Skip to content

Wirasm/PRPs-agentic-eng

BranchesTags

Repository files navigation

PRP (Product Requirement Prompts)

A collection of prompts for AI-assisted development with Claude Code.

Video Walkthrough

https://www.youtube.com/watch?v=KVOZ9s1S9Gk&lc=UgzfwxvFjo6pKEyPo1R4AaABAg

Support This Work

Found value in these resources?

Buy me a coffee: https://coff.ee/wirasm

I spent a considerable amount of time creating these resources and prompts. If you find value in this project, please consider buying me a coffee to support my work.

Transform Your Team with AI Engineering Workshops

Ready to move beyond toy demos to production-ready AI systems?

Book a workshop: https://www.rasmuswiding.com/

What you'll get:

  • Put your team on a path to become AI power users
  • Learn the exact PRP methodology used by top engineering teams
  • Hands-on training with Claude Code, PRPs, and real codebases
  • From beginner to advanced AI engineering workshops for teams and individuals

Perfect for: Engineering teams, Product teams, and developers who want AI that actually works in production

Contact me directly at [email protected]

What is PRP?

Product Requirement Prompt (PRP) = PRD + curated codebase intelligence + agent/runbook

The minimum viable packet an AI needs to ship production-ready code on the first pass.

A PRP supplies an AI coding agent with everything it needs to deliver a vertical slice of working software—no more, no less.

How PRP Differs from Traditional PRD

A traditional PRD clarifies what the product must do and why customers need it, but deliberately avoids how it will be built.

A PRP keeps the goal and justification sections of a PRD yet adds AI-critical layers:

  • Context: Precise file paths, library versions, code snippet examples
  • Patterns: Existing codebase conventions to follow
  • Validation: Executable commands the AI can run to verify its work

Quick Start

Option 1: Copy Commands to Your Project

# From your project root
cp -r /path/to/PRPs-agentic-eng/.claude/commands/prp-core .claude/commands/

Option 2: Clone Repository

git clone https://github.com/Wirasm/PRPs-agentic-eng.git
cd PRPs-agentic-eng

Commands

The .claude/commands/prp-core/ directory contains the core PRP workflow commands:

Core Workflow

Command Description
/prp-prd Interactive PRD generator with implementation phases
/prp-plan Create implementation plan (from PRD or free-form input)
/prp-implement Execute a plan with validation loops

Issue & Debug Workflow

Command Description
/prp-issue-investigate Analyze GitHub issue, create implementation plan
/prp-issue-fix Execute fix from investigation artifact
/prp-debug Deep root cause analysis with 5 Whys methodology

Git & Review

Command Description
/prp-commit Smart commit with natural language file targeting
/prp-pr Create PR with template support
/prp-review Comprehensive PR code review

Autonomous Loop

Command Description
/prp-ralph Start autonomous loop until all validations pass
/prp-ralph-cancel Cancel active Ralph loop

Ralph Loop (Autonomous Execution)

Based on Geoffrey Huntley's Ralph Wiggum technique - a self-referential loop that keeps iterating until the job is actually done.

How It Works

/prp-ralph .claude/PRPs/plans/my-feature.plan.md --max-iterations 20
  1. Claude implements the plan tasks
  2. Runs all validation commands (type-check, lint, tests, build)
  3. If any validation fails → fixes and re-validates
  4. Loop continues until ALL validations pass
  5. Outputs <promise>COMPLETE</promise> and exits

Each iteration, Claude sees its previous work in files and git history. It's not starting fresh - it's debugging itself.

Setup

The stop hook must be configured in .claude/settings.local.json:

{
  "hooks": {
    "Stop": [
      {
        "hooks": [
          {
            "type": "command",
            "command": ".claude/hooks/prp-ralph-stop.sh"
          }
        ]
      }
    ]
  }
}

Usage

# Create a plan
/prp-plan "add user authentication with JWT"

# Let Ralph loose
/prp-ralph .claude/PRPs/plans/add-user-auth.plan.md --max-iterations 20

# Cancel if needed
/prp-ralph-cancel

Tips

  • Always use --max-iterations (default: 20) to prevent infinite loops
  • Works best with plans that have clear, testable validation commands
  • State is tracked in .claude/prp-ralph.state.md
  • Progress and learnings are captured in the implementation report

Workflow Overview

Large Features: PRD → Plan → Implement

/prp-prd "user authentication system"
    ↓
Creates PRD with Implementation Phases table
    ↓
/prp-plan .claude/PRPs/prds/user-auth.prd.md
    ↓
Auto-selects next pending phase, creates plan
    ↓
/prp-implement .claude/PRPs/plans/user-auth-phase-1.plan.md
    ↓
Executes plan, updates PRD progress, archives plan
    ↓
Repeat /prp-plan for next phase

Medium Features: Direct to Plan

/prp-plan "add pagination to the API"
    ↓
Creates implementation plan from description
    ↓
/prp-implement .claude/PRPs/plans/add-pagination.plan.md

Bug Fixes: Issue Workflow

/prp-issue-investigate 123
    ↓
Analyzes issue, creates investigation artifact
    ↓
/prp-issue-fix 123
    ↓
Implements fix, creates PR

Artifacts Structure

All artifacts are stored in .claude/PRPs/:

.claude/PRPs/
├── prds/              # Product requirement documents
├── plans/             # Implementation plans
│   └── completed/     # Archived completed plans
├── reports/           # Implementation reports
├── issues/            # Issue investigation artifacts
│   └── completed/     # Archived completed investigations
└── reviews/           # PR review reports

PRD Phases

PRDs include an Implementation Phases table for tracking progress:

| #   | Phase | Description | Status      | Parallel | Depends | PRP Plan |
| --- | ----- | ----------- | ----------- | -------- | ------- | -------- |
| 1   | Auth  | User login  | complete    | -        | -       | [link]   |
| 2   | API   | Endpoints   | in-progress | -        | 1       | [link]   |
| 3   | UI    | Frontend    | pending     | with 4   | 2       | -        |
| 4   | Tests | Test suite  | pending     | with 3   | 2       | -        |
  • Status: pending → in-progress → complete
  • Parallel: Phases that can run concurrently (in separate worktrees)
  • Depends: Phases that must complete first

PRP Best Practices

  1. Context is King: Include ALL necessary documentation, examples, and caveats
  2. Validation Loops: Provide executable tests/lints the AI can run and fix
  3. Information Dense: Use keywords and patterns from the codebase
  4. Bounded Scope: Each plan should be completable by an AI in one loop

Project Structure

your-project/
├── .claude/
│   ├── commands/prp-core/   # PRP commands
│   ├── PRPs/                # Generated artifacts
│   └── agents/              # Custom subagents
├── PRPs/
│   ├── templates/           # PRP templates
│   └── ai_docs/             # Library documentation
├── CLAUDE.md                # Project-specific guidelines
└── src/                     # Your source code

Parallel Development with Worktrees

When PRD phases can run in parallel:

# Phase 3 and 4 can run concurrently
git worktree add -b phase-3-ui ../project-phase-3
git worktree add -b phase-4-tests ../project-phase-4

# Run Claude in each
cd ../project-phase-3 && claude
cd ../project-phase-4 && claude

Resources

Templates (PRPs/templates/)

  • prp_base.md - Comprehensive PRP template
  • prp_story_task.md - Story/task template
  • prp_planning.md - Planning template

AI Documentation (PRPs/ai_docs/)

Curated documentation for Claude Code context injection.

Legacy Commands

Previous command versions are preserved in old-prp-commands/ for reference.

License

MIT License

Support

I spent a considerable amount of time creating these resources and prompts. If you find value in this project, please consider buying me a coffee to support my work.

Buy me a coffee: https://coff.ee/wirasm

The goal is one-pass implementation success through comprehensive context.

About

Prompts, workflows and more for agentic engineering

Resources

Readme
Activity

Stars

1.9k stars

Watchers

63 watching

Forks

568 forks
Report repository

Releases

No releases published

Packages

No packages published

Contributors 4

Languages