HOL Guard

AI coding agents such as Claude Code, Codex and Cursor now execute commands, read files, call MCP tools and install packages on your own computer. HOL Guard adds a local-first runtime control step in front of those actions. Routine actions keep flowing; when an agent tries something that looks risky, like reading credentials, running a destructive shell command, calling an unfamiliar MCP tool or installing an unexpected dependency, Guard holds it and asks you to approve or block it, and keeps a local record of each decision. The core runtime is open source under Apache-2.0; an optional paid Guard Cloud exists for teams. HOL Guard is not a gateway, not a network firewall and not a complete prompt-injection preventer, and its scans are not a safety guarantee. Install: pipx install hol-guard && hol-guard init