The spec your coding agent actually follows.
One sentence in. Out: the build spec — stack, schema, auth, pages, security, legal —
and the AGENTS.md, CLAUDE.md and Cursor rules that keep Claude Code,
Cursor or Lovable on it.
Free spec, then the deep spec
Real output for the same project (Ledgerly) — nothing retouched. The free spec frames the build; the deep spec designs every table, route and milestone.
### Business-specific entities
Create tables for the following entities (with appropriate relationships):
Client, Invoice, LineItem, Expense, Receipt, PaymentReminder, Report
For each entity:
- Add `id`, `created_at`, `updated_at`
- Add appropriate foreign keys
- Add indexes on frequently queried columns
The template knows your entities by name. The columns, types and relations are left to your agent.
### `invoices`
Multi-currency invoice with immutable totals once issued
| Column | Type | Null | Notes |
|---|---|---|---|
| `id` | TEXT PK | no | |
| `org_id` | TEXT | no | |
| `client_id` | TEXT | no | |
| `number` | VARCHAR(32) | yes | NULL while draft; assigned from organizations.next_invoice_seq on issue |
| `status` | TEXT | no | ENUM draft\|sent\|partially_paid\|paid\|overdue\|void |
| `currency` | CHAR(3) | no | Invoice currency, immutable after issue |
| `fx_rate_to_base` | NUMERIC(18,8) | no | Rate at issue date, frozen; 1 if currency = base_currency |
| `fx_rate_date` | DATE | yes | Rate source date |
| `subtotal_cents` | BIGINT | no | Minor units of currency |
| `tax_cents` | BIGINT | no | |
| `discount_cents` | BIGINT | no | DEFAULT 0 |
| `total_cents` | BIGINT | no | subtotal - discount + tax |
| `amount_paid_cents` | BIGINT | no | DEFAULT 0; sum of payments |
| `issue_date` | DATE | yes | In org time zone |
| `due_date` | DATE | yes | issue_date + client terms |
| `sent_at` | TIMESTAMPTZ | yes | |
| `paid_at` | TIMESTAMPTZ | yes | |
| `voided_at` | TIMESTAMPTZ | yes | Issued invoices are voided, never deleted |
| `notes` | TEXT | yes | Shown on PDF |
…
Written for the project: every table, typed columns, rules in the notes, relations and indexes. Then routes, acceptance criteria and the build plan.
How it works
3 steps. 5 minutes. A perfect specification.
Describe it, or pick a type
One paragraph and the AI drafts every answer — or choose among SaaS, e-commerce, marketplace, blog, web app, landing page, business site.
Check the answers
The questions developers always forget: auth, payments, edge cases, SEO, security. Correct what the draft got wrong.
Copy, paste, done
Download the spec and one config file per agent. Drop them at the root of your repo; every agent finds its own.
7 supported project types
Each type has its own questions. We only ask what's relevant.
SaaS Application
Web application with subscription, dashboard, auth
E-commerce
Online store, cart, payment, shipping
Business Website
Presentation website, portfolio, landing page
Marketplace
Multi-vendor platform, commissions, reviews
Blog / Media
Blog, magazine, editorial content, newsletter
Web App / Tool
Online tool, calculator, generator, converter
Landing Page
Single conversion page, email collection, launch
Why use a prompt generator?
The problem isn't the tool. It's the specification.
The right questions
We ask the questions you'd forget. What happens when the subscription expires? Who pays Stripe fees? Refund policy?
Complete specification
Not a vague brief. An exhaustive document: DB schema, auth flows, SEO, security, design system, file structure.
One copy-paste
Paste the specification into Claude Code, Cursor or Bolt. Everything is there to generate a working site on the first try.
Adapted to your project
Questions change based on project type. A SaaS doesn't have the same needs as an e-commerce store.
Security by default
CSRF, XSS, SQL injection, bcrypt, HTTPS... all best practices are built into every specification.
Built-in SEO
Meta tags, schema.org, OpenGraph, sitemap, performance... your site will be optimized from day one.
Inside Claude Code, Cursor and your terminal
The same generator, where you already work. Open source, free to install.
MCP server
Your agent reads the questionnaire, answers it from your repo and the conversation, and writes AGENTS.md and CLAUDE.md itself.
claude mcp add --transport http specpack https://prompt-generator-website.com/mcp
Audit in CI
Scores CLAUDE.md, AGENTS.md, .cursorrules for bloat — the lines that cost context on every session and change nothing. Offline.
npx specpack audit --max-bloat 40
API
Types, questionnaire, specs and AI drafts over plain JSON. Two specs free per address, keys for more.
curl https://prompt-generator-website.com/api/v1/types
Simple, transparent pricing
Choose your plan and start generating specifications in minutes.
- 2 template specs
- 3 AI drafts of the questionnaire
- 8-file export pack
- CLI audit & MCP server
- Deep spec (tables, routes, build plan)
- 5 deep specs written for your project
- Unlimited specs for 7 days
- 30 AI drafts
- Saved history & API key
- Never renews — passes stack
- 10 deep specs a month
- Unlimited specs
- 100 AI drafts a month
- Saved history & API keys
- Cancel anytime
Ready to generate your first specification?
No account needed. Describe it, or answer the questions yourself.
Start with a description