ai-core-kit

Forkable standard · Claude Code · Spec-first

One interview. A spec-first repo that ships.

Fork the kit, run a single archetype-first interview, and /ack-init renders a complete repo — tech specs, a plan, a best-in-class CLAUDE.md, a design system, wired .claude/ tooling, and offline cost telemetry. Deterministic. Zero LLM in the render loop.

$npx @arthurghz/create-ack@latest —archetype fullstack

What one interview gives you

Your answers freeze into a single manifest. A deterministic render engine turns it into a spec-first repo — the same answers always produce the same output.

Step 1Archetype-first interviewAnswer only what applies
Step 2Frozen project.manifest.yamlJSON-Schema validated · source of truth
Step 3Deterministic render · zero LLM

Your fork lands with

Tech specsPRD · ARCHITECTURE · DOMAIN · REQUIREMENTS · NON-GOALS · ROADMAP
PLANSequenced delivery roadmap
Best-in-class CLAUDE.mdContext-engineered, not generic
Design system + requirementsTokens · components · brand · MCP
Wired .claude/ toolingSkills · agents · commands · hooks

Start from the right shape

Pick an archetype and answer only the questions that apply. More stacks are on the roadmap.

Batteries included

Cross-cutting capabilities ship wired into the fork — opt in per project, never imposed.

From zero to shipping in three steps

Fork the standard

Start a brand-new repository from the kit — no manual file shuffling, no copy-paste from someone else’s project.

Run the interview

One archetype-first interview writes a frozen project.manifest.yaml, then the deterministic engine renders your repo from it.

Land spec-first

You arrive with the specs, the plan, a battle-tested CLAUDE.md, the right .claude/ tooling — and minimal code to maintain.

Ready to ship spec-first?

Fork the standard and run your first interview in minutes.