OpenClaw Security Guard

The missing security layer for your OpenClaw installation.
Audit. Monitor. Protect. All from one tool.

Quick Start • Why This Tool • Features • Dashboard • CLI • Contributing

Built by Miloud Belarebia at 2PiData — for the OpenClaw community.

Upstream integration: The core runtime guardrails from this tool are being integrated as an official OpenClaw extension — see PR #17273 and Discussion #17275. This standalone CLI tool will continue to be maintained for additional features (dashboard, secrets scanning, auto-hardening, dependency auditing).

---

The Problem

You installed OpenClaw. It works great. But ask yourself:

• Are your API keys exposed in config files or skills?
• Is your sandbox mode properly configured?
• Do your MCP servers come from verified sources?
• Could a prompt injection be hiding in your workspace files?
• Are your npm dependencies free of known vulnerabilities?

If you're not sure, you need this tool.

---

Quick Start

# Install globally
npm install -g openclaw-security-guard

# Run your first audit
openclaw-guard audit

# Fix issues automatically
openclaw-guard fix --auto

# Launch the live dashboard
openclaw-guard dashboard

30 seconds. That's all it takes to know your security posture.

---

Why Security Guard?

There are other security tools in the OpenClaw ecosystem. Here's how they compare:

	Built-in security audit	ClawSec	OpenClaw Shield	Security Guard
Secrets scanning (API keys, tokens)	Basic	Yes	Yes	15+ formats + entropy
Config hardening & auto-fix	No	Partial	Partial	Full auto-fix, 3 modes
Prompt injection detection	No	No	Basic	50+ patterns
MCP server verification	No	No	No	Allowlist-based
npm dependency scanning	No	No	No	Yes
Live web dashboard	No	No	No	Real-time scoring
API cost monitoring	No	No	No	Daily/monthly
Pre-commit hooks	No	No	No	Yes
Multi-language (EN/FR/AR)	No	No	No	Yes
Zero telemetry guaranteed	Unknown	Unknown	Unknown	100% local

Security Guard doesn't replace these tools. It fills the gaps they leave.

---

What You Get

5 Application Scanners + Infrastructure Module

Scanner	What it catches
Secrets Scanner	API keys, tokens, passwords, private keys, webhook URLs. Uses pattern matching + Shannon entropy analysis.
Config Auditor	Weak sandbox mode, open DM policy, public gateway binding, disabled rate limiting, elevated mode risks.
Prompt Injection Detector	50+ patterns: instruction overrides, role hijacking, data exfiltration, delimiter manipulation, jailbreak attempts.
Dependency Scanner	Known CVEs in your npm dependency tree.
MCP Server Auditor	Unverified MCP servers not in the community allowlist.

Infrastructure Security (opt-in)

For production deployments, add --infra to scan your server:

openclaw-guard audit --infra

Scanner	What it checks
Network Security	Firewall status (UFW/firewalld/macOS), public port exposure, gateway bound to 0.0.0.0 vs 127.0.0.1
SSH Access Control	Password authentication, root login, fail2ban status, failed login attempts (24h)
System Hardening	Security updates available, config file permissions (600 vs 777), unattended upgrades
TLS / Certificates	Reverse proxy (Caddy/Nginx), Tailscale/WireGuard VPN status
Resource Security	Disk usage (DoS risk at >90%), memory usage monitoring

The infra module catches what application scanning misses:

• App score says 85/100, but gateway is on 0.0.0.0 with SSH password auth enabled = actually vulnerable
• With --infra, the combined score reflects the real picture

Auto-Hardening

Three modes to fix issues your way:

openclaw-guard fix               # Interactive: review each fix
openclaw-guard fix --auto        # Automatic: fix everything, backup first
openclaw-guard fix --dry-run     # Preview: see what would change

Every fix creates a timestamped backup. Nothing is irreversible.

Security Score (0-100)

One number that tells you where you stand:

Score	Meaning
80-100	You're in good shape.
60-79	Some issues to review.
Below 60	Action required now.

Deductions: critical finding = -10, high = -5, medium = -2. Sandbox off = -20. Open DM policy = -15.

---

Live Dashboard

openclaw-guard dashboard

Opens a password-protected web dashboard at http://localhost:18790 with:

• Real-time security score that updates as threats are detected
• Request monitoring with requests-per-minute tracking
• Cost tracking for API usage (daily/monthly limits)
• Threat feed showing prompt injection attempts, blocked requests, rate limit hits
• Config overview at a glance

The dashboard runs on localhost only, uses PBKDF2 password hashing (100k iterations, SHA-512), and connects to your OpenClaw gateway via WebSocket.

---

CLI Reference

Command	Description
openclaw-guard audit	Full security audit
openclaw-guard audit --deep	Deep scan with entropy analysis
openclaw-guard audit --quick	Fast scan, skip advanced checks
openclaw-guard audit --infra	Include infrastructure scanning (network, SSH, system)
openclaw-guard audit --ci	CI mode (exit code 1 on critical issues)
openclaw-guard fix	Interactive fix mode
openclaw-guard fix --auto	Automatic fix with backup
openclaw-guard fix --dry-run	Preview fixes without applying
openclaw-guard dashboard	Start live dashboard
openclaw-guard dashboard -p 3000	Custom port
openclaw-guard scan secrets	Scan for secrets only
openclaw-guard scan config	Audit config only
openclaw-guard scan prompts	Detect prompt injections only
openclaw-guard report -f html	Generate HTML report
openclaw-guard report -f json	Generate JSON report
openclaw-guard hooks install	Install pre-commit hook
openclaw-guard hooks status	Check hook status

Global Options

Option	Description
-c, --config <path>	Custom config file path
-l, --lang <lang>	Language: en, fr, ar
-v, --verbose	Verbose output
-q, --quiet	No banner

---

Configuration

Create .openclaw-guard.json in your project root or home directory:

{
  "scanners": {
    "secrets": {
      "enabled": true,
      "exclude": ["*.test.js", "node_modules/**"]
    },
    "config": {
      "enabled": true,
      "strict": true
    },
    "prompts": {
      "enabled": true,
      "sensitivity": "high"
    },
    "mcpServers": {
      "allowlist": [
        "mcp-server-filesystem",
        "mcp-server-fetch",
        "mcp-server-memory",
        "mcp-server-sqlite",
        "mcp-server-git",
        "mcp-server-github"
      ],
      "blockUnknown": false
    }
  },
  "monitors": {
    "cost": {
      "dailyLimit": 10,
      "monthlyLimit": 100
    }
  },
  "dashboard": {
    "port": 18790
  }
}

---

Programmatic API

Use Security Guard as a library in your own projects:

import { quickAudit, checkPromptInjection } from 'openclaw-security-guard';

// Full audit
const results = await quickAudit('~/.openclaw');
console.log(`Score: ${results.securityScore}/100`);
console.log(`Critical: ${results.summary.critical}`);

// Check a message for injection
const check = await checkPromptInjection('ignore all previous instructions');
console.log(check.safe); // false

// Individual scanners
import { SecretsScanner, ConfigAuditor, McpServerAuditor } from 'openclaw-security-guard';

const scanner = new SecretsScanner({});
const result = await scanner.scan('/path/to/openclaw');

---

Privacy & Security

This is a security tool. It should be secure itself. Here's the commitment:

Telemetry	None. Zero.
Network requests	None (except local WebSocket to your gateway)
Data collection	None
Cloud dependency	None
Dashboard access	localhost only, password-protected
Password storage	PBKDF2, 100k iterations, SHA-512
Input validation	Zod schemas everywhere
Report generation	XSS-safe, path-traversal-safe

See SECURITY.md for the full security policy.

---

Documentation

Language	Link
English	docs/en/README.md
French	docs/fr/README.md
Arabic	docs/ar/README.md

---

Contributing

PRs welcome. See CONTRIBUTING.md.

git clone https://github.com/2pidata/openclaw-security-guard.git
cd openclaw-security-guard
npm install
npm test    # 25 tests, 0 warnings

---

License

MIT

---

Created by Miloud Belarebia at 2PiData
Inspired by detect-secrets, trivy, and OWASP.

If this tool helps you, leave a star. It helps others find it too.
github.com/2pidata/openclaw-security-guard

Report a bug • Request a feature • 2pidata.com