security-hardening

name	security-hardening
description	Harden OpenClaw security configuration. Use when: (1) Setting up security for new OpenClaw installation, (2) Configuring exec approvals and allowlists, (3) Securing gateway access, (4) Setting up tool policies, (5) User asks about OpenClaw security or hardening.

OpenClaw Security Hardening Skill

This skill provides a comprehensive guide for securing your OpenClaw installation.

Security Checklist

Area	Configuration	Status
🌐 Gateway	Bind to loopback	Required
🔑 Auth	Token/password authentication	Required
📱 Channels	Allowlist policy	Recommended
⚡ Exec	Allowlist + approvals	Recommended
🛡️ Elevated	Allowlist only	Recommended
🧰 Tools	Deny dangerous tools	Optional

1. Gateway Security

Bind to Loopback (Required)

Never expose the gateway to public networks without authentication.

{
  "gateway": {
    "mode": "local",
    "bind": "loopback",  // Only localhost access
    "port": 18789,
    "auth": {
      "mode": "token",
      "token": "<strong-random-token>"
    }
  }
}

Generate a strong token:

openssl rand -hex 24

Remote Access (If needed)

For remote access, use Tailscale instead of exposing the gateway:

{
  "gateway": {
    "tailscale": {
      "mode": "serve"  // or "funnel" for public access
    }
  }
}

2. Channel Security

Allowlist Policy (Recommended)

Restrict who can interact with your agent:

{
  "channels": {
    "telegram": {
      "dmPolicy": "allowlist",
      "allowFrom": [123456789],  // Your Telegram user ID
      "groupPolicy": "allowlist",
      "allowGroups": []  // Specific group IDs
    },
    "whatsapp": {
      "dmPolicy": "allowlist",
      "allowFrom": ["[phone]"]
    }
  }
}

3. Exec Approvals

Configure Exec Approvals File

Create ~/.openclaw/exec-approvals.json:

{
  "version": 1,
  "defaults": {
    "security": "allowlist",
    "ask": "on-miss",
    "askFallback": "deny",
    "autoAllowSkills": true
  },
  "agents": {
    "main": {
      "security": "allowlist",
      "ask": "on-miss",
      "askFallback": "deny",
      "autoAllowSkills": true,
      "allowlist": [
        { "pattern": "/opt/homebrew/bin/*" },
        { "pattern": "/usr/bin/*" },
        { "pattern": "/bin/*" },
        { "pattern": "/usr/local/bin/*" }
      ]
    }
  }
}

Security Modes

Mode	Description
deny	Block all exec requests
allowlist	Allow only allowlisted commands
full	Allow everything (dangerous!)

Ask Modes

Mode	Description
off	Never prompt
on-miss	Prompt when not in allowlist
always	Always prompt

4. Tool Policies

Exec Tool Configuration

{
  "tools": {
    "exec": {
      "host": "sandbox",           // Default to sandbox
      "security": "allowlist",     // Require allowlist
      "ask": "on-miss",            // Prompt for unknown commands
      "safeBins": ["jq", "grep", "cat", "echo"]  // Safe stdin-only tools
    }
  }
}

Elevated Access Control

Only allow elevated access from specific users:

{
  "tools": {
    "elevated": {
      "enabled": true,
      "allowFrom": {
        "telegram": ["123456789"],
        "whatsapp": ["[phone]"]
      }
    }
  }
}

Deny Dangerous Tools

For high-security environments, deny certain tools:

{
  "agents": {
    "defaults": {
      "tools": {
        "deny": ["gateway", "browser"]
      }
    }
  }
}

5. Approval Forwarding

Forward exec approval requests to your chat channel:

{
  "approvals": {
    "exec": {
      "enabled": true,
      "mode": "both",
      "targets": [
        { "channel": "telegram", "to": "123456789" }
      ]
    }
  }
}

Approve via chat:

/approve <id> allow-once
/approve <id> allow-always
/approve <id> deny

6. Verification

Run Security Audit

openclaw security audit --deep

Check Configuration

openclaw doctor --non-interactive

Expected Output

• 0 critical issues
• No channel security warnings

Quick Setup Script

Run the included setup script:

./scripts/harden.sh

Common Issues

"Permission denied" errors

Your exec allowlist may be too restrictive. Add the needed binary paths.

Can't run commands

Check if security is set to deny. Change to allowlist.

Approval timeout

If no UI is available, requests will time out. Set askFallback appropriately.

Security Best Practices

1. Principle of least privilege - Only allow what's needed
2. Regular audits - Run openclaw security audit periodically
3. Monitor logs - Check ~/.openclaw/logs/ for suspicious activity
4. Keep updated - Run openclaw update regularly
5. Backup config - Keep your openclaw.json backed up