rpikit

A plugin implementing the Research-Plan-Implement (RPI) framework for disciplined software engineering.

Philosophy

Understand before acting. Plan before coding. Implement with discipline.

This plugin enforces a structured workflow that prevents premature implementation and ensures human oversight at critical decision points.

Installation

Step 1: Add the marketplace

/plugin marketplace add bostonaholic/rpikit

Step 2: Install the plugin

/plugin install rpikit

Quick Start

Slash Command Purpose
/rpikit:research-plan-implement End-to-end research, plan, and implement pipeline
/rpikit:brainstorming Explore ideas when requirements are unclear
/rpikit:researching-codebase Understand the codebase and gather context
/rpikit:writing-plans Create an actionable implementation plan
/rpikit:implementing-plans Execute the plan with discipline
/rpikit:reviewing-code Review changes for quality and maintainability
/rpikit:security-review Review changes for security vulnerabilities
/rpikit:documenting-decisions Record architectural decisions as ADRs

Workflow

Full Pipeline

flowchart LR
    rpi["/rpikit:research-plan-implement"] -->|automated| research[Research]
    research -->|approval| plan[Plan]
    plan -->|approval| implement[Implement]
    implement --> done((done))

The /rpikit:research-plan-implement skill runs the full pipeline in a single session using parallel subagents, with approval gates between phases. Implementation runs in an isolated worktree to protect the main branch.

Individual Skills

flowchart LR
    brainstorm["/rpikit:brainstorming"] -.->|optional| research["/rpikit:researching-codebase"]
    research -->|approval| plan["/rpikit:writing-plans"]
    plan --> implement["/rpikit:implementing-plans"]
    plan -.->|optional| decision["/rpikit:documenting-decisions"]
    implement -->|approval| done((done))

For more control, run each phase as a separate skill. Each phase produces artifacts in docs/plans/ and requires human approval before transitioning to the next phase.

Brainstorming vs. Research

Both commands start by asking clarifying questions before acting. The key difference is their purpose:

Brainstorming Research
What should we build? How does it work?
Explores design approaches Explores existing code
Vague idea → clear design Clear topic → codebase understanding

Use Brainstorming when:

Use Research when:

Common flow: Brainstorm first (if unclear) → Research → Plan → Implement

Output Structure

docs/plans/
├── YYYY-MM-DD-<topic>-research.md   # Research findings with file:line references
└── YYYY-MM-DD-<topic>-plan.md       # Implementation plan with tasks and criteria

docs/decisions/
└── NNNN-decision-title.md           # Architecture Decision Records

Usage Examples

Run the entire research-plan-implement workflow in a single session:

/rpikit:research-plan-implement Add OAuth login with Google and GitHub providers

This spawns parallel research subagents, synthesizes findings, creates a plan, and implements it — with approval gates between each phase.

Step-by-Step Workflow

For more control, run each phase separately:

/rpikit:researching-codebase I want to add OAuth login - what auth patterns exist?

Review the research output in docs/plans/, then create a plan from it:

/rpikit:writing-plans @docs/plans/2025-01-07-oauth-login-research.md

Review and approve the plan, then implement from it:

/rpikit:implementing-plans @docs/plans/2025-01-07-oauth-login-plan.md

Ad-hoc Code Review

Review current changes for quality issues:

/rpikit:reviewing-code

Review for security vulnerabilities:

/rpikit:security-review

Recording Decisions

After planning or design work, record the decision as an ADR:

/rpikit:documenting-decisions @docs/plans/2025-01-07-oauth-login-design.md

Stakes-Based Planning

The framework adapts to change complexity:

Skills

The plugin includes methodology skills that guide disciplined development:

Core RPI Workflow

Development Discipline

Workflow Support

Advanced Patterns

Architecture

For a detailed overview of how skills, agents, and hooks connect, see Architecture.

Inspired By

License

MIT