Working with AI can sometimes feel unpredictable. The AI Governor Framework provides a development workflow designed to fix that, for both new and existing projects.
It provides a structured, sequential process that transforms your AI from a simple code generator into a reliable engineering partner. By following these five protocols, you ensure that both you and the AI are always aligned, moving from a high-level idea to a well-implemented feature with clarity, control, and consistency.
The goal is to make AI-powered development:
- Predictable: Each step has a clear purpose and output.
- Controllable: You are always in the loop for key decisions.
- Efficient: The AI does the heavy lifting, you provide the strategic direction.
This workflow guides you through the entire development process, from initial setup to continuous improvement. Each protocol assigns a specific role to the AI, ensuring a structured and predictable collaboration with intelligent session management.
Role: The AI acts as a Project Analyst.
First, the AI analyzes your entire codebase to build a "Context Kit"—a set of foundational READMEs and project-specific rules. This is a one-time protocol that turns a generic AI into a project-specific expert.
Apply instructions from dev-workflow/0-bootstrap-your-project.md
(For best results, Cursor users should use Max Mode for this step.)
Role: The AI acts as a Product Manager.
Next, you define the "what" and "why" of your feature. The AI interviews you to create a detailed Product Requirements Document (PRD), ensuring all requirements are captured before any code is written.
Apply instructions from dev-workflow/1-create-prd.md
Here's the feature I want to build: [Describe your feature in detail]
(For best results, Cursor users should use Max Mode for this step.)
Role: The AI acts as a Tech Lead.
The AI then transforms the PRD into a granular, step-by-step technical execution plan. This ensures that both you and the AI are aligned on the implementation strategy.
Apply instructions from dev-workflow/2-generate-tasks.md to @prd-my-feature.md
(Note: Replace @prd-my-feature.md with the actual filename of your PRD.)
(For best results, Cursor users should use Max Mode for this step.)
Role: The AI acts as a Paired Developer with built-in quality validation.
Here, the AI implements the plan one parent task at a time, within a dedicated chat session. Each session includes implementation + quality audit + retrospective for optimal context preservation.
🔄 Per Parent Task (New Chat Session):
-
Start a parent task in a fresh session:
Apply instructions from dev-workflow/3-process-tasks.md to @tasks-my-feature.md. Start on task 1.(Note: Replace
@tasks-my-feature.mdwith the actual filename of your task list.) -
Implementation with Quick Reviews:
- AI implements sub-tasks with integrated quick validation
- Critical issues are caught and fixed immediately
- You review and approve: "yes" or "continue"
-
Protocol 4: Unified Quality Review (Same Session) Once the parent task is complete, use the unified review system by calling the central orchestrator:
🎯 Tool-Agnostic Command:
Apply instructions from dev-workflow/4-quality-audit.mdThe orchestrator provides an interactive selection interface:
- ☐ Code Review - DDD compliance + Code quality (quick feedback)
- ☐ 🚀 Run All - Comprehensive 6-layer validation
- ☐ Security Check - Focus on auth/data/multi-tenant validation
- ☐ Architecture Review - Focus on performance + DDD architecture
- ☐ Design System - Component usage + Visual consistency
- ☐ UI/UX & Accessibility - WCAG compliance + Responsive design
- ☐ Pre-Production Security - Complete security validation
Automatic Fallback: Custom protocols → Generic protocols (tool-agnostic)
-
Protocol 5: Implementation Retrospective (Same Session) Capture learnings while context is fresh:
Apply instructions from dev-workflow/5-implementation-retrospective.md -
Start the next parent task in a new chat:
Apply instructions from dev-workflow/3-process-tasks.md to @tasks-my-feature.md. Start on task 2.
This streamlined 5-protocol workflow with a unified review system eliminates tool-specific complexity while maintaining all quality benefits. Key improvements:
- Simplified entry point with a single, central orchestrator
- Tool-agnostic design works across Claude Code, Cursor, Aider
- Context optimization through intelligent session management
- Smart protocol selection with automatic custom/generic fallback
- Interactive selection prevents tool-specific confusion
- 6-layer validation including Design + UX coverage
Tool-Specific Usage:
Claude Code: Use the /review command for interactive selection.
Cursor: Use @review for interactive selection.
Aider: Use /load .ai-governor/dev-workflow/4-quality-audit.md
Universal: All tools use the same underlying protocol system.
The initial structured approach pays off exponentially as the AI learns your codebase patterns and becomes a true expert companion for your project with predictable, high-quality output across any development tool.