Afterburn

One command finds every bug on your website. Zero config. Generates a report your AI coding tool can use to fix them automatically.

npx afterburn-cli https://your-site.com

Afterburn v1.0.2

[ok] Checking browser...
[ok] Crawling site...
[ok] Testing workflows...
[ok] Analyzing results...
[ok] Generating reports...

Health: 62/100 - 20 issues found (0 high, 17 medium, 3 low)

Reports saved:
  HTML:     afterburn-reports/1770564079516/report.html
  Markdown: afterburn-reports/1770564079516/report.md

Afterburn crawls your site, fills out forms, clicks every button, and tells you what's broken -- in plain English for you, in structured Markdown for Claude, Cursor, or ChatGPT to auto-fix.

The AI workflow: Afterburn finds the bugs. You paste the Markdown report into Claude, Cursor, or ChatGPT. It reads the reproduction steps and fixes them. You re-scan. Done.

The AI-readable report (the killer feature)

Every scan generates two reports:

1. HTML Report -- A health score and prioritized bug list in plain English. For you to read.
2. Markdown Report -- Structured with reproduction steps and fix suggestions. Feed it directly to Claude, ChatGPT, or Cursor and your AI coding tool can fix the bugs automatically.

You: "Here's my Afterburn report. Fix these bugs."
AI:  *reads structured report, fixes 12 issues across 4 files*

This is the workflow: Afterburn finds the bugs, your AI fixes them. Zero manual debugging.

What the AI report looks like

The Markdown report is structured so AI tools can parse and act on it:

## Issues (Prioritized)

| # | Priority | Category | Summary | Location |
|---|----------|----------|---------|----------|
| 1 | HIGH | Workflow Error | Form "form#contact-form" submission isn't working | /contact |
| 2 | MEDIUM | Dead Button | Button "Get Started Free" doesn't do anything when clicked | / |
| 3 | MEDIUM | Console Error | An image failed to load | /images/icon-deploy.png |

Each issue includes a category, priority, plain-English summary, and location. Paste this into Claude or Cursor and it knows exactly what to fix.

Sample reports

See what Afterburn produces when scanning a real app with 20+ intentional defects:

• Sample HTML report -- Health score, prioritized issues, fix suggestions
• Sample Markdown report -- Structured for AI coding tools
• Terminal output -- What you see in the CLI

What it catches

• Broken forms that don't submit
• Dead buttons that do nothing when clicked
• JavaScript errors and console crashes
• HTTP errors (404s, 500s, failed API calls)
• Broken images and missing assets
• Accessibility violations (WCAG 2.1 AA via axe-core)
• Slow page loads and poor performance scores

Quick start

# No install needed -- run directly
npx afterburn-cli https://your-site.com

# Or install globally
npm install -g afterburn-cli

Afterburn downloads a browser automatically on first run (~200MB, one-time).

Try it now

Scan any public website -- no signup, no API key, no config:

npx afterburn-cli https://en.wikipedia.org/wiki/Main_Page

Or point it at your own site to see what's broken.

Common options

# Test login flows
npx afterburn-cli https://your-site.com --email [email protected] --password mypass123

# Hint at specific workflows
npx afterburn-cli https://your-site.com --flows "signup, checkout, profile edit"

# Map bugs to your source code (file:line)
npx afterburn-cli https://your-site.com --source ./src

# Watch the browser do its thing
npx afterburn-cli https://your-site.com --no-headless

Reports are saved to ./afterburn-reports/{timestamp}/.

How it works

Your URL --> Crawl all pages --> Plan test workflows --> Execute tests --> Analyze failures --> Generate reports

1. Crawl -- Discovers all pages, forms, buttons, and links. Handles SPAs (React, Next.js, Vue, Angular, Svelte).
2. Plan -- AI generates smart test workflows (or heuristic fallback with no API key).
3. Execute -- Runs each workflow in a real browser: fills forms, clicks buttons, captures screenshots.
4. Analyze -- Diagnoses root causes. AI vision analysis with Gemini, or pattern matching without.
5. Report -- Health score (0-100) and prioritized fix list in HTML + Markdown.

AI is optional

Afterburn works without any API keys. Set GEMINI_API_KEY for smarter test planning and AI-powered diagnosis, but the core tool runs a full test suite either way.

# Optional: enable AI-powered planning and diagnosis
export GEMINI_API_KEY=your-key-here

How is this different?

| Feature | Afterburn | Lighthouse | axe-core | Manual testing | |---------|-----------|------------|----------|----------------| | Zero config (no test writing) | Yes | Yes | Needs integration | Yes | | Crawls entire site | Yes | Single page | No | Tedious | | Form filling and submission | Yes | No | No | Manual | | Dead button detection | Yes | No | No | If you notice | | Plain English reports | Yes | Partial | No | N/A | | AI-ready report for auto-fix | Yes | No | No | No | | Free and open source | Yes | Yes | Yes | Free (your time) | | CI/CD integration | GitHub Action | CI plugin | CI plugin | No |

Health score

The health score (0-100) is weighted:

| Category | Weight | What it measures | |----------|--------|------------------| | Workflows | 40% | Did test workflows complete without errors? | | Errors | 30% | HTTP errors, console crashes, broken resources | | Accessibility | 20% | WCAG 2.1 AA violations | | Performance | 10% | Page load times, LCP scores |

Pre-flight check

npx afterburn-cli doctor

Checks Node.js version, browser installation, API key, and network connectivity.

Testing

# Build CLI + bundled GitHub Action artifacts
npm run build:release

# Fast default lane (unit tests)
npm test

# Deterministic end-to-end lane (local fixture server)
npm run test:e2e

# Unit coverage gate with thresholds
npm run test:coverage

# Optional real-internet smoke test (only runs when URL is provided)
AFTERBURN_EXTERNAL_SMOKE_URL=https://example.com npm run test:smoke:external

CLI reference

Usage: afterburn [options] [command] <url>

Automated testing for vibe-coded websites

Arguments:
  url                    URL to test

Options:
  -V, --version          output the version number
  --source <path>        Source code directory for pinpointing bugs
  --email <email>        Login email (or set AFTERBURN_EMAIL env var)
  --password <password>  Login password (tip: use AFTERBURN_PASSWORD env var
                         to avoid shell history exposure)
  --output-dir <path>    Custom output directory
                         (default: ./afterburn-reports/{timestamp})
  --flows <hints>        Comma-separated workflow hints (e.g., "signup, checkout")
  --max-pages <n>        Max pages to crawl (default: 50, max: 500)
  --no-headless          Show browser window (useful for debugging)
  --verbose              Show detailed progress output
  -h, --help             display help for command

Commands:
  doctor                 Check if your environment is ready to run Afterburn

Known limitations

• Visual analysis requires GEMINI_API_KEY. Without it, UI analysis relies on axe-core plus pattern matching.
• SPA support is experimental. Complex client-side routing may not be fully exercised.
• Desktop viewport only. Tests run at 1920x1080. Mobile testing planned for v2.

Three interfaces

CLI

npx afterburn-cli https://your-site.com

GitHub Action

name: Afterburn
on: [pull_request]

jobs:
  test:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: gods-strongest-vibecoder/afterburn@main
        with:
          url: https://your-staging-site.com
          fail-on: high

Posts a summary comment on your PR and uploads reports as artifacts.

MCP Server (for AI coding assistants)

{
  "mcpServers": {
    "afterburn": {
      "command": "npx",
      "args": ["afterburn-mcp"]
    }
  }
}

Exposes a scan_website tool that returns structured results your AI assistant can act on.

Tech stack

Playwright (browser automation) | Gemini 2.5 Flash (AI vision, optional) | axe-core (accessibility) | Commander.js (CLI) | Handlebars (HTML reports) | ts-morph (source mapping) | sharp (screenshot compression)

License

MIT

Built for the BridgeMind Vibeathon 2026.