Agent tooling

jev-permission-gate

A Claude Code experiment asks Jev eight risk questions before allowing, denying, or deferring a tool call to the built-in classifier.

Bookmark Keep this in your collection.

Leave a noteWhat would you try with this?

Only you can see your notes.

Source screenshot of jev-permission-gate — SOURCE SCREENSHOTFull screenshot ↗

What it does

This is a security experiment, not a safety guarantee. It needs Claude Code mods access and sends recent requests and pending commands to TypeSafe, where inline secrets can slip through its filters.

How you can use it

Your developer can borrow the idea of checking an AI’s planned actions before letting it act. Start by recording what the extra check would allow, without changing the usual decisions. Compare the two on your own examples. Do not treat an AI’s answer as proof that an action is safe.