Block unsafe agent actions and scan newly added skills with AgentGuard

Add a runtime guard that evaluates agent actions, blocks dangerous commands or secret exposure, and audits new skills before they run.

Prerequisites

Node.js, supported agent runtime such as Claude Code or OpenClaw, local skill directories and agent action hooks

Installation

Use the upstream install or setup path that matches your environment:

• npm install -g @goplus/agentguard
• git clone https://github.com/GoPlusSecurity/agentguard.git
• npm install @goplus/agentguard

Requirements and caveats from upstream:

• Requires the local OpenClaw Gateway at 127.0.0.1:18789.

• | 3 | Network Exposure | Detects dangerous ports bound to 0.0.0.0 (Redis, Docker API, MySQL, etc.), checks firewall status, flags suspicious outbound connections |
• Note: Patrol requires an OpenClaw environment. For non-OpenClaw setups, use /agentguard scan and /agentguard report for manual security checks.

Basic usage or getting-started notes:

• 8 comprehensive security checks run on a configurable schedule

• skills, and reports matches back. Run in cron / on boot.

• Optional: after one subscribe run, install an OpenClaw isolated cron job that

• Source: https://github.com/GoPlusSecurity/agentguard

• Extracted from upstream docs: https://raw.githubusercontent.com/GoPlusSecurity/agentguard/HEAD/README.md

Documentation

• https://github.com/GoPlusSecurity/agentguard

Source

• Agent Skill Exchange