I audit AI-generated code from Cursor, Lovable, Replit, Bolt.new, v0, and Claude Code, and tell you what's wrong before it ships: exposed secrets, auth checks that don't check anything, silent errors, risky dependencies, and data your server should verify instead of trusting the client.
I run a production AI agent handling real customer email, and "looks right" has never been enough to ship alone. Its regression suite runs 25 test cases, sampled 3 times each, with safety-critical behavior gated all-or-nothing. That's the bar I apply to your code: read like a person, prove the behavior holds.... Read more