Config security scan

Skill vibeeval/vibecosystem/skills/config-security-scan

AI software team for Claude Code - 138 agents, 295 skills, 73 hooks. Self-learning, multi-agent swarm, autonomous skill evolution.

Install

npx -y skills add vibeeval/vibecosystem --skill config-security-scan

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

What its author says it does

Copied from the file, not written here

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

SKILL.md

3.4 KB, as published. Nobody here has run it

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

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

# Step 2: Secret patterns
grep -rn "api[_-]?key\|password\|secret\|token\|bearer\|sk-\|pk_\|ghp_\|gho_\|xoxb-\|xoxp-" .claude/

# Step 3: Permission checks
grep -rn "dangerouslySkipPermissions\|allowedTools.*Bash\|shell_exec\|eval(" .claude/

# Step 4: MCP config review
cat .mcp.json 2>/dev/null | jq '.mcpServers | keys'

# Step 5: Hook review
ls .claude/hooks/ 2>/dev/null

Output Format

# Config Security Scan Report
Scanned: [path]
Date: [timestamp]

## Summary
- CRITICAL: X issues
- HIGH: Y issues
- MEDIUM: Z issues
- LOW: W issues

## CRITICAL Issues

### [Issue Title]
**File:** [path]
**Line:** [number]
**Issue:** [description]
**Fix:** [remediation]

## Recommendations
1. [Action item]

Hard Exclusion List (Skip These)

These are NOT security issues in the .claude/ context:

Environment variable references (not actual values)
Test/example credentials clearly marked as such
Public API keys meant to be public
SHA hashes used as identifiers
Base64-encoded non-secret data
localhost/127.0.0.1 URLs

Integration

security-reviewer: Calls this skill during security audits
verifier: Includes config scan in pre-commit checks
shipper: Runs before deployments