Turn a product idea into an AI-agent-ready spec in 5 minutes.

Interview → SPEC.md + agent instructions → hand it to your coding agent.

Quick start · How it works · Output · FAQ

✨ Why?

Every new project starts the same way: you have a vague idea, you open an AI coding agent, and you spend the next 90 minutes ping-ponging about what stack to use, whether you need auth, what the core flows are — before a single file is written.

quickStart compresses that to a 5-minute interview. You answer a branching set of questions. It writes a structured spec your agent reads before it writes code. Fewer hallucinations. Less rework. Actual focus.

  💡 idea  →  ❓ interview  →  📄 SPEC.md  →  🤖 your agent builds it

🚀 Quick start

npx quickstart-ai

That's it. Answer the questions. Drop the folder into Claude Code, Cursor, Windsurf, Copilot, or Aider. Done.

🧠 How it works

1. Pick your agent(s). Claude Code, Cursor, Windsurf, Copilot, Aider, or a vendor-neutral AGENTS.md. Select one or many.
2. (Optional) Smart pre-fill. If ANTHROPIC_API_KEY is set, paste a 2–5 sentence description and Claude drafts the answers for you. You still step through each one — accept, edit, or override.
3. Choose your comfort level. Non-technical mode asks product questions in plain English and lets your agent pick stack/deploy details. Technical mode keeps every knob.
4. 10-section interview. Branching questions across Identity, Users & Auth, Platforms, Stack, Data, Core Features, Integrations, Design, Deploy, Ops. Irrelevant sections skip themselves (no DB questions if you said no database; no mobile questions if you said web-only).
5. Two files get written. SPEC.md (the product spec) and one correctly-named instruction file per agent you selected.

📦 Output

A ready-to-commit folder:

my-new-project/
├── SPEC.md                    # Structured product spec (the source of truth)
├── CLAUDE.md                  # Claude Code instructions
└── .cursorrules               # Cursor instructions

SPEC.md contains

• Overview · audience · problem · success metric
• Approved tech stack and design decisions
• Full feature list and primary user journey
• Suggested build order that adapts to your answers (scaffold → DB → auth → features → deploy)
• Open Questions — every "not decided" flagged for your agent to ask about

Which agents are supported?

| Agent | File written | | ------------------ | ------------------------------------ | | 🟣 Claude Code | CLAUDE.md | | ⚫ Cursor | .cursorrules | | 🌊 Windsurf | .windsurfrules | | 🐙 GitHub Copilot | .github/copilot-instructions.md | | 🛠️ Aider | CONVENTIONS.md | | 🌐 Universal | AGENTS.md (vendor-neutral) |

🤖 Smart pre-fill (optional)

Set ANTHROPIC_API_KEY in your environment and quickStart will offer to bootstrap the interview from a short paragraph:

export ANTHROPIC_API_KEY=sk-ant-...
npx quickstart-ai

You describe your project in 2–5 sentences. Claude drafts answers for every question it can confidently answer. You step through each one with the draft as the default — accept, edit, or skip. Typically cuts interview time in half for a clear idea.

Without the key set, quickStart runs fully offline — nothing is sent anywhere.

🎯 Next step

cd my-new-project
claude            # or: cursor . / windsurf . / aider, etc.

Your agent reads its instruction file → follows the pointer to SPEC.md → starts scaffolding. First task usually comes from the "Suggested Build Order".

❓ FAQ

How long does it take? 5–10 minutes depending on how many questions you skip. Smart pre-fill cuts it further.

Do I have to use Claude? No. Smart pre-fill uses Claude, but the whole interview-and-write-spec flow is 100% offline and works with any agent. Pick whichever agent(s) in the multiselect and it writes the right file(s).

Can I re-run it on an existing project? Yes. Point it at the folder and confirm the overwrite. The Appendix — Raw Answers in SPEC.md stores your previous answers as JSON for reference.

Can I customize the questions? Fork the repo and edit src/questions.ts. It's a plain array of 100 objects with conditional skipIf predicates — easy to extend.

Does it work on Windows? Yes. Built on Node 18+; tested on Windows, macOS, and Linux.

🛠️ Development

git clone https://github.com/NijeMatija/quickStart.git
cd quickStart
npm install
npm run dev       # run locally with tsx (no build)
npm run build     # compile TypeScript to dist/
npm start         # run the built CLI

📜 License

MIT — do whatever you want. Attribution appreciated but not required.

If this saved you time, a ⭐ on GitHub would mean a lot — and if it saved you a lot of time, you can ☕ buy me a coffee.