DEV Community

Eduardo fernandes da silva
Eduardo fernandes da silva

Posted on

I Built a Free Client-Side Tool to Audit AI-Generated Code Before Production

The problem nobody talks about after adopting Copilot

If you've shipped with Cursor, Copilot, or ChatGPT in the last year, you've felt this: velocity goes up, but so does the cleanup work after merge.

Hardcoded API keys. Empty catch {} blocks. innerHTML assignments. Copy-pasted duplicate functions the model "forgot" it already wrote.

Research backs it up — AI-assisted code carries roughly 1.7× more bugs than human-written code. The speed win is real. The debt is real too.

What I built: DebtScan AI

DebtScan AI is a 100% client-side scanner. Paste a snippet, get a technical debt score + security flags in ~10 seconds. Nothing leaves your browser.

Try it free: https://produtosagentes.vercel.app/debtscan_ai.html

What it catches today

Category Patterns
Security eval(), innerHTML, hardcoded secrets, OpenAI key patterns (sk-...)
Debt TODO/FIXME comments, debug console.log, empty catch blocks
Quality Duplicate code blocks (copy-paste artifacts)

You get a 0–100 debt score with severity-tagged issues. Score ≥80 = safe to ship. Below 50 = do not merge.

Why client-side only?

Developers won't paste proprietary code into a random SaaS backend. DebtScan runs entirely in the browser — no uploads, no accounts, no trust issues.

Monetization (optional)

The core scan is free forever. There's a $19 full PDF remediation report with line-by-line fixes and a 30-day paydown plan for teams that need audit trails.

How the scoring works

Each pattern has a severity weight:

  • Critical (eval, XSS, secrets): −15 pts each
  • High (empty catch): −10 pts
  • Medium (TODOs, duplicates): −4 to −6 pts
  • Low (console.log): −2 pts

Score = max(0, 100 - total penalty).

Would love your feedback

  1. What patterns am I missing? (e.g. SQL injection strings, dangerouslySetInnerHTML, missing input validation)
  2. Would you use this in CI/pre-merge review? I can open-source the pattern engine.
  3. Is the scoring fair? Happy to tune weights based on real-world feedback.

Built solo this weekend. No VC, no signup wall — just a tool I wished existed before my last AI-assisted PR.


Links:

ai #webdev #javascript #security #showdev

Top comments (0)