RPI Skills

A set of Claude Code skills implementing the Research-Plan-Implement (RPI) workflow for AI-assisted software development with human-in-the-loop checkpoints.

Skills

Skill	Command	Gate	Description
Researching Feature	/rpi-researching-feature	Human	Structured research producing research.md
Planning Implementation	/rpi-planning-implementation	Human	Implementation plan producing plan.md
Implementing Plan	/rpi-implementing-plan	Human	Executes the plan, writes code
Reviewing Gated	/rpi-reviewing-gated	Auto	Self-review with severity-rated pass/revise verdict
Documenting Decision	/rpi-documenting-decision	Auto	ADR-style decision records
Logging Implementation	/rpi-logging-implementation	Auto	Post-implementation devlog
Running RPI	/rpi-running-autonomous	None	Autonomous orchestrator, runs the full chain unattended

Installation

Per-project

Copy the skills/ folder into your project's .claude/ directory:

# All skills
cp -r /path/to/rpi-skills/skills/ /path/to/your-project/.claude/skills/

# Or individual skills
cp -r /path/to/rpi-skills/skills/rpi-researching-feature /path/to/your-project/.claude/skills/

Global (all projects)

Copy to your home directory to make skills available to every Claude Code session:

cp -r /path/to/rpi-skills/skills/ ~/.claude/skills/

Usage

Interactive (human-in-the-loop)

Run skills individually. The three major phases ask for confirmation before starting:

/rpi-researching-feature Add user authentication with OAuth2
/rpi-planning-implementation
/rpi-implementing-plan

Decisions can be recorded during research or planning, and implementation automatically triggers the devlog and gated review.

Autonomous

Run the full RPI cycle without interaction:

/rpi-running-autonomous Add user authentication with OAuth2 supporting Google and GitHub providers

Workflow

/rpi-researching-feature ──→ human approval
                                │
                                ▼
                              /rpi-planning-implementation ──→ human approval
                                                                  │
                                                                  ▼
                                                              /rpi-implementing-plan ──→ /rpi-logging-implementation
                                                                                          │
                                                                                          ▼
                                                                                    /rpi-reviewing-gated

/rpi-documenting-decision can be invoked at any point during research or planning when multiple valid approaches exist.

Artifacts

All artifacts are written to the project root:

File	Produced by
research.md	/rpi-researching-feature
plan.md	/rpi-planning-implementation
decision-*.md	/rpi-documenting-decision
devlog-NNN.md	/rpi-logging-implementation
code-review-NNN.md	/rpi-reviewing-gated

Dependencies between skills

Skills are designed to work independently but produce better results together:

• /rpi-planning-implementation checks for research.md (soft dependency)
• /rpi-implementing-plan checks for plan.md (effectively required)
• /rpi-reviewing-gated is auto-invoked after implementation and can also be run manually against research, planning, or code artifacts
• /rpi-documenting-decision reads research.md and plan.md for context if available
• /rpi-logging-implementation reads plan.md and recent git changes