rpi
Original:🇨🇳 Chinese
Translated
General RPI (Research, Plan, Implement, Iterate) execution skill. It is used for engineering tasks where users require "research first, then plan, then implement, and finally iterate", or when tasks are highly complex, high-risk, or have unclear impact. This skill does not rely on specific command-line tools or platforms, and is applicable to any AI Agent that supports skill mechanisms.
1installs
Sourcefirede/agent-skills
Added on
NPX Install
npx skill4agent add firede/agent-skills rpiTags
Translated version includes tags in frontmatterSKILL.md Content (Chinese)
View Translation Comparison →RPI
Overview
- Uses a four-phase process: .
Research -> Plan -> Implement -> Iterate - Strictly maintain phase boundaries: No code modification during Research/Plan phases, verification is mandatory during the Implementation phase.
- Produce structured results at each stage to ensure direct reuse in the next phase.
Usage
Advance phase by phase in the same session:
- Enter the phase, only collect facts and constraints.
Research - Enter the phase, output a reviewable execution plan.
Plan - After the plan is confirmed, enter the phase, execute changes and verify.
Implement - After implementation is completed, enter the phase, conduct a retrospective and define improvement priorities.
Iterate
Workflow Decision-Making
- When requirements are unclear or context is insufficient, start with .
Research - After key facts are clarified, create a and wait for confirmation.
Plan - After the plan is confirmed, execute .
Implementation - After implementation is completed, execute .
Iteration - When key prerequisites change, roll back to the or
Researchphase.Plan
Phase Specifications
1) Research
Objective: Establish credible context, no code modification.
Actions:
- Clarify problem definition, success criteria, and constraint boundaries.
- Identify relevant code, dependencies, configurations, and existing patterns.
- Distinguish between facts, assumptions, and unknowns.
- Extract decision-making information for entering the Plan phase.
Output Structure:
- Summary
- Key Findings
- Context
- Pending Confirmations
- Recommendations
2) Plan
Objective: Form an executable, verifiable, reviewable plan, no code modification.
Actions:
- Break down steps based on confirmed facts.
- Mark affected files, risks, and rollback ideas.
- Define verification schemes and completion criteria.
- Wait for confirmation before entering the Implementation phase.
Output Structure:
- Objectives
- Prerequisites & Assumptions
- Scope & Involved Files
- Execution Steps
- Risks & Mitigation
- Verification Scheme
3) Implement
Objective: Execute changes according to the plan and conduct continuous verification.
Actions:
- Implement step by step, control the scope of changes.
- Verify immediately after each batch of changes (testing, lint, type-check, function check).
- When deviating from the plan, first explain the reason and update the plan.
- Record progress and remaining work.
Output Structure:
- Completed Changes
- Verification Results
- Deviation Explanation
- Remaining Work
4) Iterate
Objective: Conduct a systematic retrospective and define the next round of optimizations.
Actions:
- Evaluate correctness, robustness, maintainability, and observability.
- Check boundary conditions, failure paths, performance, and security risks.
- Identify test gaps and technical debt.
- Output prioritized improvement items.
Output Structure:
- Strengths
- Issues
- Boundary Cases & Gaps
- Prioritized Improvement Items
Quality Gates
- Direct code modification is prohibited during the and
Researchphases.Plan - Skipping verification and directly declaring completion is prohibited during the phase.
Implement - Only giving conclusions without executable improvement items is prohibited during the phase.
Iterate - A structured result must be output at the end of each phase.
References
- Read for detailed checklists and prompt templates.
references/rpi-playbook.md