A software development lifecycle for developers who build fast with AI assistance but need the result to survive production: real users, real costs, real teams, real failure modes.
This is a skill file (for use with Claude / AI coding assistants) that teaches you to vibe smarter from the start. It doesn't ask you to stop vibecoding — it asks you to draw boundaries before you write code, so the system you build fast is also the system you can extend, debug, split apart, and operate in production.
Core insight: A monolith written with internal service boundaries is already halfway to a microservice architecture. The split is a mechanical operation, not a rethink.
- The Monolith Trap — why vibecoding defaults to monoliths and how to avoid it
- Boundary Map First — name your services before you write code
- SDLC Stages — discipline for each phase of development
- LLM-Specific Patterns — streaming, structured outputs, evals, caching, vendor fallback
- Observability Contract — what to monitor and how
- Security Baseline — minimum viable security for production
- Decision Framework — when to split, when to stay monolith
- CI/CD Gate Reference — what to check before shipping
- Failure Modes — common ways vibecoded products break
Use this when you're vibecoding anything that involves user tiers, permissions, LLM API calls, billing, multi-service integrations, or a team of more than one person shipping to it.
Don't use for: pure frontend UI work, simple scripts, one-off data processing, or single-user local tools.
v2.0 — adds complete boundary maps, corrected data layer classification, multi-tenancy guidance, expanded LLM Gateway interface, dynamic prompt assembly, async job architecture, minimum viable test pyramid, and CI/CD gate guidance.