PRD Skill for Claude Code

A Claude Code custom command that generates structured, AI-prototyping-ready Product Requirements Documents. Drop it into any project and run /prd to start.

---

What it does

This skill turns a product idea into a complete PRD optimized for two audiences simultaneously:

1. Your cross-functional team — PMs, engineers, and designers get the context they need to ask good questions and make confident decisions
2. AI prototyping tools — Cursor, v0, Lovable, bolt.new, and Replit get structured sections (component inventory, data models, API specs, file structure) they can use as a build brief without additional clarification

The skill guides you through a conversational discovery process, then generates a 15-section PRD structured around the Jobs to Be Done (JTBD) framework and atomic User Stories.

---

The approach: JTBD + User Stories

Most PRDs describe what to build. This skill starts with why people need it.

Jobs to Be Done (JTBD)

JTBD frames each user need as a situational job:

"When [situation], I want to [motivation], so I can [expected outcome]."

This matters for AI tools because it prevents technically-correct-but-wrong solutions. When Cursor or Lovable knows why a feature exists, it makes better architectural choices — not just syntactically valid ones.

User Stories

User Stories translate JTBD jobs into atomic, implementable features:

"As a [role], I want to [action], so that [benefit]."

Each story gets a unique ID (US1, US2…) that cross-references:

• The JTBD job it serves
• The acceptance criteria that verify it's done

How they work together

JTBD operates at the level of human motivation. User Stories operate at the level of product features. Together, they give your team and your AI tools a clear line from "why this exists" → "what to build" → "how to verify it's correct."

---

Why this PRD format works with AI prototyping tools

Each structural section serves a specific purpose for AI tools:

Section	What AI tools use it for
AI Build Summary	The tl;dr brief — the first thing Cursor or Replit reads to understand the build target
Component Inventory	v0, Lovable, and bolt.new generate UI at the component level — this gives them the full list
Data Models	TypeScript interfaces AI tools use to generate type-safe code without guessing schema
API / Integration Surface	Endpoint specs AI tools use to generate API clients and server routes
State Management Map	Tells AI tools what lives in local vs. server state vs. URL params — prevents architectural mistakes
Tech Stack Recommendation	Locks in the stack so AI tools don't mix incompatible libraries
Suggested File Structure	An ASCII tree AI tools use as a scaffolding target
Acceptance Criteria	Binary pass/fail checklists AI agents can verify against generated code

---

Installation

Option 1: Global (available in all projects)

mkdir -p ~/.claude/commands
cp SKILL.md ~/.claude/commands/prd.md

Option 2: Project-local (this project only)

mkdir -p .claude/commands
cp SKILL.md .claude/commands/prd.md

That's it. Claude Code will detect the file and register /prd as a custom command.

---

Usage

/prd

Starts the conversational discovery flow. The skill asks about the problem, users, tech stack, and constraints before generating the full PRD.

/prd "checkout flow redesign"

Pre-populates the PRD title and jumps straight to clarifying questions. Use this when you already know the feature name.

---

Example output

See examples/example-prd.md for a fully filled-out PRD demonstrating every section.

---

Customization tips

For simple features: The skill omits sections that don't apply — just answer "not applicable" when asked, and those sections won't appear in the output.

For complex multi-team efforts: Add a "Dependencies" section after Section 3 to map inter-team handoffs, and expand Section 14 (Open Questions & Risks) with specific team owners.

For design-heavy features: Expand Section 6 (Proposed Experience) with more detail — annotated screen lists, interaction flows, and specific accessibility requirements.

For API-first features: Expand Sections 8 and 9 (Data Models and API Surface) with request/response examples and error codes.

Adjusting the tech stack defaults: Edit the prompt in SKILL.md to change the default stack recommendation the skill suggests when users have no preference.

---

How to update the skill

When you want to modify the PRD template:

1. Edit SKILL.md in this repo
2. Re-copy it to wherever you installed it: cp SKILL.md ~/.claude/commands/prd.md

---

Compatible tools

Tested and optimized for output that works well with:

• Cursor
• v0 by Vercel
• Lovable
• bolt.new
• Replit
• Any Claude Code instance