Framework for governing AI agent behavior in software development. Enforces consistency and quality across AI-generated code with plan gating, self-healing, and drift detection.
What I Learned
Constitutional AI principles apply beyond language models — they work for code generation too
The framework needs to learn, not just enforce
Plan gating is the single most effective quality control
The Problem
AI coding tools generate inconsistent code. Without governance, you get drift in patterns, violations of standards, and no auditability. Hard to adopt AI coding at scale without controls.
The Approach
Created a governance framework that wraps AI coding tools with rules. Every AI-generated change goes through plan gating, validation, and evolution tracking. The framework learns from the codebase over time.
Architecture Decisions
Plan gating system (MVP, Surgical, Systemic)
Self-healing blueprint engine with drift detection
Evolution story tracking (learns from patterns)
Cross-framework learning engine
Why I Built It This Way
Built as a governance layer, not a replacement for AI tools
Inspired by Anthropic's constitutional approach
Made framework-agnostic to work with any AI coding tool
Technology Stack
TypeScriptBunVitePlaywrightVitestCLIDocker
Outcomes
Constitutional governance for AI code generation
Automated quality enforcement
Reproducible blueprint system
Started: 2024-08
6 months to v2.5