Built this after getting frustrated with AI-generated codebases that look fine on the surface but have serious issues underneath. The free roast runs a single security agent — finds hardcoded secrets, injection vulnerabilities, auth flaws. Curious what people think about the approach and whether the findings are useful in practice.