VibeCodeXray

SAMPLE REPORT

What a VCX audit actually gives you

Not a vague AI review. Not a linter dump. VibeCodeXray returns deterministic findings with files, evidence, severity, and plain-English context so you can decide what to fix before shipping AI-generated code.

1

critical security issue

7

medium/high findings

42

files inspected

100%

evidence-backed findings

FINDINGS

Representative issues VCX catches

These examples mirror the patterns AI assistants produce constantly: code that runs, demos nicely, and quietly stores a rake in the grass for production.

Criticalsrc/app/api/search/route.ts:42

User-controlled SQL reaches a database query

EVIDENCE

The handler concatenates searchParams.get("q") directly into a raw SQL string before calling db.query().

WHY IT MATTERS

AI-generated CRUD endpoints often look harmless in demos. In production, this pattern can leak customer data or allow destructive queries.

FIX DIRECTION

Use parameterized queries and validate the accepted search shape before it reaches the database layer.

Highsrc/app/api/projects/[id]/delete/route.ts:27

Protected action has no ownership check

EVIDENCE

The route checks for a session, then deletes by project id without verifying that project.userId matches session.user.id.

WHY IT MATTERS

This is the classic vibe-coded auth bug: login exists, but authorization is missing where the dangerous action happens.

FIX DIRECTION

Scope the delete query by both project id and authenticated user id, then return 404 for non-owned records.

Mediumsrc/app/dashboard/page.tsx:88

N+1 database reads in dashboard rendering

EVIDENCE

The component loads projects, then fetches findings separately inside a loop for each project card.

WHY IT MATTERS

It works with three demo rows. It becomes slow and expensive once a real user has a few dozen scans.

FIX DIRECTION

Fetch aggregated finding counts in one query and pass the summary into the dashboard cards.

Lowsrc/lib/utils.ts:1

Generated utility file mixes unrelated responsibilities

EVIDENCE

The file contains formatting helpers, auth helpers, validation rules, and billing constants in one shared module.

WHY IT MATTERS

AI assistants frequently dump convenient helpers into one file. That makes future changes riskier because everything imports everything.

FIX DIRECTION

Split auth, billing, formatting, and validation helpers into focused modules before the dependency graph hardens.

REMEDIATION PLAN

What the report helps you do next

A useful audit should shorten the path from scary finding to owner-ready fix. VCX turns raw scan output into a practical review packet for founders and engineering teams working with AI-generated code.

STEP 1

Fix-now queue

Critical and high findings are grouped by production risk first, not by whatever file happened to be scanned first.

OUTPUT

Priority list with owner-ready file paths and fix direction

STEP 2

Evidence trail

Each item includes the observed pattern and the concrete code location so reviewers can reproduce the concern quickly.

OUTPUT

Finding cards with evidence, affected route/module, and severity

STEP 3

Launch decision

The report separates blocking issues from cleanup work so teams can decide whether to ship, pause, or scope a hardening sprint.

OUTPUT

Ship / fix-before-launch / monitor guidance for the repo

HANDOFF CHECKLIST

Turn the audit into a fixable work queue

The sample report is not just a list of warnings. It shows the operating order for an AI-built app: block launch risk first, keep evidence attached, assign the work, and verify the repository again after fixes.

BLOCKERS FIRST

Critical and high findings get fixed before launch claims, investor demos, or payment flows are treated as production-ready.

EVIDENCE RETAINED

Each fix keeps the original file path, rule, observed pattern, and reviewer note attached so the next pass can confirm the risk changed.

OWNER-READY WORK

Findings are grouped into security, dependency, performance, and maintainability lanes so a founder can hand the right work to the right engineer.

RE-SCAN DECISION

After fixes, teams rerun the audit against the updated repository instead of assuming the AI-generated patch removed the original risk.

COVERAGE

Built for the weird failure modes of AI code

AI-generated projects fail differently: missing authorization after beautiful authentication, generated data access patterns that do not scale, and duplicated glue code that nobody remembers asking for. VCX checks those patterns directly.

  • Security patterns: injection, XSS, unsafe auth, secrets, risky redirects
  • Performance patterns: N+1 reads, missing awaits, expensive loops, bundle traps
  • Dependency checks: known vulnerabilities, deprecated packages, risky licenses
  • Code quality: dead exports, God files, duplicated logic, fragile generated glue

Run this against your own repo.

Free tier, no credit card, and enough audits to find out whether your AI-built app is production-ready or just wearing a nice landing page as a disguise.

Start free