LLMs are biased toward complex patterns (trained on blogs where Saga/Event Sourcing/Hexagonal are overrepresented); they suggest sophistication that doesn't fit the problem.
The Rule Set
1. Strategic DDD first, tactical patterns never by default
More from matheusbuniotto/skills-library
deep-agents
Build, review, and reason about LangChain Deep Agents systems. Use when working with `deepagents`, `create_deep_agent`, agent harness design, planning with `write_todos`, virtual filesystem backends, permissions, subagents, context management, skills, memory, sandboxes, interpreters, human-in-the-loop flows, or when deciding whether Deep Agents is the right abstraction for a long-running agent.
2grug
Pair on engineering work with a pragmatic staff-engineer style that favors simplicity, legibility, reversible decisions, and working artifacts over premature abstraction. Use when the user asks for implementation help, architecture choices, code review, decomposition of a large task, or when analysis paralysis, perfectionism, scope creep, or excessive stack selection is slowing progress.
1