Usage & Commands

Commands

Default (interactive)

npm create ai-native-project

Runs the guided wizard. Behavior depends on the current directory:

• New project — if the current directory isn't an AI-native project, the CLI asks for a folder name and scaffolds a fresh project into it.
• Existing project — if it detects a project (or an .ai-native-project.json), it adds the setup to the current directory. Existing files are never overwritten; composed files (CLAUDE.md/AGENTS.md, package.json, docker-compose.yml, CI) are appended to / merged so new selections actually land.

--boot <name>

npx create-ai-native-project --boot my-app

Creates the my-app folder and scaffolds into it. If that folder already contains a project, it's treated as an existing project (adds, never clobbers).

update

npx create-ai-native-project update

Refreshes the local template registry cache (clones it if missing, otherwise fetches and hard-resets to the configured ref). See Configuration.

The interactive flow

target (new / existing / --boot)
  → project name + brief     (injected into CLAUDE.md / AGENTS.md)
  → agentic coding tool(s)   (Claude Code / Codex / OpenCode / Cline)
  → project type             (single / monorepo)
  → tech stack(s)
  → database(s) · vector-db · ORM · storage · auth · IaC · security standard(s)
  → MCP server(s)            (chrome-devtools, context7, playwright, github, …)
  → skills · agents
  → docs folder?  (Docusaurus / Fumadocs / Starlight)
  → CI?           (GitHub Actions / GitLab CI)
  → Docker?       (docker-compose)
  → generate
  → optional dependency install

Multi-select steps (stacks, databases, skills, …) let you pick any combination. On an existing project, already-installed options are filtered out of the pickers, and the fixed project type is reused without asking again.

What gets generated

• An instructions file — CLAUDE.md and/or AGENTS.md, composed from the project-type base plus a section per stack/database/vector-db/storage/auth/IaC/ security selection. The project name and brief you enter at the start are substituted into the file's title, description, and Overview so the agent knows what the project is from the first line.
• Security standards — pick from 7 priority groups (Essential, Infrastructure, Enterprise Governance, Testing & Validation, Identity & Access, Supply Chain, Industry Compliance); each appends a ## Security: section mapping its standards (OWASP, NIST, CIS, SLSA, ISO, …) to concrete guidance. Picking any group also installs the security-standards skill and the threat-modeler + security-auditor agents.
• MCP servers — pick from a curated set (Chrome DevTools, Playwright, Context7, GitHub, Filesystem, Git, Fetch, Sequential Thinking, Memory). The CLI writes each selected tool's own MCP config — .mcp.json (Claude Code), the mcp block of opencode.json (OpenCode), .codex/config.toml (Codex) — and appends a short section per server. See MCP Servers.
• Skills & agents in each selected tool's layout (see Agentic Coding Tools).
• A runnable package.json (project-type base + stack + ORM fragments, merged).
• Optional docker-compose.yml, CI pipeline, and docs site.
• The core set on every project: the engineering-standards, knowledge-base, and using-create-ai-native-project skills, plus the code-reviewer and security-reviewer agents.
• .ai-native-project.json — the project manifest.

After writing, the CLI offers to install dependencies (only if it wrote a package.json).