vibesec-skill

name	VibeSec-Skill
description	This skill helps Claude write secure web applications. Use this when working on any web application or when a user requests a scan or audit to ensure security best practices are followed.
paths	**/*.html,**/*.htm,**/*.js,**/*.ts,**/*.jsx,**/*.tsx,**/*.vue,**/*.svelte,**/*.py,**/routes*,**/auth*,**/views*,**/controllers*,**/api*,**/nginx*,**/Dockerfile*,**/docker-compose*

Secure Web App Checklist

Principles: Defense in depth, fail closed, least privilege, validate server-side, encode output contextually.

Access Control

• Verify user owns resource on EVERY request (server-side, not client)
• Use UUIDv4, not sequential IDs
• Check org membership for multi-tenant apps
• On account removal: revoke all tokens/sessions immediately
• Watch for: IDOR, privilege escalation, horizontal/vertical access, mass assignment

XSS Prevention

Sanitize ALL user-controllable inputs:

• Direct: forms, search, file names, rich text
• Indirect: URL params/fragments, headers, third-party API data, WebSocket/postMessage, localStorage
• Overlooked: error messages reflecting input, PDF generators, email templates, admin log viewers, JSON rendered as HTML, SVG uploads, markdown with HTML

Defenses:

• Output encode per context (HTML entities, JS escape, URL encode, CSS escape)
• Use framework auto-escaping (React JSX, Vue {{ }})
• CSP: default-src 'self'; script-src 'self'; frame-ancestors 'none' — avoid unsafe-inline/unsafe-eval
• Sanitize with DOMPurify; whitelist tags for rich text
• Headers: X-Content-Type-Options: nosniff, X-Frame-Options: DENY

CSRF Protection

• Protect ALL state-changing endpoints (POST/PUT/PATCH/DELETE) AND login/signup/password reset/OAuth callbacks
• Use crypto-random tokens tied to session + validate server-side + regenerate on auth change
• Set SameSite=Strict; Secure; HttpOnly on session cookies
• Missing token = rejected request (don't check presence optionally)
• JSON APIs are NOT immune — validate Origin/Referer AND use tokens

Secrets Exposure

Never in client-side code: API keys, DB strings, JWT secrets, encryption keys, OAuth secrets, internal URLs Check for leaks in: JS bundles, source maps, HTML comments, hidden fields, data attributes, localStorage, SSR hydration data, NEXT_PUBLIC_*/REACT_APP_* env vars

Open Redirect

• Allowlist valid redirect domains, or accept relative paths only (starts with /, no //)
• Block bypasses: @ symbol, subdomain abuse, javascript: protocol, double URL encoding, backslash, null byte, data URLs, protocol-relative //, unicode homographs

Password Security

• Min 8 chars (12+ recommended), no max (or 128), allow all chars, don't require specific types
• Hash with Argon2id, bcrypt, or scrypt — NEVER MD5/SHA1/plain SHA256

SSRF Prevention

• Allowlist approach preferred (pre-approved domains only)
• Resolve DNS before request, validate IP is not private/internal, pin resolved IP
• Block cloud metadata: 169.254.169.254, metadata.google.internal
• Block bypasses: decimal/octal/hex IP, IPv6 localhost [::1], DNS rebinding, CNAME to internal, redirect chains
• Limit/disable redirect following; validate each hop; set timeouts and response size limits

File Upload Security

• Validate: extension (allowlist), magic bytes, file size (server-side)
• Block: double extension .php.jpg, null byte %00, MIME spoofing, polyglot files, SVG with JS, XXE via DOCX/XLSX, ZIP slip ../, ImageMagick exploits
• Handle: rename to UUID, store outside webroot, serve with Content-Disposition: attachment + nosniff, use CDN/separate domain

SQL Injection

• PRIMARY: parameterized queries / prepared statements — never concatenate user input
• ORDER BY / table names can't be parameterized — whitelist only
• Escape LIKE wildcards (%, _)
• DB user must have minimum required permissions

XXE Prevention

• Disable DTD processing, external entities, XInclude
• Applies to: SOAP, XML-RPC, XML uploads, RSS/feeds, DOCX/XLSX/PPTX (ZIP+XML), SVG, SAML
• Python: defusedxml library or etree.XMLParser(resolve_entities=False, no_network=True)

Path Traversal

• Never use user input directly in file paths
• Use indirect references (mapping): files.get(user_input)
• If path needed: canonicalize with os.path.abspath(os.path.realpath(...)), verify commonpath with base dir
• Reject .., absolute paths, whitelist allowed chars

Security Headers

Strict-Transport-Security: max-age=31536000; includeSubDomains; preload
Content-Security-Policy: default-src 'self'; script-src 'self'; frame-ancestors 'none'
X-Content-Type-Options: nosniff
X-Frame-Options: DENY
Referrer-Policy: strict-origin-when-cross-origin
Cache-Control: no-store  # sensitive pages

JWT Security

• Explicitly specify algorithm on verify — reject alg: none, never derive from token
• Use 256+ bit random secrets, always set exp, use short-lived tokens (15min) + refresh rotation
• Store in httpOnly/Secure/SameSite=Strict cookies — NEVER localStorage

API Security

• Mass assignment: whitelist allowed fields — never Model.update(req.body) directly
• GraphQL: disable introspection in prod, limit query depth (10), enforce cost limits, limit batch size
• Rate limiting: apply to auth endpoints, API calls, file uploads

General Rules

1. Validate all input server-side
2. Use parameterized queries
3. Encode output per context
4. Auth + authz checks on every endpoint
5. Handle errors securely (no stack traces to users)
6. Keep dependencies updated
7. When unsure, choose the more restrictive option