Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-engineering

Coding-agent tools on Jev with public benchmarks: a strict command gate, a code-review router Action, and use cases.

MCP serverbetaMIT
View on GitHub ↗

Overview

jev-engineering collects tools built on Jev, each linked to a rerunnable benchmark, including one where a built-in tool beats the author's. review-router is a GitHub Action that decides whether an AI reviewer must read a whole pull request or just new commits. On 561 commits from FastAPI, Express and Django, the README reports a filename rule sent 13 CVE fixes to quick review, while review-router sent all 13 to full review and still let 42 percent of commits skip it.

jev-gate is a strict second lock on coding-agent commands that judges only the command, never the conversation: deterministic hard rules first, then one Jev call, exposed as a Claude Code PreToolUse hook, an MCP server, a loopback service and a team policy that personal overrides can tighten but not loosen. In the author's test, Claude Code's auto mode let 8 of 12 dangerous commands run after a session claimed authorisation, and jev-gate let none. A 300-call injection kit reports blunt injections moved 0 of 30 dangerous commands but caused 10 percent false denials on safe ones, while authority framing moved 3 of 30.

What the decision model decides

TypeQuestionAnswers
Yes / noDoes this diff need a full review?(inferred)—
ChoiceShould this command be allowed, prompted or denied?(inferred)allow · ask · deny

How it works

  1. Apply deterministic hard rules to the command or diff
  2. Ask Jev one typed question about the command or what the diff adds
  3. Allow clearly safe commands, block clearly dangerous ones with a reason, prompt otherwise
  4. Route PRs to full or quick AI review

Strengths

  • Judges only the command, so the conversation cannot talk it into anything
  • Public, rerunnable benchmarks including unfavourable ones
  • Team policy cannot be loosened by personal overrides

Caveats

  • Injection kit shows some false denials and some successful authority-framing attacks
  • Intended as a second check, not a replacement for built-in approval

Getting started

Add `uses: eugeniughelbur/jev-engineering/review-router@v1` with an OpenRouter key to a workflow, or install jev-gate as a Claude Code hook.

At a glance

Type
MCP server
Status
beta
Language
TypeScript
Runtime
Claude Code hook, MCP server, GitHub Action
Has code
yes
Has tests
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Agent safety: gates coding-agent tool calls with deterministic rules first and one typed Jev call second, then publishes a rerunnable 300-call injection test showing what the gate catches and what walks past it.

    yibie/awesome-jev

  • Decision layer for coding agents: deterministic rules before any model call, then one Jev request, as a Claude Code hook, an MCP server, a loopback service and a shared team policy. Ships the 300-call injection test behind its own numbers.

    AnotiaWang/awesome-decision-models

  • A Claude Code tool-call gate. Clearly safe commands run with no permission prompt, clearly dangerous ones are blocked with the reason, and unclear ones get the normal prompt. Also ships review-router, a GitHub Action that routes AI code review to a full or quick read. Disclosure: I built and maintain it.

    logicrw/awesome-jev-projects

  • Decision layer for coding agents: deterministic hard rules, then a Jev call, exposed as a Claude Code PreToolUse hook, an MCP server, a loopback service and a shared team policy. Ships a 300-call injection kit and its results: blunt injections moved 0 of 30 dangerous commands but caused 10% false denials on safe ones, authority framing moved 3 of 30.

    cobanov/awesome-jev

  • Decision layer for coding agents: deterministic rules before any model call, then one Jev request, as a Claude Code PreToolUse hook, an MCP server, a loopback service and a team policy that personal overrides can tighten but not loosen. Ships the 300-call injection kit behind its own numbers.

    hellogumbo/awesome-jev

Frequently asked questions

What does Jev decide?

Whether an agent command is safe and whether a pull request needs a full review.

Does it replace Claude Code's auto mode?

No, the author recommends it as a second lock alongside built-in approval.

How resistant is it to injection?

The kit reports 0 of 30 moved by blunt injections and 3 of 30 by authority framing.

Related projects

  • jev-axi

    Agent-ergonomic Jev CLI and PreToolUse hook that scores shell commands for hazards and screens fetched text for injection.

    PluginbetaTypeScript
  • jev-firewall

    PreToolUse firewall for Claude Code and Codex: shell-aware rules first, then Jev returns allow, ask or block, failing closed.

    HookbetaTypeScript
  • jev-guard (leepokai)

    Security hook giving Codex, Copilot, Gemini, Cursor, pi, OpenCode and ACP an auto-mode-like guard via three Jev questions.

    LibrarybetaTypeScript
  • pi-jev-sentinel

    Jev Sentinel: guard for Pi, Claude Code and Codex that checks tool calls, tool outputs and replies, failing closed.

    PluginbetaTypeScript
  • reflex (ursuciprian)

    Pre-execution hook for coding agents that lets shell commands run, asks a human, or blocks them, with optional Jev classification.

    ProjectbetaJavaScript
  • jevc

    Compiler that turns agent rules in CLAUDE.md and similar files into testable Jev programs: typed questions plus a coded verdict.

    ProjectbetaTypeScript