rpi

RPI Workflow (Research - Plan - Implement)

← Back to Claude Code Best Practice

---

A systematic development workflow with validation gates at each phase. You describe a feature, a team of specialized agents researches feasibility, plans the implementation, and builds it -- with GO/NO-GO checkpoints preventing wasted effort.

See the full workflow with diagrams: rpi-workflow.md

---

Quick Start

1. Install: copy the .claude folder (containing agents/ and commands/rpi/) into your repository root.
2. Create the plans directory: mkdir -p rpi/plans
3. Describe your feature to Claude Code. It generates rpi/plans/{feature-slug}.md.
4. Create the feature folder and move the plan:

   mkdir -p rpi/{feature-slug}
   cp rpi/plans/{feature-slug}.md rpi/{feature-slug}/REQUEST.md

5. Research: /rpi:research rpi/{feature-slug}/REQUEST.md
6. Plan: /rpi:plan {feature-slug}
7. Implement: /rpi:implement {feature-slug}

---

Agent Roster

Eight specialized agents collaborate across the three phases.

Agent	Role	Phases
requirement-parser	Extracts structured requirements from the feature request	Research
product-manager	Defines user stories, acceptance criteria, and GO/NO-GO verdict	Research, Plan
senior-software-engineer	Evaluates technical feasibility, writes architecture specs, implements code	Research, Plan, Implement
technical-cto-advisor	Reviews strategic alignment and technical risk	Research
ux-designer	Designs UI flows and interaction patterns	Plan
documentation-analyst-writer	Produces research reports, plan documents, and implementation records	Research, Plan
code-reviewer	Reviews implementation for quality, correctness, and standards compliance	Implement
constitutional-validator	Validates agent outputs against project constraints and rules	All

---

Phase Overview

Phase 1: Research (/rpi:research)

Agents analyze the feature request against the existing codebase. Output: research/RESEARCH.md with a GO or NO-GO verdict.

• Parses requirements into structured format
• Evaluates technical feasibility
• Checks strategic alignment
• Identifies risks and dependencies

Phase 2: Plan (/rpi:plan)

Agents produce a complete implementation roadmap. Output:

File	Contents
plan/pm.md	User stories and acceptance criteria
plan/ux.md	UI/UX flows and wireframes
plan/eng.md	Technical architecture and specifications
plan/PLAN.md	Phased task breakdown with test gates

Phase 3: Implement (/rpi:implement)

Agents build the feature phase-by-phase, running test gates after each phase. Output: implement/IMPLEMENT.md with phase-by-phase results.

---

When to Use This Workflow

Scenario	Why RPI helps
New features with unclear scope	Research phase forces clarity before any code is written
Cross-cutting changes (auth, database schema, API contracts)	Multi-agent planning catches integration issues early
Features requiring product and engineering alignment	PM, UX, and engineering agents produce aligned artifacts
Onboarding new team members	The artifact trail (REQUEST, RESEARCH, PLAN, IMPLEMENT) serves as living documentation
High-risk changes where rollback is expensive	GO/NO-GO gate prevents wasted effort on non-viable features

---

Feature Folder Structure

rpi/{feature-slug}/
├── REQUEST.md              # Initial feature description
├── research/
│   └── RESEARCH.md         # GO/NO-GO analysis
├── plan/
│   ├── PLAN.md             # Implementation roadmap
│   ├── pm.md               # Product requirements
│   ├── ux.md               # UX design
│   └── eng.md              # Technical specification
└── implement/
    └── IMPLEMENT.md        # Implementation record

---

Contents

• rpi-workflow.md -- Full workflow diagram and example walkthrough
• .claude/agents/ -- Agent definitions (8 agents)
• .claude/commands/rpi/ -- Slash commands (research, plan, implement)

Last Updated: 2026-04-09