create-ai-native-project

Interactive CLI that makes a project AI-native: it scaffolds an instructions file (CLAUDE.md / AGENTS.md), reusable skills and agents, runnable stack boilerplate, docker-compose, and CI for Claude Code, OpenAI Codex, and OpenCode — all composed from a central template registry.

📖 Documentation → · ▶ More demos →

Usage

# new or existing project — interactive wizard
npm create ai-native-project

# create a folder and scaffold into it
npx create-ai-native-project --boot my-app

# refresh the local template registry cache
npx create-ai-native-project update

Also published as the alias create-ai-native-starter — npm create ai-native-starter runs the exact same wizard.

• Bare run — detects whether you're in an existing project. New folder → asks for a name and scaffolds; existing project → adds to it, never overwriting your files (new selections are appended/merged into CLAUDE.md, package.json, docker-compose.yml, and CI).
• --boot <name> — creates that folder and scaffolds into it.

What it asks

1. Agentic coding tool(s) — Claude Code, OpenAI Codex, OpenCode, and/or Cline. One source is retargeted per tool: CLAUDE.md/AGENTS.md, .claude/.opencode/ .codex agents, and shared skills. See Agentic Coding Tools.
2. Project type — single or monorepo (fixed once set).
3. Tech stack(s) — React, Next.js, React Native, Flutter, NestJS, Laravel, Python FastAPI, Python Streamlit — each ships runnable boilerplate, a Dockerfile, CI job fragments, and a specialist agent.
4. Databases (Postgres/MySQL/MongoDB), storage (MinIO/AWS S3), auth (JWT/Clerk) — each contributes instructions sections and compose services.
5. MCP servers — a curated developer set (Chrome DevTools, Playwright, Context7, GitHub, Filesystem, Git, Fetch, Sequential Thinking, Memory) written to each tool's own config (.mcp.json, opencode.json, .codex/config.toml), merged so servers you already have are never overwritten.
6. Skills / agents, an optional docs site (Docusaurus), CI (GitHub Actions / GitLab CI — one job per stack/app), and Docker (docker-compose composed from your selections).

Every project also gets the core set automatically: the engineering-standards, knowledge-base, and using-create-ai-native-project skills plus the code-reviewer and security-reviewer agents.

Monorepo layout

Apps are grouped under apps/ (Turborepo + pnpm workspaces):

apps/
  frontend/website/      # one stack per app
  backend/admin-api/
  mobile/app/
packages/                # shared code (tsconfig, ui, …)
turbo.json
pnpm-workspace.yaml

Docker services (build: ./apps/<group>/<name>, collision-free ports) and CI jobs (scoped to each app dir) are generated per app. Add more apps later by re-running the CLI — it reads the project state and appends the new app's service, CI job, and CLAUDE.md section.

Project state

Every scaffolded project carries .ai-native-project.json recording its type, stacks, apps, databases, storage, auth, CI, docker, docs, skills, and agents. Re-runs read it to skip what's installed, keep the project type fixed, and merge new selections. Claude reads it too (via the injected skill) before extending the project.

Configuration

Templates come from a git registry, cloned to ~/.cache/create-ai-native-project/registry on first use (update refreshes it).

| Env var | Default | Purpose | | --- | --- | --- | | CLAUDE_SETUP_REGISTRY | https://github.com/GeekyAnts/agentic-coding-registry.git | Registry repo URL | | CLAUDE_SETUP_REGISTRY_REF | main | Branch/ref to use |

Set them in a .env in your working directory (see .env.example).

Development

npm install
npm run dev        # run the CLI from source (tsx)
npm run typecheck
npm test           # vitest
npm run build      # tsup → dist/

Requires Node ≥ 20.12. See CLAUDE.md for the full project knowledge base and the template registry's README for how to contribute templates.

Changelog

Notable releases — newest first. Full history and release notes on GitHub Releases.

• 1.4.2 — Scaffolded instructions (CLAUDE.md / AGENTS.md) now include a Knowledge Base (OKF) section across all project shapes — how to maintain the knowledge/ bundle (frontmatter/body conventions, reserved files, guardrails, okflint), plus a directive to read knowledge/ before searching the codebase (registry PR #34).
• 1.4.1 — Windows dependency-install fix; dependency refresh (clack 1.x, commander 15, TypeScript 7); Cline added to the package description.
• 1.4.0 — OKF (Open Knowledge Format) knowledge/ bundle scaffolded into every project.
• 1.3.0 — MCP-servers kind: preconfigure Model Context Protocol servers per tool.
• 1.2.1 — Terminal demos (VHS) + published docs site.
• 1.2.0 — Project name + brief prompts, injected into the generated instructions file.
• 1.1.0 — Security-standards kind (7 priority groups) + security skill/agents.
• 1.0.0 — v1 registry expansion: AI-native stacks, vector-db, ORM, IaC, the Cline target, and more frameworks.

Built by GeekyAnts

create-ai-native-project is built and maintained by GeekyAnts, a product engineering studio trusted by leading enterprises.

Shipping an AI-native product and need a team that moves fast? We do AI-powered product engineering — hire dedicated engineers or talk to us about consulting →

License

MIT © GeekyAnts Inc