JevMade Sign in
← Back to experiments

Benchmarks & research

jev-sec-bench

jev-sec-bench tests Jev blindly on prompt-injection and vulnerable-code detection tasks.

Source screenshot of jev-sec-bench
SOURCE SCREENSHOTFull screenshot ↗

What it does

The Go benchmark is built on the jev-go client.

How you can use it

Before writing code, gather examples of incoming messages and code snippets that your project handles. Label which ones are safe, suspicious, or dangerous. Write down what your software is supposed to do. Sharing that purpose helps spot trick attempts much more accurately.

A developer can then adapt this testing tool to score your own incoming text. Your app can send text to receive risk numbers, blocking clear threats, sending mild risks to human review, and letting safe text through. Remember that automated scores can make mistakes, so never treat them as absolute proof.