Agentguard

What This Skill Does

AgentGuard, developed by manas-io-ai, serves as a robust security and monitoring layer for OpenClaw agents. It provides real-time surveillance of an agent's interactions with the file system, network, and external APIs. By employing ML-lite behavioral analysis, AgentGuard establishes a baseline of 'normal' agent activity and identifies deviations that could indicate malicious intent, system compromise, or unauthorized data exfiltration. The skill maintains comprehensive audit logs, offers automated alert triggers for suspicious events, and generates actionable daily reports, ensuring your agent environment remains secure and compliant.

Installation

You can integrate AgentGuard into your OpenClaw environment by executing the following command in your terminal: clawhub install openclaw/skills/skills/manas-io-ai/agentguard Once installed, customize the configuration file located at config/agentguard.yaml to define your specific monitoring directories, trusted API domains, and alert channels.

Use Cases

AgentGuard is ideal for developers and enterprises building autonomous agents that handle sensitive data. It is particularly effective for:

• Protecting API keys and environment variables from accidental exposure.
• Monitoring agent activity in production environments where compliance and audit trails are mandatory.
• Preventing unauthorized file exfiltration from sensitive local directories.
• Detecting potential supply chain attacks or prompt injections that trigger unknown API calls.

Example Prompts

1. "AgentGuard, please run a security audit for today and generate a summary of any detected API anomalies."
2. "AgentGuard, start monitoring the /production/data directory for any unexpected write operations."
3. "AgentGuard, check the current status of file system monitoring and report any high-sensitivity alerts from the last 24 hours."

Tips & Limitations

• Sensitivity Tuning: Start with 'medium' sensitivity to avoid alert fatigue, adjusting to 'high' only when strictly necessary.
• Trusted Domains: Always populate your trusted_domains list to minimize false positives from legitimate background services.
• Performance: While lightweight, continuous monitoring of thousands of files may impact performance; consider using exclude_patterns to ignore large log files or node_modules folders.
• Note: AgentGuard is a monitoring tool; while it can block calls if configured to do so, it should be paired with strong host-level firewall and permission settings for maximum security.