Vibrant

AI coding assistants are genuinely useful. They're also responsible for a new category of subtle bugs easy to miss in review: hardcoded API keys, empty catch blocks that swallow errors silently, leftover console.logs, and shortcuts that look fine but fall apart under real conditions. Vibrant runs 15+ static analysis rules tuned to the patterns AI assistants tend to produce. When rules aren't enough, it sends flagged code to an LLM (OpenAI, Claude, Gemini, or Ollama) for deeper contextual analysis. No install required: just `npx vibrant .`. The interesting challenge was building a rule engine that's extensible without being complicated, and a CLI experience that feels good in both local terminals and CI pipelines.