System Design Skill

Your coding agent will happily design a Kafka cluster for your 2,000-user startup. It will confidently explain the "best" architecture with zero capacity math, zero failure analysis, and zero interest in the codebase you already run.

This skill fixes that. It makes agents design the way senior engineers actually do — **numbers first, trade-offs explicit, failures walked, complexity justified** — and it does it on the problems you actually have: the existing codebase, the overdue migration, the design doc that's about to get reviewed by people who've seen this before.

Why this exists

Two failure modes define AI-generated architecture today:

1. Scale theater. "We'll need Kafka, Kubernetes, and microservices — for scale." For a product with 2,000 users. The skill's right-sizing gate classifies your real scale into one of six tiers and deletes or justifies anything above it.
2. Hand-waving. No numbers, no failure modes, no non-goals, no cost. "It's scalable because we use microservices." The skill's four gates make these omissions failures, not opinions.

The result is a difference you can hear:

You:  "Design a rate limiter for our API across 100 gateway nodes."

Wi