AI Product Starter

Turn your idea into a real plan — before anyone writes a line of code.

Got an idea for an app, a product, a tool, a startup? This gives your AI assistant (Claude, Cursor, Copilot, ChatGPT/Codex, Gemini…) the good habit of stopping to plan it with you first instead of jumping straight into code.

It asks you simple questions — one at a time, in plain language — and turns your answers into a clear, written plan you (or a developer, or an AI) can actually build from. You don't need to be technical. It works out what your project needs and writes only that — no jargon-filled documents nobody reads.

What you get

A plan, in plain files you can read and share:

• A roadmap — what to build, and in what order, broken into phases.
• A feature list and spec — what the first version does, as plain "a user can…" stories.
• A timeline and a rough budget — phases with rough durations, the tools and services you'll need, and what it costs to run (it suggests these; you adjust).
• A money plan — pricing, revenue, and an investor pitch deck — only if you're selling or raising. Skipped otherwise.
• A short hand-off brief — so a developer or an AI can pick it up and build without you re-explaining everything.

It adapts: a quick weekend experiment gets a tiny plan; a real startup gets the full set. Anything that doesn't fit your project is simply left out.

Who it's for

• Founders, product managers, makers, non-coders — get a solid plan you can act on or hand to a builder. It stops at the docs; no coding required.
• Developers — keep going past the plan into a ready-to-code repo (see For developers).

Example output

What actually lands in your project — adapted to who you are and what you build.

A founder planning a commercial app with a UI — plan path, stops at the docs:

my-barber-app/
├── AGENTS.md             # short brief to hand a developer or AI later
├── README.md
└── docs/
    ├── ROADMAP.md        # phases + what ships when
    ├── SPEC.md           # features + the key things it tracks, in plain language
    ├── DESIGN-SYSTEM.md  # look & feel (it has a UI)
    ├── MONETIZATION.md   # pricing + a running-cost estimate
    └── BACKLOG.md        # the feature / milestone plan

No git, no tests, no CI — nothing to maintain. Hand AGENTS.md + BACKLOG.md to a builder when you're ready.

A developer building a SaaS now — build path, a ready-to-code repo:

my-saas/                      # git repo, branches: main + staging
├── AGENTS.md                 # operating model: you decide, the AI builds, humans merge
├── CLAUDE.md                 # entry pointer
├── README.md
├── .gitignore
├── .github/
│   ├── pull_request_template.md
│   └── workflows/claude.yml  # @claude review
└── docs/
    ├── ROADMAP.md
    ├── SPEC.md
    ├── ARCHITECTURE.md       # stack + module boundaries
    ├── DATABASE.md           # data model
    ├── DESIGN-SYSTEM.md
    ├── WORKFLOW.md           # git/PR/review/release + test-first
    └── BACKLOG.md            # tickets with status tracking

Different project, different output: a CLI skips DESIGN-SYSTEM, an internal tool skips MONETIZATION + the pitch deck, a weekend hack gets just AGENTS + BACKLOG.

Install

Add it to your AI assistant once, then just talk about your idea.

Any tool — install once, then run init (Cursor, Copilot, Gemini, Codex, Claude, …). Either run it one-off with npx, or install the CLI globally first:

# one-off, no install:
npx ai-product-starter init

# or install the CLI globally, then run it in any project:
npm install -g ai-product-starter
ai-product-starter init

init drops the skill into your project and wires the entry files your assistant reads, so it works whatever tool you use. Add --tool cursor|copilot|gemini|claude to wire just one. Then tell your assistant to plan your project (see Try it).

Claude Code (plugin)

/plugin marketplace add bosznrt/ai-product-starter
/plugin install ai-product-starter@ai-product-starter

Manual — clone the repo and tell your assistant: "Read skills/ai-product-starter/SKILL.md and use it to plan my new project."

Updating

Update it the same way you update any plugin for your assistant.

Claude Code

/plugin marketplace update ai-product-starter
/plugin install ai-product-starter@ai-product-starter

Installed via npx or globally — re-run the installer; it's safe to run again:

npx ai-product-starter@latest init
# if you installed it globally:
npm install -g ai-product-starter@latest && ai-product-starter init

Manual clone — git pull, and your assistant picks up the new skill and templates on the next run.

Try it

Once it's installed, just say what you're dreaming up:

"I want to make an app that helps people find and book a barber nearby — can you help me plan it?"

It interviews you, proposes features, a timeline, and a budget, then writes your plan. You stay in control — it checks the file list with you before writing anything, and never builds or ships on its own.

For developers

Building it now? It doesn't stop at docs — it also sets up:

• Technical docs — architecture, data model, and an AI strategy (when relevant).
• An operating model — you're the Product Owner; the AI orchestrates and reviews but doesn't merge or release — humans do.
• A real git workflow — feature → staging → production, semver, test-first, and independent AI code review with no self-review (Claude, Copilot, or Codex — your call).
• A bootstrapped repo — first commit, main/staging branches, PR template.

It's tool-agnostic: generated repos use an AGENTS.md as the source of truth (most AI coding tools read it) and keep the review/workflow vendor-neutral, so you can use any assistant — or several.

Docs

docs/ has the full guides:

• Getting started — install and your first run.
• How it works — the interview, the two finish lines, and how it chooses files.
• Customizing — tailor the templates and the workflow.
• FAQ.

Contributing

Issues and PRs welcome — improve the skill, the templates, the docs, or add support for another AI tool. See CONTRIBUTING.md.

License

MIT — see LICENSE. Built by BoSz Narathip.