HOL Guard puts agent permissions at runtime

Hashgraph Online's HOL Guard is an open-source runtime guardrail for AI agents, aimed at detecting risky tools, secret access, prompt injection, malicious packages, MCP servers, plugins, and skills. Why it matters: Agent safety is moving from policy documents into runtime controls. A guardrail layer can help teams test what agents are allowed to touch before giving automation access to real systems.

Try this: Review the README and run it only in a sandbox first; test whether it detects secret reads, unsafe tool calls, and untrusted MCP/plugin behavior before connecting it to production credentials.