PRP Manager Skill

Purpose

Create comprehensive PRPs (Product Requirements Prompts) that enable AI agents to implement features with sufficient context and validation loops for one-pass implementation success.

When to Use

• Initialize: Setting up PRPs for a new project (creates custom template)
• Generate: Planning a new feature or enhancement
• Execute: Implementing an existing PRP file

Core Principles

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. Progressive Success: Start simple, validate, then enhance
5. One-Pass Success: Goal is working code through comprehensive context

References

• Default Template: See assets/templates/prp_base.md
• Usage Examples: See references/examples.md
• Customization Guide: See references/customization.md
• Project Template: PRPs/templates/prp_base.md (generated per project)

---

Workflow 1: Initialize PRPs (RECOMMENDED FIRST)

Run this first in any new project to create a customized PRP template.

When to Suggest

• User asks to create a PRP but PRPs/templates/prp_base.md doesn't exist
• User explicitly asks to initialize or setup PRPs
• First time using PRPs in a project

Step 1: Analyze Project

Read these files (in order of priority):

AI Agent Configuration:
  - AGENTS.md                              # Universal AI agent guidance (agents.md standard)
  - CLAUDE.md or .claude/settings.json    # Claude-specific rules
  - GEMINI.md                              # Gemini-specific rules

Project Documentation:
  - README.md                              # Project overview
  - CONTRIBUTING.md                        # Contribution guidelines

Package/Dependencies:
  - package.json                           # Node.js projects
  - pyproject.toml or requirements.txt     # Python projects
  - Cargo.toml                             # Rust projects
  - go.mod                                 # Go projects

Code Quality:
  - .eslintrc* / biome.json               # JS/TS linting
  - ruff.toml / pyproject.toml [ruff]     # Python linting
  - mypy.ini / pyproject.toml [mypy]      # Python type checking
  - tsconfig.json                          # TypeScript config

Testing:
  - jest.config.* / vitest.config.*       # JS/TS testing
  - pytest.ini / pyproject.toml [pytest]  # Python testing
  - tests/ or __tests__/ structure        # Test patterns

Step 2: Detect Stack & Conventions

Extract from analysis:

• Language/Framework (Python/FastAPI, Node/Express, etc.)
• Package manager (uv, npm, pnpm, yarn, cargo)
• Linting tools and commands
• Type checking tools and commands
• Testing framework and commands
• Project structure conventions
• Any special rules from AGENTS.md/CLAUDE.md/GEMINI.md

Step 3: Generate Custom Template

Create directory and template:

mkdir -p PRPs/templates

Generate PRPs/templates/prp_base.md with:

• Project-specific validation commands
• Correct package manager syntax
• Actual linting/testing tools used
• Codebase tree format matching project structure
• Any gotchas from AGENTS.md/CLAUDE.md/GEMINI.md
• Test patterns from existing tests

Step 4: Confirm Setup

Output created:

• PRPs/templates/prp_base.md - Customized for this project

Summary to user:

• Stack detected
• Validation commands configured
• Any special rules applied
• Ready to generate PRPs

---

Workflow 2: Generate PRP

Pre-check

If PRPs/templates/prp_base.md doesn't exist:

"I notice this project doesn't have a PRP template yet. Would you like me to initialize PRPs first? This will create a customized template based on your project's stack and conventions."

Step 1: Understand the Request

• What is the feature/enhancement?
• What is the expected end state?
• Any specific patterns to follow?

Step 2: Research Phase

Codebase Analysis:

• Search for similar features/patterns
• Identify files to reference
• Note existing conventions
• Check test patterns

External Research (if needed):

• Library documentation (include URLs)
• Implementation examples
• Best practices and pitfalls

Step 3: Generate the PRP

Check existing PRPs:

ls -1 PRPs/*.md 2>/dev/null | grep -E '^PRPs/[0-9]{3}--' | sort -r | head -5

Naming: PRPs/{NNN}--{feature-name}.md

• 3-digit padding (001, 002, ...)
• kebab-case for feature name

Use template from: PRPs/templates/prp_base.md

Step 4: Quality Check

• All context included for one-pass implementation
• Validation gates are executable
• References existing patterns
• Clear implementation path
• Gotchas documented

Score (1-10): Confidence for one-pass success

---

Workflow 3: Execute PRP

Step 1: Load and Understand

• Read PRP completely
• Understand all context
• Extend research if gaps found

Step 2: Plan

• Think before executing
• Break into manageable steps
• Use task tracking if available
• Identify patterns from existing code

Step 3: Execute

• Follow implementation blueprint
• Implement in task order
• Mark tasks as completed

Step 4: Validate

• Run each validation command
• Fix failures
• Re-run until all pass

Step 5: Complete

• All checklist items done
• Final validation suite passed
• Re-read PRP to verify
• Report completion status

---

Best Practices

DO:

• Initialize PRPs first for new projects
• Include comprehensive context for AI agents
• Reference real files and patterns
• Provide executable validation commands
• Document known gotchas
• List tasks in execution order

DON'T:

• Skip initialization (generic template is less effective)
• Create new patterns when existing ones work
• Skip validation
• Ignore failing tests
• Hardcode values that should be config

---

Output Summary

For Initialize

✅ PRPs Initialized for [Project Name]

Stack Detected:
- Language: Python 3.11
- Framework: FastAPI
- Package Manager: uv
- Linting: ruff
- Type Checking: mypy
- Testing: pytest

Created:
- PRPs/templates/prp_base.md

Ready to generate PRPs!

For Generate

📄 PRPs/005--feature-name.md

Confidence: 8/10
Key Notes: [implementation highlights]

For Execute

✅ PRP Execution Complete

Tasks: 5/5 completed
Validation: All passing
Status: SUCCESS