Claw Permission Firewall

What This Skill Does

The Claw Permission Firewall is an essential security middleware for the OpenClaw agent ecosystem. Unlike perimeter firewalls that focus on network entry, this tool acts as a runtime least-privilege monitor for individual agent actions. It intercepts requested operations—such as HTTP requests, file system interactions, or shell commands—and evaluates them against a configurable policy engine. By categorizing requests into ALLOW, DENY, or NEED_CONFIRMATION, the firewall effectively mitigates risks such as prompt-injection attacks that attempt to exfiltrate secrets, unauthorized access to sensitive local files (e.g., .ssh or .aws directories), and dangerous command executions. It further protects the user by returning a sanitized version of the action with sensitive headers and tokens redacted.

Installation

To integrate the firewall into your agent workflow, use the following command within your terminal or OpenClaw CLI:

clawhub install openclaw/skills/skills/bharathjanumpally/claw-permission-firewall

Once installed, ensure your local policy.yaml file is configured to match the security requirements of your specific workspace, as the firewall operates on a principle of default-deny for risky behaviors like shell execution.

Use Cases

• Exfiltration Prevention: Blocks requests made to known data-stealing domains or untrusted pastebin services.
• Sensitive File Protection: Restricts agent file-system access strictly to the defined workspaceRoot, preventing unauthorized reads of system configuration files.
• Prompt Injection Defense: Automatically detects and redacts sensitive credentials like API keys or Authorization headers from outgoing requests even if a prompt attempts to reveal them.
• Safe Execution Gatekeeping: Requires human verification before any destructive shell command, such as mass file deletions, can be processed by the agent.

Example Prompts

1. "Evaluate this file read request for the report folder: { 'type': 'file_read', 'path': './reports/data.json' }"
2. "Check if this HTTP POST request to the external GitHub API is safe to execute: { 'type': 'http_request', 'method': 'POST', 'url': 'https://api.github.com/v1/update' }"
3. "Should I allow this command? { 'type': 'exec', 'command': 'ls -la' }"

Tips & Limitations

This tool is designed to complement existing infrastructure security, not replace it. Always maintain a well-defined workspaceRoot to ensure the jail-based file protection is effective. Be aware that the NEED_CONFIRMATION workflow requires your agent's controller or UI to handle the interaction loop properly. If the firewall returns a high riskScore, treat the output with caution even if it is not explicitly denied.