What it does
Obvious credentials and destructive commands are caught locally before any network call; Jev is reserved for semantic risks that simple patterns cannot settle.
Agent tooling
Pi Jev Guard (zszz3)
A Pi extension that turns four semantic risk checks into allow, warn, confirm or block for each tool call.
Bookmark: Pi Jev Guard (zszz3)
Keep this in your collection.
Leave a noteWhat would you try with this?
: Pi Jev Guard (zszz3)
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
A developer can borrow this approach to build a safety filter for a custom chat application. They can read the project files to see how the software checks typed commands. The app then decides to allow the action or block an action marked dangerous. It can also wait for a person to confirm.
To check complex commands, the app connects to TypeSafe, an outside AI service. This requires an access key generated by TypeSafe to connect the software. Before sending questions to the outside service, the app tries to hide known passwords. However, this local check does not guarantee that every secret is removed.