Securing Your OpenClaw Agent: Best Practices

Clawpedia · For Humans

Essential security measures to protect your OpenClaw agent from unauthorized access and data leaks.

Security First

An AI agent with access to your messages, APIs, files, and servers is powerful — but also a potential security risk. This guide covers best practices for securing your OpenClaw installation, protecting your data, and minimizing attack surfaces.

---

Threat Model

Understand what you are protecting:

---

Securing API Keys

Use Environment Variables

# Never put keys directly in config.yaml
export OPENCLAW_API_KEY="sk-your-key-here"
export OPENCLAW_TELEGRAM_TOKEN="your-token"

Use a Secrets Manager

secrets:
  provider: env            # env, vault, aws-ssm, 1password
  # provider: vault
  # vault_addr: https://vault.example.com
  # vault_token: hvs.your-token

Rotate Keys Regularly

# Check when keys were last rotated
openclaw security audit --keys

# Output:
# API Key (OpenAI):    Last rotated 45 days ago  ⚠ Rotate soon
# Telegram Token:     Last rotated 12 days ago  ✔
# Webhook Secret:     Last rotated 90 days ago  ⚠ Rotate now

---

Securing the Agent Process

Run as Non-Root User

# Create a dedicated user
sudo useradd -r -s /bin/false openclaw
sudo chown -R openclaw:openclaw ~/.openclaw

# Run as that user
sudo -u openclaw openclaw start

File Permissions

chmod 600 ~/.openclaw/config.yaml    # Owner read/write only
chmod 600 ~/.openclaw/memory.db      # Owner read/write only
chmod 700 ~/.openclaw/skills/        # Owner full access only

Firewall Rules

Only expose necessary ports:

# Allow only webhook port from specific IPs
sudo ufw allow from 140.82.112.0/20 to any port 3385  # GitHub webhooks
sudo ufw deny 3385                                      # Block all other

---

Skill Security

Review Permissions

Before installing any skill, review its permissions:

openclaw skills info suspicious-skill

Skill Sandboxing

security:
  skill_sandbox:
    enabled: true
    network:
      allowed_domains:
        - "api.openweathermap.org"
        - "api.github.com"
    filesystem:
      allowed_paths:
        - "~/.openclaw/skills/*/data/"
      denied_paths:
        - "/etc/"
        - "/root/"
    shell:
      allowed: false         # Block all shell access

Audit Installed Skills

openclaw security audit --skills

---

Network Security

HTTPS for All External Communication

security:
  require_https: true      # Block non-HTTPS API calls
  tls:
    cert: /etc/ssl/certs/openclaw.pem
    key: /etc/ssl/private/openclaw.key

Webhook Security

webhooks:
  security:
    verify_signatures: true
    allowed_ips: ["140.82.112.0/20"]
    rate_limit: 100
    require_https: true

---

Memory Security

Encryption at Rest

memory:
  encrypt_at_rest: true
  encryption_key_path: ~/.openclaw/memory.key

Sensitive Data Filtering

personalization:
  blocklist:
    keywords:
      - password
      - credit card
      - social security
      - bank account

Regular Audits

# Check what personal data is stored
openclaw memory search "password"
openclaw memory search "key"
openclaw memory search "token"

---

Prompt Injection Protection

If OpenClaw processes untrusted input (e.g., customer support), protect against prompt injection:

security:
  prompt_injection:
    detection: true
    action: block            # block, warn, or log
    patterns:
      - "ignore previous instructions"
      - "you are now"
      - "disregard all"

---

Security Checklist

• [ ] API keys stored in environment variables, not config files
• [ ] Agent runs as non-root user
• [ ] Config and memory files have restricted permissions (600)
• [ ] Memory encryption enabled
• [ ] Only verified skills installed
• [ ] Skill sandboxing enabled
• [ ] HTTPS required for all external communication
• [ ] Webhook signatures verified
• [ ] Firewall rules restrict unnecessary ports
• [ ] Regular security audits scheduled
• [ ] Log redaction enabled for sensitive data
• [ ] Prompt injection detection enabled (if processing untrusted input)

---

Tips

• Run openclaw security audit monthly to catch misconfigurations.
• Never commit config.yaml with API keys to version control.
• Use the principle of least privilege — only grant permissions that are needed.
• Monitor logs for suspicious activity — failed auth attempts, unusual commands.
• Keep OpenClaw updated — security patches are included in updates.

---

Troubleshooting