prd-creator — Claude Code skill: PRDs and task lists
This skill turns a rough app idea into a PRD with verifiable pass criteria, then breaks it into JSON tasks. Useful when a feature needs writing down before code starts.
npx skills add PageAI-Pro/ralph-loop --skill prd-creator -a claude-codeWhat prd-creator does
The skill runs in two stages. First the agent asks clarifying questions through AskUserQuestion to pull out requirements you have not spelled out yet. Then it builds PRD.md: app overview, target audience, success metrics, competitor research via WebSearch, features and user flows, stack recommendations, security notes, assumptions and dependencies. Along the way it checks prerequisites — access, MCP servers, docs, environment variables and test users — and creates or updates .env.local with placeholders only.
The second stage produces tasks.json. Every task gets a type (functional, ui-ux, api-endpoint, security and others) plus verification steps. TASK-1 always comes first and verifies prerequisites before feature work. Tasks stay small: the source asks for a maximum of 10 minutes each, and anything bigger gets split. All tasks are generated with "passes": false.
The skill will not write real secrets into .env.local — placeholders only, you fill in the values yourself. It also does not generate code. Output lands in PROJECT_ROOT/.agent/prd/PRD.md, PROJECT_ROOT/.agent/prd/SUMMARY.md and PROJECT_ROOT/.agent/tasks.json. If those files are missing, it warns you and offers to create them.
What it helps with
- document an app idea before any code is written
- get a task list with verification steps for each item
- check access and environment variables before starting work
- split a large feature into 10-minute tasks
- produce a short project summary in SUMMARY.md
Example prompts for the agent
- Create a PRD for an expense tracking app
- Break the finished PRD into tasks and save them to tasks.json
- Update the PRD: we are adding Google sign-in
How to install prd-creator
1Skills CLI (Vercel)
npx skills add PageAI-Pro/ralph-loop --skill prd-creator -a claude-code2Manually: put the files in place
~/.claude/skills/prd-creator/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/prd-creator/
1Skills CLI (Vercel)
npx skills add PageAI-Pro/ralph-loop --skill prd-creator -a codex2Manually: put the files in place
.agents/skills/prd-creator/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add PageAI-Pro/ralph-loop --skill prd-creator -a cursor2Manually: put the files in place
.cursor/skills/prd-creator/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install PageAI-Pro/ralph-loop prd-creator2Manually: put the files in place
.github/skills/prd-creator/SKILL.md1Skills CLI (Vercel)
npx skills add PageAI-Pro/ralph-loop --skill prd-creator -a gemini-cli2Platform CLI
gemini skills install https://github.com/PageAI-Pro/ralph-loop --path .agent/skills/prd-creatorManually: put the files in place
.devin/skills/prd-creator/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add PageAI-Pro/ralph-loop --skill prd-creator -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about prd-creator
How do I install prd-creator in Claude?
Run npx skills add PageAI-Pro/ralph-loop --skill prd-creator -a claude-code. The skill lands at ~/.claude/skills/prd-creator/SKILL.md.
What does prd-creator do?
It runs a structured interview about your idea, verifies project prerequisites and writes PRD.md with metrics, competitive analysis and stack recommendations. Then it generates tasks.json with tasks and pass steps.
Does the skill write real secrets into .env.local?
No. It creates or updates .env.local with placeholders only and tells you to fill in real values by hand. It does not generate code either.
Source text
SKILL.md · MIT
---
name: prd-creator
description: Guides creation of comprehensive Product Requirement Documents (PRDs) for software projects through structured questioning and validation, then generates implementation task lists in JSON format. Use when users want to document a software idea, create specifications for development, plan a new application feature/bug, or break down requirements into actionable tasks. Transforms ideas into implementation-ready documents with verifiable pass criteria.
license: MIT
metadata:
author: pageai
version: '1.0.1'
tags: prd, product requirements, software development, documentation, task generation
website: https://pageai.pro/blog/long-running-ai-coding-agents-ralph-loop#step-2-write-your-requirements
---
# PRD Creation Assistant
Transform software ideas into comprehensive PRDs and actionable implementation tasks through a two-part process.
## Overview
This skill helps beginner-level developers.
1. Receive an implementation description from the user
2. Create detailed PRD documents through structured questioning
3. Verify implementation prerequisites, including access, MCPs, docs, env variables, and test users
4. Generate implementation task lists in JSON format for developers
5. Write an overall description of the project. An executive summary that gives a high level overview of the app and its main features.
### Part 1: Implementation Description
You will receive a lacking implementation description from the user.
The main goal is to comprehend the intent and think about the larger architecture and a robust way to implement it, filling in the gaps.
### Part 2: PRD Creation
**File**: [PRD.md](PRD.md)
You will need to ask clarifying questions to get a clear understanding of the implementation.
**When to use**: User wants to document a software idea or create feature specifications
**What it does**:
- Guides structured questioning to gather all requirements
- Verifies project prerequisites before PRD finalization
- Creates/updates `.env.local` with placeholder values only
- Creates executive summary for validation
- Researches competitive landscape
- Generates comprehensive PRD.md with:
- App overview and objectives
- Target audience
- Success metrics and KPIs
- Competitive analysis
- Core features and user flows
- Technical stack recommendations
- Prerequisites and access
- Security considerations
- Assumptions and dependencies
**Process**:
1. Ask clarifying questions using `AskUserQuestion` tool
2. Verify prerequisites and create/update `.env.local` placeholders
3. Create executive summary for user approval
4. Research competition via WebSearch
5. Generate complete PRD
6. Iterate based on feedback
**Read [PRD.md](PRD.md) for complete instructions.**
---
### Part 3: Implementation Task Generation
**File**: [JSON.md](JSON.md)
You will need to analyze the completed PRD and generate a comprehensive task list in JSON format.
**When to use**: After PRD is complete and approved, or user requests task breakdown
**What it does**:
- Analyzes the completed PRD
- Generates `TASK-1` as mandatory prerequisite verification
- Generates a complete list of implementation tasks in JSON format, covering all features and requirements from the PRD
- Keeps the tasks small and manageable
- Categorizes tasks by type (functional, ui-ux, api-endpoint, security, etc.)
- Defines verification ('pass') steps for each task
- Creates developer-ready checklist
**IMPORTANT**:
- Each task should be simple enough to be completed in maximum 10 minutes.
- If a task is too complex, it should be split into smaller tasks.
**Read [JSON.md](JSON.md) for complete instructions.**
## Part 4: Overall Description
You will need to read the completed PRD and generate an overall description of the project in `PROJECT_ROOT/.agent/prd/SUMMARY.md`.
The description should be short, concise and contain:
- An overall description of the project
- The main features of the app
- Key user flows
- A short list of key requirements
## Quick Start
**If user wants to create a PRD:**
1. Read [PRD.md](PRD.md)
2. Follow the PRD creation workflow
3. Verify prerequisites and create/update `.env.local` with placeholder values only
4. If needed, update the overall description [SUMMARY.md](SUMMARY.md)
5. After PRD completion, ask: "Would you like me to generate implementation tasks? See Part 2."
**If user wants implementation tasks for an existing PRD:**
1. Read [JSON.md](JSON.md)
2. Read the PRD file
3. Generate comprehensive task list in JSON format, starting with `TASK-1` prerequisite verification
4. Save as `tasks.json`
**If user wants both:**
1. Complete PRD creation first [PRD.md](PRD.md), including prerequisite verification and `.env.local` placeholders
2. Get user approval on PRD
3. If needed, update the overall description [SUMMARY.md](SUMMARY.md)
4. Proceed to generate implementation tasks [JSON.md](JSON.md)
**If a user want to update the PRD:**
1. Read [PRD.md](PRD.md)
2. Update the PRD
3. Save as `PRD.md`
4. If needed, update the overall description [SUMMARY.md](SUMMARY.md)
5. Ask user if they want to generate implementation tasks
**If a user want to update the implementation tasks:**
1. Read [JSON.md](JSON.md)
2. Update the implementation tasks
3. Save as `tasks.json`
4. Ask user if they want to update the PRD again
**If user wants to update both the PRD and the implementation tasks:**
1. Update the PRD first [PRD.md](PRD.md)
2. If needed, update the overall description [SUMMARY.md](SUMMARY.md)
3. Update the implementation tasks [JSON.md](JSON.md)
4. Save as `PRD.md` and `tasks.json`
## After completion
Ensure the required files are present:
- PROJECT_ROOT/.agent/prd/PRD.md
- PROJECT_ROOT/.agent/prd/SUMMARY.md
- PROJECT_ROOT/.agent/tasks.json
If they are not present, warn the user and ask if they would like to create any of them.
## Important Constraints
- Do not generate code - focus on documentation and task specification
- Use AskUserQuestion extensively in Part 1 to clarify requirements
- Never write real secret values to PRD, tasks, chat, logs, or `.env.local`; use placeholder values and tell the user to fill real values manually
- In Part 2, generate comprehensive task lists (50-200+ tasks for typical projects)
- In Part 2, always generate `TASK-1` as prerequisite verification before feature work
- Always initialize tasks with `"passes": false` - never mark tasks complete during generation
- Use available tools: AskUserQuestion, WebSearch, Sequential Thinking, Read
© pageai · MIT · PageAI-Pro/ralph-loop