What it does
The tool only reviews text rather than attachments, and it requires either a paid TypeSafe API key or a local Kev server to judge whether a message violates your custom rules.
Agent tooling
openclaw-jev-leakguard
This OpenClaw plugin checks outgoing message text against destination rules and can hold back suspected leaks. OpenClaw's delivery hooks can still send messages if a handler throws or takes too long.
Bookmark: openclaw-jev-leakguard
Keep this in your collection.
Leave a noteWhat would you try with this?
: openclaw-jev-leakguard
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
Start with the documented client-name and chat-channel rules: mark client names locally and distinguish private channels from public announcements. The plugin checks outgoing message text against its destination. Hosted Jev checks send that text to TypeSafe before deciding whether it is sensitive; use a local judge if that disclosure is unacceptable.
The plugin tries to hold messages back when the model is unavailable and stops waiting after 12 seconds. OpenClaw itself skips handlers that throw or exceed 15 seconds and sends the message, so this is not guaranteed leak prevention. It checks text, not attachments; the author's synthetic tests are not an adversarial security audit.