Source code Security & compliance Code quality Releases Stability

AI-generated code risks

The mistakes coding agents make confidently, and repeatedly.

Overview

Generated code compiles, passes review at a glance, and carries a recognisable set of defects. This check reads your source for the patterns that show up when auth, crypto, or query-building was written by a model and merged without a close read.

What it looks for

  • JWTs decoded instead of verified, or verified without pinning an algorithm
  • MD5, SHA-1, or low-cost bcrypt on passwords, and deprecated crypto primitives
  • eval and dynamic exec reached by user input
  • String-built SQL, path traversal, open redirects, and SSRF sinks
  • Secrets left in comments, TLS verification switched off, scaffold defaults never changed
  • Calls to APIs that do not exist, and @ts-ignore hiding the type error that would have caught them

Why it matters

These are not style problems. A JWT verified without an algorithm check accepts a token signed with none. A password hashed with MD5 is a password already leaked. Both look completely ordinary in a diff.

Rules in this check

How the fixes arrive

Pull requestLow-risk changes arrive as a pull request on a lumioguard branch, with a breakage analysis attached. You review and merge it.
GuideChanges only you can make safely arrive as a step-by-step guide you can follow yourself or hand to your coding agent.

Run them all on your app

Connect your repo and your live services with read-only scopes. The first scan is free, and nothing changes without your approval.