rpi
Use when implementing features from Jira tickets, PRDs, or user requirements. Orchestrates Research-Plan-Implement workflow with quality gates for hallucination, overengineering, and underengineering detection.
Security Assessment
About rpi
The rpi skill is an orchestrator for implementing features from Jira tickets, PRDs, and user requirements using a structured Research-Plan-Implement workflow. Its purpose is to guide feature work through a defined sequence of stages so implementation is grounded in source requirements rather than guesswork. It addresses common delivery risks by introducing explicit quality gates that check for hallucination, overengineering, and underengineering before work proceeds to the next stage.
Its main capabilities include coordinating multiple individual skills in sequence, creating a session for tracking work, generating research and planning artifacts, auditing those artifacts against defined thresholds, and moving approved work into implementation and code review. The documented workflow includes research output, audit gates for confidence, hallucination, traceability, and balance, a user approval step before implementation, and a final review stage that reports findings using P0, P1, and P2 severity levels. It also includes compatibility guidance for different environments, including output directory conventions, differences between Claude Code and Codex CLI invocation patterns, and alternatives for task tracking tools.
This skill is suited for developers and technical teams who need a disciplined workflow for turning requirements into code changes. It is especially relevant when the input is a Jira issue key, a Confluence PRD URL, or a direct feature request such as "implement" or "build." Target users include engineers working in Claude Code or Codex CLI environments who want a repeatable process with checkpoints, documented intermediate outputs, and an explicit approval gate before implementation begins.
FAQ
When should I use the rpi skill?
Use it when the input is a Jira issue key, a Confluence PRD URL, or a described feature request, especially when the user asks to implement, build, or create a feature.
What workflow does the rpi skill follow?
It follows a Research-Plan-Implement sequence with session creation, research, audit of research, planning, audit of plan, user approval, implementation, and code review.
How does compatibility differ between Claude Code and Codex CLI?
The skill uses `.claude/output` for Claude Code and `.codex/output` for Codex CLI. Slash command syntax such as `/rpi` is described as Claude Code syntax, while Codex CLI should invoke the skill by name in the prompt. For task tracking, Claude Code uses TodoWrite, while Codex CLI uses `update_plan` or a simple checklist.
Does the skill require approval before implementation?
Yes. The documented workflow includes a user approval step after the plan audit, where a plan summary is presented and explicit approval is required before implementation.
What quality gates are defined in the workflow?
The research audit requires confidence of at least 60 percent and hallucination of at most 20 percent. The plan audit requires 100 percent traceability and balance of at least 70 percent.
Install rpi
Quick Setup:
- Copy the skill folder to
.claude/skills/ - Claude will automatically detect and use the skill
Repository
ferdiangunawan/rpi-stack