HOL Guard sits between any AI agent and the tools it wants to run. Approve or deny shell commands, secret reads, and MCP changes before they execute.