# Config Security Scan

Scan .claude/ directory for security misconfigurations, exposed secrets, unsafe...

- Repository: https://github.com/vibeeval/vibecosystem (541 stars)
- Category: Security
- License: MIT
- Last updated: 2026-08-08
- Runs scripts: no, instructions only

## Install

Claude Code: `npx skills add vibeeval/vibecosystem --skill config-security-scan -g -a claude-code -y`
Codex: `npx skills add vibeeval/vibecosystem --skill config-security-scan -g -a codex -y`
Cursor: `npx skills add vibeeval/vibecosystem --skill config-security-scan -g -a cursor -y`
Gemini CLI: `npx skills add vibeeval/vibecosystem --skill config-security-scan -g -a gemini-cli -y`

## Description

Scan .claude/ directory for security misconfigurations, exposed secrets, unsafe permissions

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# Config Security Scan

Scan your `.claude/` directory and related configuration files for security issues. Inspired by AgentShield pattern - checks CLAUDE.md, settings.json, MCP configs, hooks, and agent definitions for misconfigurations, exposed secrets, and unsafe permissions.

## Usage

```
/config-security-scan [path]
```

Default path: `.claude/` in current project.

## What It Checks

### 1. Secrets Detection (CRITICAL)
```
- API keys, tokens, passwords in CLAUDE.md
- Hardcoded credentials in hook scripts
- Secrets in MCP server configs
- Bearer tokens in agent definitions
- .env files committed to git
```

### 2. Permission Escalation (HIGH)
```
- dangerouslySkipPermissions in settings.json
- Overly broad tool permissions (all tools for simple agents)
- MCP servers with filesystem write access
- Hooks with shell execution and no validation
- Agents with Bash tool that don't need it
```

### 3. MCP Server Security (HIGH)
```
- Unknown/untrusted MCP servers
- MCP servers with network access + filesystem access
- Missing authentication on MCP endpoints
- MCP servers running as root/admin
- Unverified npm packages in MCP configs
```

### 4. Hook Security (MEDIUM)
```
- Hooks that execute user input
- Hooks without error handling
- Hooks that modify git config
- Hooks that access external networks
- Hooks with hardcoded paths
```

### 5. Agent Definition Security (MEDIUM)
```
- Agents with unnecessary tools
- Agents with system-level Bash access
- Agent descriptions that could enable prompt injection
- Agents without clear scope boundaries
```

### 6. Configuration Hygiene (LOW)
```
- Unused MCP server configs
- Deprecated settings
- Conflicting rules
- Missing recommended security settings
```

## Scan Procedure

```bash
# Step 1: Find all config files
find .claude/ -type f \( -name "*.json" -o -name "*.md" -o -name "*.yml" -o -name "*.yaml" -o -name "*.js" -o -name "*.mjs" -o -name "*.ts" \)

The rest: https://github.com/vibeeval/vibecosystem/blob/HEAD/skills/config-security-scan/SKILL.md