Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
- Renamed the skill from "skill-scanner" to "moltguard" and updated branding to "MoltGuard — OpenClaw Security Guard". - Comprehensive rewrite of documentation and user instructions for clarity and onboarding. - Expanded description of protections to include prompt injection, data exfiltration, malicious commands, and behavioral risks. - Added detailed onboarding steps, enterprise enrollment, API key management, and plan comparison table. - Provided clear command reference for status, dashboard, account linking, and updating/uninstalling MoltGuard. - Clarified contact and support information for users.
What This Skill Does
MoltGuard is a security plugin for OpenClaw agents that detects prompt injection attacks, data exfiltration attempts, and malicious commands hidden in files and web content. It analyzes three risk surfaces: prompt and instruction content, agent behavior, and data exposure. Installation is automatic and a free tier provides 500 detections per day.
It intercepts threats at the agent execution layer, catching injected instructions in external files and web content before the agent acts on them.
When to use it
- Blocking hidden prompt injection attacks in email attachments
- Preventing credential leakage before sensitive data leaves the agent
- Catching malicious file deletion commands embedded in web content
- Stopping PII from being sent to external LLMs
- Detecting unauthorized API calls triggered by injected instructions
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Read and summarize this email attachment
- 1Agent reads the email file
- 2MoltGuard sends the content to Core for risk analysis
- 3Core detects a hidden prompt injection instruction inside the email body
- 4MoltGuard blocks the injected instruction from executing
- 5Agent reports the detected attack back to the user
User is alerted to the injection attempt; the malicious instruction is not followed