skill-firewall

What This Skill Does

The skill-firewall is an essential security layer for OpenClaw users who interact with external, third-party skills. It acts as a mandatory sanitization gateway, ensuring that any code or logic imported from untrusted repositories (like ClawHub, GitHub, or public scripts) is thoroughly neutralized before it ever touches your local environment. Rather than blindly executing or copying external code—which is a primary vector for prompt injection and supply chain attacks—this skill mandates a 'read-and-rewrite' policy. It decomposes the external skill into its core functional requirements and discards the implementation, forcing the AI to draft a fresh, clean, and transparent version of the tool. This effectively strips away hidden HTML injection attacks, malicious Unicode obfuscation, and unauthorized remote-execution calls.

Installation

To ensure your agent is protected, you should initialize this skill by adding it to your agent’s local configuration path. Once activated, the skill-firewall will automatically intercept any command referencing external skill imports. No complex setup is required; it acts as an agent middleware that triggers whenever the user attempts to add an external component to their library.

Use Cases

Use this skill whenever you are browsing repositories like ClawHub and want to test a tool developed by a community member. It is particularly critical when the skill promises functionality that requires network access, system configuration, or file system modifications. Whether you are adding a data-scraping tool or a productivity workflow, the skill-firewall provides the necessary audit trail for human oversight, ensuring you know exactly what the code does before it runs.

Example Prompts

1. "I found this cool task-automation skill on ClawHub, can you install it for me?"
2. "Please add this external script from GitHub to my current workspace to help me organize files."
3. "OpenClaw, I want to use the latest trending skill for web analysis I saw on the forums, here is the link."

Tips & Limitations

Always review the 'Skill Firewall Report' generated by the agent. While the firewall prevents automated injection, it relies on the agent's ability to correctly identify the legitimate purpose of the code. If you find the rewritten version confusing, ask the agent to explain the logic in plain, non-technical language. Remember: the firewall is a defense-in-depth measure, not a silver bullet. Always exercise caution when granting administrative permissions or network access to any skill, even those that have been rewritten.