An RPI inspired CONTRIBUTING.md to help AI's work and keep humans in the loop (gist.github.com)

A new Research-Plan-Implement (RPI) workflow has been proposed to enhance AI-assisted software development while ensuring human oversight. This framework emphasizes structured phases—research, planning, and implementation—each featuring checkpoints for AI self-reviews and human approval. By organizing work into dedicated plan folders linked to specific Git branches, teams can streamline their development processes and maintain documentation integrity. The document specifically outlines how AI and human roles are delineated in tasks, reinforcing a collaborative dynamic that keeps humans "in the loop." The significance of this approach lies in its emphasis on quality assurance through a structured iterative process. Known as Context Engineering, the practice allows AI tools to operate within limited scope by confining their focus to relevant artifacts, thus preventing context pollution. The RPI workflow not only provides a clear guideline for maintaining high standards in software development but also promotes transparency and accountability within AI-assisted projects. Overall, the incorporation of systematic human review checkpoints is poised to improve software quality outcomes while supporting the evolving role of AI in development.