◆ Spec-driven development

Stop your AI agent from
building the wrong thing.

GenPRD turns your idea into a testable spec — EARS acceptance criteria and requirement IDs — then hands it to Claude Code or Cursor. The agent builds what you actually meant, the first time.

No credit card · Model-agnostic · Your spec, any agent

Hands off to the tools you already use

Claude Code·Cursor·GitHub Copilot·Windsurf·Cline·Aider·Zed·Gemini CLI·Codex CLI·Continue·Roo Code·Kilo Code·Kiro·Trae·Augment·Bolt·Lovable·v0·Replit·Devin
✕ Prompt-only workflow

“Add login” is wildly underspecified. The agent guesses scope, hallucinates edge cases, drifts on large tasks — and you find out after burning hours, then throw the work away.

✓ Spec-driven with GenPRD

A precise spec is the source of truth. Testable criteria, explicit scope, traceable IDs. Teams report 3–10× higher first-pass success when the agent builds from a real spec.

From idea to buildable spec in 3 steps

The whole point: no more guessing between your head and the agent.

01

Describe the idea

One paragraph. No template gymnastics. Optionally answer a few sharp clarifying questions so nothing is left to guessing.

02

Get a rigorous spec

EARS acceptance criteria (WHEN… the system SHALL…), stable requirement IDs (REQ-/AC-/NFR-), non-goals, and NFRs — graded to the risk of the feature.

03

Hand it to your agent

Export a clean tasks list that cites every requirement. Drop it into Claude Code or Cursor and the agent builds what you actually meant.

spec.md — generated
REQ-3  Passwordless sign-in via WhatsApp OTP
  AC-3.1  WHEN a user submits a valid phone number,
          the system SHALL send a 6-digit OTP within 10s.
  AC-3.2  IF an OTP is entered incorrectly 5 times,
          THEN the system SHALL lock further attempts for 15 min.
NFR-sec-1  OTP codes SHALL be single-use and expire in 5 minutes.

# tasks.md
- [ ] Implement POST /auth/wa/request   // impl REQ-3 | AC-3.1
- [ ] Verify + rate-limit OTP            // impl REQ-3 | AC-3.2 | NFR-sec-1

Not another PRD generator

Most tools stop at a pretty document. GenPRD produces a spec your agent can actually build from.

EARS acceptance criteria

Unambiguous, LLM-parseable "WHEN/IF-THEN… SHALL" statements — the format Kiro & Spec Kit converged on.

Requirement traceability

Every requirement gets a stable ID, and every task/test cites it. Drift becomes mechanically detectable.

Graded rigor

Tier 0 lean for CRUD, deep spec + invariants for money/auth. No oceans of markdown for a settings toggle.

Clean agent handoff

Emit tasks.md + spec that Claude Code and Cursor consume directly — the spec becomes the agent’s constraint, not a doc that rots.

Projects & versioning

Organize specs into projects, keep every version, compare changes. Git for your product intent.

Export anywhere

Markdown and PDF now; Claude Code skill / MCP handoff on the roadmap.

Simple pricing

Start free. Upgrade when specs become part of how you ship.

Free
$0

5 specs / week

  • ✓ Full spec engine (EARS + IDs)
  • ✓ Markdown export
  • ✓ 3 saved projects
Start free
Starter
$15/mo

30 spec credits / month

  • ✓ Everything in Free
  • ✓ 20 projects + versioning
  • ✓ Agent handoff exports
Pro
Most popular
$29/mo

Unlimited · fair-use 100 / week

  • ✓ Unlimited specs & projects
  • ✓ Premium model (Claude) + PDF export
  • ✓ Versioning, diff & agent handoff

Regional pricing available. Final pricing set at launch.

Why a spec beats a prompt

Raw promptGeneric PRD toolGenPRD
Testable acceptance criteria (EARS)
Stable requirement IDs (REQ / AC / NFR)
Traceability into tasks & tests
Explicit non-goals & NFRspartial
Rigor graded to risk (CRUD vs money/auth)
Agent-ready handoff (tasks.md, CLAUDE.md)
Structural quality validation

FAQ

How is this different from asking ChatGPT to write a PRD?+

A chat gives you prose. GenPRD enforces structure a coding agent can execute: EARS acceptance criteria, stable requirement IDs, explicit non-goals and NFRs — then validates the result and packages an agent handoff. The spec is checkable, not just readable.

Which coding agents does it work with?+

Any agent that reads Markdown: Claude Code, Cursor, Copilot, Windsurf, Cline, Aider and more. Export tasks.md or a CLAUDE.md / .cursorrules build guide and drop it into your repo.

What is "graded rigor"?+

Not every feature deserves the same depth. A CRUD screen gets a lean spec; anything touching money, auth or user data gets invariants, threat modeling and stricter acceptance criteria. You pick the tier per spec.

Do I need to be technical?+

No. Describe the idea in plain language; the discovery questions pull out the details an engineer would ask. The output is precise enough for an agent yet readable by anyone.

Can I use it for free?+

Yes — the free tier includes full spec generation with EARS + requirement IDs and Markdown export. Paid plans add premium models, unlimited projects, versioning and PDF export.

Ship what you meant to build.

Turn your next idea into a spec your AI agent builds right the first time.

Generate your first spec →