jevbooks

← All projects

jev-gate

eugeniughelbur/jev-gate · Homepage

Safety gate for AI coding agents, backed by TypeSafe's Jev.

What it is

A tool-call gate for Claude Code, Codex, Cursor and other AI coding agents. It answers allow, ask or deny in about 400 milliseconds for two hundredths of a cent, so every action can be checked instead of approving everything or nothing.

How it uses Jev

Jev answers two typed questions in parallel about the proposed tool call, returning a destructive probability and confidence. The gate compares these to thresholds to produce allow, ask or deny. Hard rules and a read-only allowlist run before Jev.

Primitives:score

Technique worth stealing

Hard rules run before the model, because the model was least reliable where a regex is easiest to write.

Try it

claude plugin marketplace add eugeniughelbur/jev-gate; claude plugin install jev-gate@jev-gate

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.89
  • Shows a System One patternyes0.79
  • Handles uncertaintyno0.13
  • Measuredyes0.69
  • Runnableyes0.93
  • Worth recommendingno0.32
  • Model replicano0.05
  • Problem scopescore on a 0–2 scale1.74
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.