01 / ARCHITECTURE AS CODE
@kekkai/blueprint
Turn architectural decisions from documented consensus into an executable system shared by engineers, linting, CI, and AI Agents.
When AI begins generating code at scale, architecture—not functionality—is often the first thing quietly eroded. Blueprint expresses how a system should grow as a source that can be compiled, executed, and verified.
Documentation drifts, verbal consensus is lost, and every Agent session may guess the architecture again. File placement, dependency direction, module responsibility, and package ownership cannot depend on the next person remembering them.
A single blueprint.config declares layers, dependency flow, module shape, ownership, naming, and quality rules, then generates ESLint rules, an Architecture Handbook, and an Agent Contract together. Rules and explanations come from the same source of truth, so they cannot drift independently.
Survey first outputs deterministic project facts, then a person or Agent makes architectural judgments. Baseline locks in existing debt and blocks only new violations. Adoption neither pretends a legacy system has no problems nor forces a complete rewrite at once.
Different verification layers address risks unit tests cannot see: real ESLint composition; Vite, Next.js, Vue, React, and monorepo fixtures; Windows paths; the published CLI; weekly terrain runs; and mutation testing. The point is not the number of tests, but clear failure responsibility at every layer.
@kekkai/structure-lint originally focused on one-way dependency flow. Those capabilities are now part of Blueprint’s broader Architecture as Code model. The former repository is archived: it is the origin of the evolution, not a parallel product.
Sky-1945 is an unconventional stress test for Blueprint. The view below reads the GitHub API directly to present the repository’s current src/ structure, making the shape created by applying architectural rules to a real codebase directly inspectable.