/create-prd

Create PRD Command

Interactive wizard to create a Product Requirements Document and set up your project for Ralph.

Usage

/create-prd                           # Start interactive wizard
/create-prd "Build a todo app"        # Start with description

What It Does

1. Asks for task set name (default: "initial")
2. Asks discovery questions about your project
3. Generates task list in .ralph/tasksets/{name}/tasks.json
4. Creates PRD in .ralph/tasksets/{name}/prd.md
5. Creates memories file in .ralph/tasksets/{name}/memories.md
6. Creates PROMPT.md with iteration instructions
7. Copies ralph.sh loop runner script
8. Creates symlink .ralph/current-taskset

Discovery Questions

Ask 10 discovery questions (task set name, problem, audience, features, tech stack, architecture, UI/UX, auth, integrations, success criteria). See the ralph-prd skill's WORKFLOW.md for question details and good answer examples.

Output Files

See the ralph-prd skill's SKILL.md "What It Creates" section for the full directory structure. Use templates/ for generating each file.

Implementation

When the user runs /create-prd:

Step 1: Ask for task set name

## Task Set Name

What would you like to name this task set?

Examples: "initial", "auth-feature", "test-coverage"

(Default: "initial" if you press Enter)

Validate the name:

• Lowercase only
• Alphanumeric and hyphens only
• No spaces

Step 2: Check for existing setup

TASKSET_NAME="${1:-initial}"

if [[ -d ".ralph/tasksets/${TASKSET_NAME}" ]]; then
  echo "Task set '${TASKSET_NAME}' already exists."
  echo "Run /ralph taskset switch ${TASKSET_NAME} to use it."
  echo "Or choose a different name."
  exit 0
fi

Step 3: Run discovery questions

Use AskUserQuestion tool for each discovery step:

## Discovery: Problem Definition

What problem are you solving with this project?

(Describe the pain point you're addressing)

Step 4: Generate tasks from features

Convert each feature into atomic tasks:

• Each task completable in ONE iteration
• Clear steps provided
• Verification criteria included

Step 5: Create files

1. Create directory structure:

   mkdir -p .ralph/tasksets/${TASKSET_NAME} screenshots
   

2. Generate .ralph/tasksets/${TASKSET_NAME}/tasks.json from discovery answers

3. Generate .ralph/tasksets/${TASKSET_NAME}/prd.md from discovery answers

4. Generate .ralph/tasksets/${TASKSET_NAME}/memories.md (empty template)

5. Generate PROMPT.md with tech-stack-specific commands (uses symlink paths)

6. Copy ralph.sh from plugin and make executable

7. Generate .ralph/tasksets/${TASKSET_NAME}/config.json with settings

8. Create symlink:

   ln -sfn tasksets/${TASKSET_NAME} .ralph/current-taskset
   

9. Create empty activity log:

   touch .ralph/tasksets/${TASKSET_NAME}/activity.log
   

Step 6: Provide next steps

## Setup Complete!

Task set: ${TASKSET_NAME}

Files created:
- .ralph/tasksets/${TASKSET_NAME}/tasks.json (X tasks)
- .ralph/tasksets/${TASKSET_NAME}/prd.md
- .ralph/tasksets/${TASKSET_NAME}/memories.md
- .ralph/tasksets/${TASKSET_NAME}/config.json
- .ralph/current-taskset (symlink)
- PROMPT.md
- ralph.sh

Next steps:
1. Review the generated tasks: cat .ralph/current-taskset/tasks.json
2. Start the loop: ./ralph.sh 20
3. Or run fewer iterations: ./ralph.sh 5

The loop will work through tasks until complete or max iterations reached.

To create another task set later:
/ralph taskset new "feature-name"

Quick Mode

If a description is provided, skip some questions:

/create-prd "Build a React todo app with TypeScript"

In this case:

• Infer tech stack from description
• Ask only clarifying questions
• Generate minimal viable task list
• Use default task set name "initial"

Task Categories

Categories: setup, feature, integration, styling, testing. See WORKFLOW.md for details.

Tech Stack Defaults

If not specified, default to:

Component	Default
Framework	React
Language	TypeScript
Build Tool	Vite
Styling	Tailwind CSS
Testing	Vitest

Notes

• Tasks should be atomic (completable in one iteration)
• Order tasks by dependency (setup first, then features)
• Keep scope tight for v1 (defer nice-to-haves)
• Success criteria should be testable
• Task set names should be descriptive and kebab-case