kapus-secaudit
kapus-hq/kapus-secaudit
Audit your vibe-coded app for security holes (secrets, RLS, auth, injection, config), score it, and — with your approval — fix them in your code. A Claude Code / Codex skill. By Kapus.
Instalação
npx convez-ai kapus-hq/kapus-secaudit1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
The problem: vibe coding ships working apps fast — but the security is usually missing. The
.envis committed, the Supabase table has RLS off so anyone with the anon key can read every row, the/apiroute returns data with no auth check, the service-role key is sitting in the client bundle. It runs, and it's wide open.kapus-secaudit reads your codebase (or hits your live URL), finds those holes, ranks them by severity, and — in fix mode — patches them in your code after you approve each change.
What it checks
Five areas — the ones vibe coders blow most often:
- General hacking surface (OWASP) — vulnerable deps, dynamic code execution, SSRF, open redirects, missing rate limits, error/stack leakage.
- Secret & key exposure (the #1 P0) — committed
.env, secret keys in the client bundle /NEXT_PUBLIC_*, exposed Supabaseservice_role, Stripe/OpenAI/Anthropic keys in the repo, hardcoded credentials, keys lingering in git history. - Auth & access control — Supabase RLS off, unauthenticated
/apiroutes, IDOR, client-only permission checks, weak sessions, unguarded admin routes. - Input validation & injection — SQL/NoSQL injection, XSS, unchecked file uploads, command injection, prompt injection (LLM apps), unbounded payloads.
- Config & deploy — missing security headers, wildcard CORS, production source maps, debug mode, default credentials, public storage buckets.
You get a 0–100 score, a scorecard, and a findings list with P0–P3 severity — each finding says what, where (file:line / endpoint), why it's dangerous, and how to fix it.
The rules that make it safe
This is a defensive skill. It audits and hardens — it does not attack.
- 🚫 No real exploitation. No injection payloads, no auth-bypass attempts, no brute force, no DoS against live targets — read + safe non-destructive probes only.
- 🔒 Your assets only. URL audits are for sites you own and operate. Third-party audit requests are refused.
- 🎯 No false alarms. Unsure findings are marked "needs confirmation" with how to confirm — no fear-marketing.
- 🔑 Exposed secret? Rotate first. If a key leaked into git history or the client bundle, a
.gitignoreis already too late — it warns you to revoke & reissue the key now, before any code edit. - ✅ Fixes need approval. Even in fix mode it shows you the file-by-file change plan first, and flags anything that changes behavior (adding auth that blocks an existing flow) before touching it.
The model brings the capability; the skill is the leash — the discipline that stops a strong model from "helpfully" running an actual attack or silently breaking your auth flow.
Recommended setup (for best results)
- Model: the strongest reasoning model your harness offers (e.g. a Claude Opus-class model) — better at spotting real holes and avoiding false positives.
- Harness: Claude Code or Codex — it needs to read your files (and write them in fix mode).
- Before fix mode:
git commitorstashso your tree is clean — every change is onegit restoreaway. - For live-URL audits: have the URL ready and confirm it's yours.
- Turn on extended thinking / high reasoning effort for the audit pass.
Install
git clone https://github.com/kapus-hq/kapus-secaudit
cd kapus-secaudit && ./install.sh
Or copy manually into ~/.claude/skills/kapus-secaudit.
Use
In a new Claude Code session:
/kapus-secaudit
or just:
- "is my app safe? check it for security holes"
- "audit https://myapp.com for vulnerabilities" (your own site)
- "find the security issues and fix them" (fix mode)
License
MIT. Use it, fork it, ship it. PRs welcome — especially stack-specific fix recipes and new detection signatures.
🌼 Built by Kapus
Kapus is the BYOK (bring-your-own-key) capability layer for the agent era — focused tools that run on your keys and hand off to your agent. kapus-secaudit is one of the open-source tools from that library (see also kapus-untangle, kapus-copyguard, and kapus-patch).
⭐ If this saved you from a breach, a star helps other builders find it.