SecurityNetwork Security & Checklist

Network Security: Keeping the Gateway Locked

The OpenClaw gateway is the HTTP server that handles communication between your channels, the web interface, and the agent’s core. By default, it listens on 127.0.0.1 (localhost) — meaning only your machine can access it.

Do Not Expose the Gateway

# DEFAULT (safe) - only accessible from your machine
gateway:
  host: 127.0.0.1
  port: 3000
 
# DANGEROUS - accessible from your entire network
gateway:
  host: 0.0.0.0  # DO NOT do this
  port: 3000

Binding to 0.0.0.0 means anyone on your local network (or the internet, if port-forwarded) can access your agent’s gateway. This includes:

  • The web chat interface (full agent access)
  • API endpoints (can send commands to your agent)
  • Health check and debug endpoints (can leak system info)

If You Need Remote Access

There are legitimate reasons to access your agent remotely — maybe it runs on a home server and you want to reach it from your phone. The safe way to do this:

Option 1: Tailscale (Recommended)

# Install Tailscale on your server and devices
# Your OpenClaw gateway stays on 127.0.0.1
# Access it via your Tailscale IP (e.g., 100.x.y.z:3000)

Tailscale creates an encrypted mesh VPN between your devices. The gateway never touches the public internet.

Option 2: VPN or WireGuard

Similar to Tailscale but self-hosted. Set up a WireGuard tunnel between your devices and the server running OpenClaw.

Option 3: Reverse Proxy with Auth (Advanced)

If you must expose the gateway, put it behind a reverse proxy (Nginx, Caddy) with authentication:

server {
    listen 443 ssl;
    server_name openclaw.yourdomain.com;
 
    ssl_certificate /path/to/cert.pem;
    ssl_certificate_key /path/to/key.pem;
 
    location / {
        auth_basic "OpenClaw";
        auth_basic_user_file /etc/nginx/.htpasswd;
        proxy_pass http://127.0.0.1:3000;
    }
}

Never expose the gateway directly to the internet without authentication and TLS.

Monitoring Network Activity

Keep an eye on what your agent is doing on the network:

# Watch outbound connections from the OpenClaw process
lsof -i -P -n | grep openclaw
 
# Monitor in real time
watch -n 5 'lsof -i -P -n | grep openclaw'

If you see connections to domains you do not recognize, investigate immediately. It could be a malicious skill phoning home.

Logging and Auditing

Security without visibility is guesswork. Enable logging so you can review what your agent has been doing.

What to Log

logging:
  level: info  # Options: debug, info, warn, error
  file: ~/.openclaw/logs/openclaw.log
  maxSize: 50MB
  maxFiles: 10
  logToolCalls: true      # Log every tool invocation
  logMessages: true       # Log all incoming/outgoing messages
  logApiCalls: true       # Log external API calls
  redactSecrets: true     # Mask API keys in logs

Regular Review

Make it a habit to review logs periodically:

# Check for shell command executions
grep "tool:shell" ~/.openclaw/logs/openclaw.log
 
# Check for outbound messages
grep "tool:send_message" ~/.openclaw/logs/openclaw.log
 
# Check for file writes
grep "tool:write_file" ~/.openclaw/logs/openclaw.log
 
# Look for errors or unusual activity
grep -i "error\|warn\|denied\|blocked" ~/.openclaw/logs/openclaw.log

Updates and Maintenance

Security is not a one-time setup. OpenClaw is actively developed, and new vulnerabilities are discovered regularly.

  • Keep OpenClaw updated — Security patches are released frequently
  • Review ClawHub skills after updates — A skill update could introduce malicious code
  • Rotate API keys periodically — Especially if you suspect exposure
  • Review paired contacts — Remove contacts who no longer need access
  • Check gateway binding — Make sure it has not been changed to 0.0.0.0 by a config update

The Security Checklist

Work through this checklist when setting up a new OpenClaw instance or auditing an existing one.

DM Policies and Pairing

  • Set dmPolicy: allowlist on all personal channels (iMessage, WhatsApp, Signal)
  • Set dmPolicy: approval on semi-public channels (Telegram, Discord)
  • Review your allowed contacts list — remove anyone who no longer needs access
  • Confirm that pairing is required for new contacts on sensitive channels
  • Test that unapproved contacts cannot reach your agent

Sandboxing and Isolation

  • Enable sandbox: true in your main config
  • Restrict file system access to specific directories
  • Block access to ~/.ssh, ~/.aws, ~/.config, and other sensitive directories
  • Consider running OpenClaw in Docker for stronger isolation
  • Set memory and CPU limits if using Docker

API Keys and Secrets

  • Remove all API keys from SOUL.md and other config files
  • Move API keys to environment variables or OpenClaw’s secrets manager
  • Use OAuth instead of API keys where available
  • Verify that API keys do not appear in logs (redactSecrets: true)
  • Rotate any keys that were previously in plaintext config files

Tool Permissions

  • Review which tools are enabled — disable anything you do not actively use
  • Set requireApproval: true on high-risk tools (shell, send_message, send_email)
  • Configure shell command allowlists if shell access is enabled
  • Restrict write_file to workspace directories only
  • Test that blocked commands and tools are actually blocked

ClawHub Skills

  • Review all installed skills — remove any you do not recognize or use
  • Check skill source code for suspicious patterns (outbound requests, env access, eval)
  • Verify skill authors are legitimate
  • Install new skills with --sandbox flag
  • Subscribe to OpenClaw security advisories for skill vulnerability alerts

Network Security

  • Confirm gateway is bound to 127.0.0.1 (not 0.0.0.0)
  • If remote access is needed, use Tailscale or VPN — not port forwarding
  • If using a reverse proxy, confirm TLS and authentication are configured
  • Monitor outbound network connections periodically

Logging and Auditing

  • Enable logging with logToolCalls: true and logMessages: true
  • Enable redactSecrets: true in logging config
  • Set up a schedule to review logs (weekly minimum)
  • Check for unexpected shell commands, file writes, and outbound messages

Ongoing Maintenance

  • Keep OpenClaw updated to the latest version
  • Rotate API keys on a regular schedule (quarterly minimum)
  • Review and prune paired contacts monthly
  • Re-audit installed ClawHub skills after updates
  • Test your security configuration after any major config change

What to Do If Something Goes Wrong

If you suspect your agent has been compromised:

  1. Stop the agent immediately — openclaw stop or kill the process
  2. Revoke API keys — Rotate every key the agent had access to
  3. Review logs — Check for unauthorized commands, messages, or data exfiltration
  4. Check installed skills — Look for recently installed or modified skills
  5. Review paired contacts — Remove any contacts you did not authorize
  6. Restore from backup — If files were modified, restore from a known-good backup
  7. Report to the community — If a malicious skill was involved, report it on GitHub

Next Steps

Security is the foundation. Once your agent is locked down, you are ready to connect it to the world:

ClawHub VettingOverview