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

jev-judge

Dependency-free CLI and Codex skill that send local state to Jev for typed judgments and return answers without echoing the state.

Agent skillbeta
View on GitHub ↗

Overview

jev-judge by ohmyjiro is a small, dependency-free Python CLI for delegating bounded judgements to Jev. You give it a JSON state and 1 to 16 named Noul, Choice or Score questions; it returns typed answers, usage and elapsed time but never echoes the state. The intended pattern for coding agents is to build the request from a local file or tool output before that material enters the agent's context, so the agent reads only the small answer.

The README is careful about claims: this can reduce agent context, but the API call is not free and it does not prove a workflow is faster. It enforces a 50 KB request guardrail, reports failures without printing state or key, and does not choose thresholds or take actions. An example asks a yes/no, a topic choice with an other option, and an actionability score about a saved link. A Codex skill file is included.

What the decision model decides

TypeQuestionAnswers
Yes / noYes/no judgement about a saved link (example)—
ChoiceWhich topic does it belong to (with an other option)?—
ScoreHow actionable is it?—

How it works

  1. Code builds a request from local files or tool output
  2. Run jev-judge on the request
  3. Agent reads only the typed answers
  4. Agent handles uncertain cases and any writing

Strengths

  • Keeps bulky evidence out of agent context
  • No dependencies
  • State and key never printed

Caveats

  • Sends state to TypeSafe; avoid confidential material
  • No thresholds or actions built in

Getting started

`install -m 755 bin/jev-judge "$HOME/.local/bin/jev-judge"`, set TYPESAFE_API_KEY, then `jev-judge examples/triage.json`.

At a glance

Type
Agent skill
Status
beta
Language
Python
Runtime
CLI and Codex skill
Has code
yes

Use cases and integrations

How curated lists describe it

  • Dependency-free CLI and Codex skill that send local state to Jev for batched typed judgments and return answers without echoing the state.

    hellogumbo/awesome-jev

Frequently asked questions

Why not echo the state?

So agents see only the small answer, not the bulky evidence.

How many questions per request?

1 to 16.

Does it act on answers?

No, thresholds and actions are left to the caller.

Related projects

  • askjev

    Unofficial hosted or local Jev MCP server giving agents a quick second opinion: it picks yes/no, scale or choice and returns probabilities.

    MCP serverbetaTypeScript
  • jev-skill-router

    Local Python MCP service that keeps the skill catalog out of the agent's context and lets Jev pick which SKILL.md files to load.

    MCP serverbetaPython
  • jev-stack

    One-command npx installer for a set of Jev-powered agent tools: context compaction, browser agent, prompt enhancer and skill router.

    ProjectbetaJavaScript
  • System One Connector

    Go CLI and single-binary MCP server giving Claude Code, Codex, Hermes and pi an evaluate tool backed by Jev, d1, CLM or Laya.

    MCP serverbetaGo
  • swift-jev

    Swift library and CLI for Jev with enum-typed Choice, Noul and Score questions, plus an agent skill for coding agents.

    CLIproductionSwift
  • jev-mcp (blakestone-x)

    Python MCP server wrapping Jev as classify, score, check, match and screen tools with confidence on every answer.

    MCP serverbetaPython