nurettincoban/ai-prd-workflow

What it solves

AI coding agents are highly capable at writing code but often struggle with long-term planning, maintaining consistency across large projects, and remembering project requirements. This workflow solves the problem of "scope creep" and fragmented decision-making by enforcing a structured, document-driven development process that persists across chat sessions.

How it works

The project provides a set of standardized prompts and scripts that guide an AI agent through a rigorous software development lifecycle. Instead of jumping straight into coding, the workflow forces the agent to:

1. Define: Create and verify a Product Requirements Document (PRD).
2. Plan: Extract features, establish project rules, and break the work into small, dependency-ordered Request for Comments (RFC) documents.
3. Build: Implement and review each RFC individually, ensuring code matches the written specifications.
4. Maintain: Use automated checks (like trace-check.py) to ensure that documentation, requirements, and code remain in sync throughout the project.

Who it’s for

This workflow is designed for developers building multi-week projects using AI coding assistants (such as Claude Code, Cursor, GitHub Copilot, or Devin) who want to avoid the pitfalls of unstructured, session-based development.

Highlights

• Persistence: Decisions are captured in markdown files (PRD, RFCs, Rules), ensuring they survive across different chat sessions and AI models.
• Traceability: Automated status checks verify that every feature has a corresponding RFC and that documentation remains consistent.
• Tool Agnostic: Works with various AI coding tools via a standardized skill-based installation or simple prompt copy-pasting.
• Rigorous Review: Includes a "fresh-reader" check where code is reviewed in a separate context from where it was written to catch errors that the original agent might overlook.