Skip to content

Repository files navigation

SkillBadge — In-browser trust scanner for AI agent skills

SkillBadge 🛡️

MIT License Live demo 100% client-side

Is that SKILL.md safe to install? SkillBadge is an in-browser trust scanner for AI agent skills. Paste or drag & drop any SKILL.md file and get an instant heuristic security scan — prompt-injection phrases, exfiltration URLs, curl | bash pipes, credential harvesting, base64 blobs, obfuscated code, destructive shell commands, hardcoded secrets, and more.

🔗 Live demo: https://devilking7x.github.io/skillbadge/

✨ Features

  • 20 heuristic security checks — prompt-injection / instruction overrides, secrecy directives, credential harvesting language, known exfiltration & drop domains (Discord webhooks, webhook.site, ngrok…), raw IPs in URLs, curl|bash / wget|sh pipes, rm -rf (escalated to critical for /, ~, $HOME targets), dangerous chmod, sudo escalation, eval/exec, base64 blobs, obfuscated code (\x escapes, fromCharCode, atob, reversed strings), hardcoded secrets (Stripe/OpenAI/GitHub/AWS key patterns, private keys), env & shell-history snooping, reverse-shell primitives, persistence mechanisms (cron, systemd, rc files), sensitive local paths, URL-encoded links, plain-HTTP URLs, and missing frontmatter.
  • Trust score + verdict — 0–100 score with TRUSTED / CAUTION / SUSPICIOUS / DANGEROUS verdicts, each finding graded critical / warning / info with plain-English explanations and exact line-number matches.
  • Shareable SVG badge — generate a shields-style "scanned" badge with one click, plus copy-paste Markdown and HTML embed snippets for your README.
  • CI-friendly JSON report — downloadable report with an explicit ci.fail flag and threshold, ready to gate your pipeline on untrusted skills.
  • Scan history — every scan is stored in localStorage; re-view or delete past scans anytime.
  • Bundled demo samples — one click loads a clean skill, a deliberately shady skill, or a borderline one, so you can see the scanner in action instantly.
  • 100% client-side — no uploads, no servers, no tracking. Your files never leave the browser.

🚀 How to use

  1. Open the live demo.
  2. Paste a SKILL.md, drag & drop the file, or click a demo sample (clean / shady / borderline).
  3. Review the trust score, verdict, and findings (filter by severity, expand for line-level matches).
  4. Switch to the Badge tab to download your SVG badge and copy the README snippet, or the JSON report tab for the CI-ready report.

Local development

pnpm install
pnpm dev      # http://localhost:5173
pnpm build    # outputs to dist/public

🛠️ Tech stack

  • TypeScript + React 19 + Vite 7
  • Tailwind CSS v4 (dark premium theme, Space Grotesk + JetBrains Mono)
  • lucide-react icons
  • Zero backend — static build deployed to GitHub Pages

📸 Screenshots

SkillBadge SKILL.md security scanner

🔒 Privacy

Everything runs 100% in your browser: scanned files are never uploaded, no servers are contacted, and scan history lives only in your browser's localStorage. Clear it anytime from the History tab.

⚠️ A note on heuristics

SkillBadge detects patterns, not intent. A flagged skill isn't automatically malicious (a DevOps skill may legitimately use curl), and a clean scan isn't a guarantee of safety. Always review flagged findings manually before installing a skill you don't fully trust.

📄 License

MIT — see LICENSE.

About

In-browser trust scanner for AI agent skills (SKILL.md) — security heuristics + shareable badge

Topics

Resources

Contributing

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages