VibeCheck

The trust layer between AI-generated software and production.

VibeCheck validates AI-generated code before it ships. Catches security vulnerabilities, accessibility issues, performance problems, code duplication, technical debt, missing tests, and missing documentation.

The Problem

63% of developers spend more time debugging AI-generated code than writing it. But the bigger problem is trust.

AI is lowering the barrier to creating software. Anyone can describe an idea and get code. But:

• Is this code safe?
• Is this AI hallucinating?
• Will this break?
• Is this accessible?
• Is this good enough to launch?
• Am I accidentally creating security problems?

VibeCheck answers those questions.

The Solution

npx @jordan-thirkle/vibecheck ./my-project

VibeCheck runs 6 quality checks and produces a report:

| Check | What It Finds | |-------|---------------| | 🔒 Security | Hardcoded secrets, API keys, tokens | | ♿ Accessibility | Missing alt text, labels, WCAG violations | | ⚡ Performance | Large images, render-blocking resources | | 📋 Duplication | Duplicate files, repeated code | | 🧪 Tests | Missing test files | | 📚 Documentation | Missing README, package.json description |

Output

🔍 VibeCheck — AI Code Quality Gate

✅ Security (0 issues)
⚠️ Accessibility (2 issues)
   🟠 src/index.html
      2 images without alt attribute [WCAG 1.1.1]
   🟡 src/components/Form.tsx
      1 input without label [WCAG 1.3.1]
✅ Performance (0 issues)
✅ Duplication (0 issues)
✅ Tests (0 issues)
✅ Documentation (0 issues)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

📊 Results: ✅ 5 pass | ⚠️ 1 warn | ❌ 0 fail

🏆 Quality Score: 83/100

🟢 Production ready — confidence to ship

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

💡 Recommendations:

  Accessibility: Add alt text to images, labels to inputs.
    → Run axe-core for detailed WCAG audit

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

🤖 AI-Readable Summary:

  score: 83
  ready: true
  security: 0 issues
  accessibility: 2 issues
  performance: 0 issues
  duplication: 0 issues
  tests: 0 issues
  docs: 0 issues

The Mission

AI is lowering the barrier to creating software. The new problem is not:

"Can people create software?"

The new problem is:

"Can people trust what AI creates?"

VibeCheck becomes the trust layer between AI creation and real-world usage.

Who This Is For

Phase 1 (now): Solo developers, indie hackers, AI-assisted engineers Phase 2: Founders, designers, creators, agencies, small businesses Phase 3: Anyone using AI to create software

Product Principles

Expert users need:

• Deep technical analysis
• CI integration
• Developer workflows
• Automation

Beginners need:

• Simple explanations
• Confidence scoring
• Education
• Guidance
• No-code entry points

Both experiences use the same underlying quality intelligence.

Installation

npx @jordan-thirkle/vibecheck ./your-project

Or install globally:

npm install -g @jordan-thirkle/vibecheck
vibecheck ./your-project

Flags

| Flag | Description | |------|-------------| | --json | Machine-readable output for AI tools | | --fix | Auto-fix where possible (coming soon) |

The AI-Specific Wedge

No existing tool checks for AI-generated code failure modes:

| Check | What It Finds | |-------|---------------| | 🔍 Hallucinated Dependencies | Fake packages that don't exist on npm | | 🧬 Synthetic Vulnerabilities | Subtle flaws unique to LLM output | | 📝 AI Artifact Leakage | Hidden prompts, instructions, or artifacts | | 🎯 Semantic Correctness | Code that doesn't match its stated purpose | | 🧩 AI-Specific Edge Cases | Failure modes humans don't make | | 🏗️ Architectural Coherence | Do AI-generated modules work together? |

The wedge: "Did AI generate this code, and does it actually do what it claims?"

Roadmap

• [x] CLI MVP with 6 quality checks
• [ ] Hallucinated dependency detection
• [ ] Synthetic vulnerability detection
• [ ] AI artifact leakage detection
• [ ] Semantic correctness verification
• [ ] GitHub Action
• [ ] Quality badge
• [ ] Plain English explanations (Phase 2)
• [ ] Simple web interface (Phase 2)
• [ ] AI software confidence certificate (Phase 3)
• [ ] CI/CD integration
• [ ] Custom rules
• [ ] Team dashboard

Philosophy

Build narrowly. Position broadly.

V1 is a CLI for developers. The mission is universal: anyone creating software with AI should have confidence in what they create.

License

MIT