vibecheck

Vibe code. Vibe check. 156 security checks for Claude Code.

Blocks dangerous patterns before Claude writes them to disk.

The Problem

AI coding tools generate insecure code by default. Supabase schemas without RLS. API keys in frontend bundles. JWT tokens that never expire. CORS open to the world. Passwords hashed with MD5.

Audit tools find these problems after the fact. vibecheck finds them before they exist.

You:        "Create a Stripe integration"
Claude:     const stripe = require("stripe")("sk_live_51H7abc123...");
vibecheck:  BLOCKED. Stripe live key hardcoded. Use process.env.STRIPE_SECRET_KEY.
Claude:     Rewrites with environment variable.

The vulnerability never reaches your codebase.

Quick Start

1. Clone

git clone https://github.com/Wishmakingfairy/vibecheck.git ~/.claude/plugins/vibecheck

2. Install

claude plugin add ~/.claude/plugins/vibecheck

3. Build with confidence. Every file Claude writes now goes through 156 security checks. Zero configuration needed.

Demo

Here is what happens when Claude tries to write insecure code with vibecheck installed:

$ echo '{"tool_name":"Write","tool_input":{"file_path":"app.js",
  "content":"const stripe = require(\"stripe\")(\"sk_live_abc123\");"}}' | python3 hooks/security_gate.py

BLOCKED this write. Security vulnerabilities detected:

  [SEC-004] Secrets & API Keys | Stripe live secret key detected.
  Use environment variables: process.env.STRIPE_SECRET_KEY.
  CWE: CWE-798

  Suppress if intentional: // vibecheck-disable SEC-004

$ echo '{"tool_name":"Write","tool_input":{"file_path":"schema.sql",
  "content":"ALTER TABLE users DISABLE ROW LEVEL SECURITY;"}}' | python3 hooks/security_gate.py

BLOCKED this write. Security vulnerabilities detected:

  [DB-001] Database Security | Row Level Security (RLS) is being DISABLED.
  Keep RLS enabled. Create policies for granular access control.
  CWE: CWE-862

$ echo '{"tool_name":"Write","tool_input":{"file_path":"utils.js",
  "content":"crypto.createHash(\"md5\").update(password).digest(\"hex\")"}}' | python3 hooks/security_gate.py

BLOCKED this write. Security vulnerabilities detected:

  [CRYPTO-001] Cryptography | MD5 used for password hashing.
  Use bcrypt: const hash = await bcrypt.hash(password, 12)
  CWE: CWE-328

Every block includes: what is wrong, how to fix it, and the CWE ID.

156 Checks, 15 Categories

Category	#	Examples
Secrets & API Keys	22	AWS keys, Stripe keys, GitHub tokens, .env without .gitignore, service account JSON, high-entropy strings
Authentication	20	JWT without expiry, no rate limiting, OAuth CSRF, weak passwords, localStorage tokens
Database Security	20	Supabase RLS disabled, SQL injection, mass assignment, NoSQL injection, GraphQL depth
Network & CORS	11	CORS wildcard, SSRF (IPv6/metadata), open redirects, mixed content
Injection	21	XSS in React/Vue/Angular/Svelte, command injection, path traversal, prototype pollution
Security Headers	11	Debug mode in prod, stack traces in responses, directory listing, server version
Supply Chain	15	Typosquatted packages, unpinned GitHub Actions, curl
Infrastructure	10	.env in public/, Docker as root, admin without auth, source maps in prod
AI/LLM Security	16	API keys in frontend, system prompt exposure, prompt injection, streaming as HTML
Cryptography	12	MD5 for passwords, Math.random() for tokens, ECB mode, timing attacks
Privacy	8	PII in logs, credit cards in code, user data in error responses
Business Logic	6	Race conditions, missing idempotency, enumeration attacks
Logging	5	Passwords in logs, log injection, console.log with secrets
File System	5	Path traversal, zip slip, chmod 777, symlink attacks
WebSocket	4	No auth, missing origin validation, no message size limits

Full check reference (all 156 IDs)

Secrets & API Keys (SEC)

ID	What	Severity
SEC-001	AWS access keys (AKIA pattern)	CRITICAL
SEC-002	AWS secret access keys	CRITICAL
SEC-003	GitHub tokens (ghp/gho/ghu/ghs/ghr)	CRITICAL
SEC-004	Stripe live secret keys (sk_live)	CRITICAL
SEC-005	Frontend env vars exposing secrets (NEXT_PUBLIC/VITE/REACT_APP)	CRITICAL
SEC-006	Hardcoded passwords in assignments	CRITICAL
SEC-007	Private key content (RSA/EC/DSA/SSH)	CRITICAL
SEC-008	JWT secret hardcoded	CRITICAL
SEC-009	Database connection strings with credentials	CRITICAL
SEC-010	Webhook secrets in source	WARNING
SEC-011	API keys in URL query parameters	WARNING
SEC-012	Secrets in Docker ENV directives	CRITICAL
SEC-013	High-entropy strings near assignments (Shannon > 4.5)	WARNING
SEC-014	Secrets in code comments	WARNING
SEC-015	Default credentials (admin:admin, root:root)	CRITICAL
SEC-016	.env file written without .gitignore entry	CRITICAL
SEC-017	Secrets in Terraform/IaC variables	CRITICAL
SEC-018	Secrets in GitHub Actions workflows	CRITICAL
SEC-019	Secrets in Kubernetes manifests	CRITICAL
SEC-020	Cloud metadata endpoint patterns	CRITICAL
SEC-021	Supabase service_role key in frontend	CRITICAL
SEC-022	Google/Firebase service account JSON	CRITICAL

Authentication (AUTH)

ID	What	Severity
AUTH-001	Login route without rate limiting	WARNING
AUTH-002	Missing brute-force protection	WARNING
AUTH-003	Weak password regex (< 8 chars)	WARNING
AUTH-004	No MFA on sensitive routes	INFO
AUTH-005	Session not regenerated after login	WARNING
AUTH-006	Missing session expiry	WARNING
AUTH-007	JWT.sign without expiresIn	CRITICAL
AUTH-008	JWT algorithm set to "none"	CRITICAL
AUTH-009	JWT stored in localStorage	WARNING
AUTH-010	Forms without CSRF token	WARNING
AUTH-011	Password reset token without expiry	WARNING
AUTH-012	No account lockout	INFO
AUTH-013	Direct object reference without auth	WARNING
AUTH-014	Missing role check on admin routes	INFO
AUTH-015	OAuth missing state parameter	CRITICAL
AUTH-016	Cookies missing Secure/HttpOnly/SameSite	WARNING
AUTH-017	Password in reversible encryption	CRITICAL
AUTH-018	Missing re-auth for sensitive ops	WARNING
AUTH-019	Refresh token rotation missing	WARNING
AUTH-020	Magic link tokens without expiry	WARNING

Database (DB)

ID	What	Severity
DB-001	Supabase RLS disabled	CRITICAL
DB-002	SQL string concatenation	CRITICAL
DB-003	Missing parameterized queries	WARNING
DB-004	Database credentials in frontend	CRITICAL
DB-005	Missing input validation before query	WARNING
DB-006	GRANT ALL PRIVILEGES	WARNING
DB-007	CREATE TABLE without RLS	WARNING
DB-008	Unencrypted backups	INFO
DB-009	Missing audit logging	INFO
DB-010	GraphQL resolver without auth	WARNING
DB-011	NoSQL injection ($where, $regex)	CRITICAL
DB-012	Missing encryption at rest	INFO
DB-013	Prisma $queryRaw / Drizzle raw SQL	CRITICAL
DB-014	Missing foreign key on user_id	INFO
DB-015	Admin routes without auth	CRITICAL
DB-016	Mass assignment (req.body to ORM)	WARNING
DB-017	Missing pagination (unbounded queries)	WARNING
DB-018	GraphQL without depth limit	WARNING
DB-019	No connection pooling	INFO
DB-020	Supabase Realtime without RLS	WARNING

Network & CORS (NET)

ID	What	Severity
NET-001	CORS origin: *	CRITICAL
NET-002	CORS credentials with wildcard	CRITICAL
NET-003	Overly permissive CORS methods	WARNING
NET-004	Missing HTTPS redirect	WARNING
NET-005	Missing HSTS	WARNING
NET-006	Mixed content (HTTP on HTTPS)	WARNING
NET-007	Missing cert pinning for mobile	INFO
NET-008	Open redirect	WARNING
NET-009	SSRF (unvalidated URL fetch)	CRITICAL
NET-010	Missing origin validation	WARNING
NET-011	Binding to 0.0.0.0 in prod	WARNING

Injection (INJ)

ID	What	Severity
INJ-001	dangerouslySetInnerHTML (without DOMPurify)	CRITICAL
INJ-002	innerHTML with variable	CRITICAL
INJ-003	Vue v-html	CRITICAL
INJ-004	Angular [innerHTML]	CRITICAL
INJ-005	Svelte {@html}	CRITICAL
INJ-006	Command injection (exec/spawn)	CRITICAL
INJ-007	Path traversal (../)	CRITICAL
INJ-008	Template literal in SQL/shell	WARNING
INJ-009	LDAP injection	CRITICAL
INJ-010	XXE (XML external entities)	CRITICAL
INJ-011	Unsafe deserialization (pickle, yaml.load)	CRITICAL
INJ-012	Missing Content-Type validation	WARNING
INJ-013	File upload without type whitelist	WARNING
INJ-014	File upload without size limit	WARNING
INJ-015	ReDoS vulnerable regex	WARNING
INJ-016	Header injection	WARNING
INJ-017	Prototype pollution	CRITICAL
INJ-018	CSS injection via user input	WARNING
INJ-019	GraphQL introspection in prod	WARNING
INJ-020	CRLF injection	WARNING
INJ-021	eval() / new Function() / setTimeout(string)	CRITICAL

Security Headers (HDR), Supply Chain (SUP), Infrastructure (INF), AI/LLM (AI), Cryptography (CRYPTO), Privacy (PRIV), Business Logic (BIZ), Logging (LOG), File System (FS), WebSocket (WS)

See skills/security-scan/references/ for complete documentation of all checks with regex patterns, fix examples, and false positive notes.

How It Works

vibecheck runs as a PreToolUse hook on every Write, Edit, and MultiEdit operation:

Claude tries to write a file
         |
         v
  vibecheck intercepts
         |
  Detect file extension (.js? .sql? .py? .yaml?)
         |
  Load ONLY relevant checkers (skip DB checks for .css)
         |
  Scan content against matched patterns
         |
    +---------+---------+
    |         |         |
 CRITICAL  WARNING    INFO
 exit(2)   exit(0)   silent
 BLOCKS    warns     scan only
 write     allows

Severity	What Happens	Example
`CRITICAL`	Blocks the write. Claude must fix and retry.	Hardcoded API key, RLS disabled, SQL injection
`WARNING`	Warns in stderr. Write proceeds.	JWT in localStorage, missing rate limiting
`INFO`	Silent during hooks. Shown in full scan mode.	Missing MFA, no structured logging

Performance: ~88ms average per write. Only loads checkers relevant to the file type.

Smart Detection

vibecheck is not a dumb regex scanner. It understands context:

Feature	How It Works
Test file awareness	`.test.`, `.spec.`, `__tests__/`, `fixtures/` auto-downgrade CRITICAL to WARNING
DOMPurify detection	`dangerouslySetInnerHTML` only flagged if DOMPurify is NOT imported in the same file
Public API awareness	CORS wildcard not flagged if public API patterns detected in the same file
Comment context	Comments with "BAD:", "DON'T", "NEVER", "WRONG" recognized as educational, not vulnerabilities
Entropy analysis	Catches secrets without known prefixes via Shannon entropy > 4.5 on 20+ char strings
Inline suppression	`// vibecheck-disable SEC-001` skips a specific check when you know what you are doing

Configuration

Create .vibecheck.json in your project root. Entirely optional.

{
  "severity_overrides": {
    "NET-001": "WARNING"
  },
  "disabled": [],
  "ignore_paths": ["**/*.test.*", "**/fixtures/**"],
  "inline_suppression": true,
  "framework": "auto"
}

Configuration options explained

Field	Type	Default	Description
`severity_overrides`	`object`	`{}`	Change severity for specific checks. `{ "NET-001": "WARNING" }` downgrades CORS wildcard from CRITICAL to WARNING.
`disabled`	`string[]`	`[]`	Disable specific checks entirely. `["SEC-013"]` disables entropy analysis.
`ignore_paths`	`string[]`	`[]`	Glob patterns for files to skip. Test files are already handled automatically.
`inline_suppression`	`boolean`	`true`	Enable/disable `// vibecheck-disable` comments.
`framework`	`string`	`"auto"`	Force a framework (`nextjs`, `vite`, `express`, `django`). Auto-detects from package.json.

Skills and Agent

Skills

vibecheck includes 4 skills that activate when you ask about security topics:

Skill	Trigger Phrases	What It Does
security-scan	"security scan", "vulnerability check"	Full codebase audit with categorized report
secure-keys	"secure API keys", "handle secrets"	Framework-specific secret management (Next.js, Vite, Express, Supabase)
secure-auth	"secure login", "rate limiting"	Auth hardening: rate limiting, JWT, bcrypt, CSRF, OAuth
secure-deploy	"deploy securely", "pre-launch check"	Pre-deployment security checklist

Agent

The security-scanner agent performs deep codebase analysis:

Detects stack from package.json / requirements.txt / go.mod
Discovers all source files (excludes node_modules, dist, .git)
Runs npm audit or pip audit for dependency vulnerabilities
Cross-file analysis: auth middleware on admin routes, RLS on DB tables
Outputs categorized report sorted by severity with CWE mappings

CWE Mapping

Every check maps to a Common Weakness Enumeration ID for compliance reporting:

68 CWE IDs covered

CWE	Description	Checks
CWE-798	Hardcoded credentials	SEC-001 to SEC-022
CWE-89	SQL injection	DB-002, DB-003, DB-013
CWE-79	Cross-site scripting	INJ-001 to INJ-005
CWE-862	Missing authorization	DB-001, DB-007, DB-010
CWE-918	SSRF	NET-009, SEC-020
CWE-352	CSRF	AUTH-010, AUTH-015
CWE-328	Weak hash	CRYPTO-001
CWE-330	Weak RNG	CRYPTO-004, BIZ-004
CWE-22	Path traversal	INJ-007, FS-001, FS-005
CWE-78	Command injection	INJ-006
CWE-502	Unsafe deserialization	INJ-011
CWE-306	Missing authentication	DB-015, AI-014
CWE-613	Insufficient session expiration	AUTH-007
CWE-942	Permissive CORS	NET-001, NET-002
CWE-532	Sensitive data in logs	LOG-002, LOG-005, PRIV-001
CWE-732	Incorrect permissions	FS-004
CWE-829	Untrusted functionality	SUP-004, SUP-005, SUP-011
...	68 total

How vibecheck Compares

	vibecheck	eslint-plugin-security	Snyk Code	Semgrep
Runs inside Claude Code	Yes	No	No	No
Blocks before write	Yes	No (lint after)	No (scan after)	No (scan after)
Zero config	Yes	Needs .eslintrc	Needs account	Needs rules
Supabase RLS checks	Yes	No	No	No
AI/LLM security	Yes	No	Limited	Limited
Vibe-coding aware	Yes	No	No	No
Dependencies	0 (Python stdlib)	npm	Cloud service	Binary
Latency per file	~88ms	N/A	Seconds	Seconds

vibecheck is purpose-built for AI-assisted development. It catches the specific mistakes that LLMs make when generating code.

File Structure

vibecheck/
├── .claude-plugin/plugin.json      # Plugin metadata
├── hooks/
│   ├── hooks.json                  # PreToolUse wiring
│   └── security_gate.py            # Main hook (reads stdin, dispatches, exits 0 or 2)
├── checkers/                       # 15 checker modules, one per category
��   ├── __init__.py                 # Registry + dispatcher + entropy analysis
│   ├── secrets.py                  # SEC-001 to SEC-022
│   ├── auth.py                     # AUTH-001 to AUTH-020
│   ├���─ database.py                 # DB-001 to DB-020
│   ├── network.py                  # NET-001 to NET-011
│   ├── injection.py                # INJ-001 to INJ-021
│   ├── headers.py                  # HDR-001 to HDR-011
│   ├── supply_chain.py             # SUP-001 to SUP-015
│   ├── infrastructure.py           # INF-001 to INF-010
│   ├── ai_llm.py                   # AI-001 to AI-016
│   ├── crypto.py                   # CRYPTO-001 to CRYPTO-012
│   ├── privacy.py                  # PRIV-001 to PRIV-008
│   ├─�� business_logic.py           # BIZ-001 to BIZ-006
│   ├── logging_security.py         # LOG-001 to LOG-005
│   ├── filesystem.py               # FS-001 to FS-005
���   └── websocket.py                # WS-001 to WS-004
├── skills/                         # 4 teaching skills
│   ���── security-scan/SKILL.md      # Full scan orchestrator
│   ├── security-scan/references/   # 15 detailed check docs
│   ├── secure-keys/SKILL.md        # Secret management patterns
│   ├── secure-auth/SKILL.md        # Auth hardening patterns
│   └── secure-deploy/SKILL.md      # Pre-deploy checklist
├── agents/
│   └── security-scanner.md         # Deep codebase scanner
├── LICENSE                         # MIT
└── CHANGELOG.md

Requirements

Claude Code with plugin support
Python 3.8+ (uses only stdlib; zero pip dependencies)

Troubleshooting

Hook not firing

# Check plugin is installed
claude plugin list

# Reinstall
claude plugin remove vibecheck
claude plugin add ~/.claude/plugins/vibecheck

Too many false positives

Test files are already auto-downgraded. For additional paths:

{ "ignore_paths": ["**/fixtures/**", "**/mocks/**"] }

For a specific instance:

// vibecheck-disable SEC-001
const awsKeyForTesting = "AKIA...";

To downgrade instead of block:

{ "severity_overrides": { "SEC-001": "WARNING" } }

Performance concerns

The hook only loads checkers relevant to the file extension. A .css file skips all checks. Average latency is ~88ms. If you experience slowness, check for large files being written (the hook scans the full content).

Contributing

Found a false positive? Missing a check? Open an issue.

When adding a new check:

Add the pattern to the relevant checkers/*.py module
Add a reference entry in skills/security-scan/references/
Test with both a vulnerable and clean code sample
Map to a CWE ID

Author

Built by Haralds Gabrans (GitHub)

License

MIT

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

vibecheck

The Problem

Table of Contents

Quick Start

Demo

156 Checks, 15 Categories

Secrets & API Keys (SEC)

Authentication (AUTH)

Database (DB)

Network & CORS (NET)

Injection (INJ)

Security Headers (HDR), Supply Chain (SUP), Infrastructure (INF), AI/LLM (AI), Cryptography (CRYPTO), Privacy (PRIV), Business Logic (BIZ), Logging (LOG), File System (FS), WebSocket (WS)

How It Works

Smart Detection

Configuration

Skills and Agent

Skills

Agent

CWE Mapping

How vibecheck Compares

File Structure

Requirements

Troubleshooting

Contributing

Author

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 4 Commits
.claude-plugin		.claude-plugin
agents		agents
checkers		checkers
hooks		hooks
skills		skills
CHANGELOG.md		CHANGELOG.md
LICENSE		LICENSE
README.md		README.md

Folders and files

Latest commit

History

Repository files navigation

vibecheck

The Problem

Table of Contents

Quick Start

Demo

156 Checks, 15 Categories

Secrets & API Keys (SEC)

Authentication (AUTH)

Database (DB)

Network & CORS (NET)

Injection (INJ)

Security Headers (HDR), Supply Chain (SUP), Infrastructure (INF), AI/LLM (AI), Cryptography (CRYPTO), Privacy (PRIV), Business Logic (BIZ), Logging (LOG), File System (FS), WebSocket (WS)

How It Works

Smart Detection

Configuration

Skills and Agent

Skills

Agent

CWE Mapping

How vibecheck Compares

File Structure

Requirements

Troubleshooting

Contributing

Author

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages