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

jevriel

Skill and plugin for Codex, Claude Code and other agents that helps them use Jev for bounded judgments and measure the effect.

Agent skillbetaApache-2.0
View on GitHub ↗

Overview

JEVRIEL helps an AI coding agent use TypeSafe Jev for fast structured decisions such as routing a request, ranking evidence, classifying a document or choosing the next step, and helps you build the integration and measure whether it improves the work. It installs as a native plugin for Codex or Claude Code, or as a portable Markdown skill with a local runtime and optional MCP server for any agent with file and terminal access.

There are two adoption modes. Light, the default, runs only when you ask for a pilot, a judgment or a benchmark. Deep is an explicit opt-in that adds persistent host instructions so the agent considers each workflow and uses Jev where bounded judgments fit, while exact calculations stay in code and writing stays with the LLM. The README notes Deep depends on the host following instructions and does not intercept every tool call. Jev can be reached via OpenRouter, TypeSafe directly, Cloudflare, an existing Jev MCP, a compatible endpoint or a custom adapter. The README reports measured results on 429 real documents.

What the decision model decides

TypeQuestionAnswers
ChoiceWhere should this request be routed / which step comes next?(inferred)—
ChoiceHow should this document be classified?(inferred)—
ScoreHow relevant is this piece of evidence?(inferred)—

How it works

  1. Install for Codex, Claude Code or a portable host and connect a Jev provider
  2. Choose Light (on request) or Deep (agent assesses workflows) adoption
  3. Agent identifies bounded judgments suitable for Jev
  4. Jev answers typed questions; code and the LLM handle the rest
  5. Benchmark commands compare against LLM-only workflows

Strengths

  • Works across several agent hosts and Jev providers
  • Explicit opt-in for deeper adoption
  • Built-in measurement of whether Jev helps

Caveats

  • Early release (v0.2.0)
  • Deep mode relies on the host following instructions
  • Requires a Jev provider account

Getting started

npx --yes github:thehan-co/jevriel install claude (or install codex / install portable), then npx --yes github:thehan-co/jevriel setup --provider openrouter.

At a glance

Type
Agent skill
Status
beta
Language
JavaScript
Runtime
Codex/Claude Code plugin, portable skill, CLI, MCP server
Has code
yes
License
Apache-2.0

Use cases and integrations

How curated lists describe it

  • \*\*Codex · Claude Code · Portable skill\*\* | \[Apache-2.0 code and docs\](LICENSE) | Early release

    logicrw/awesome-jev-projects

  • Give your AI JEV wings. A skill and plugin to build with TypeSafe Jev, upgrade LLM-only workflows and measure the result.

    hellogumbo/awesome-jev

Frequently asked questions

What is the difference between Light and Deep?

Light runs only when asked; Deep adds standing instructions so the agent looks for places to use Jev in every workflow.

Which Jev providers work?

OpenRouter, TypeSafe direct, Cloudflare, an existing Jev MCP, compatible HTTPS endpoints or a custom adapter.

Does it intercept every tool call?

No, the README says Deep mode does not intercept calls or require Jev for every step.

Related projects

  • jev-mcp (arunav25)

    MCP server exposing Jev as one evaluate tool, plus a harness comparing Jev's judgments with general LLMs on shared datasets.

    MCP serverbetaJavaScript
  • jev-mode (tiffygk)

    Claude Code and Codex skills plus a study course for building with Jev, including Jevaluate, which rates how well projects use Jev.

    Projectbeta
  • mcp-server-jev

    Community MCP server giving Codex, Claude and other clients two tools for typed Jev decisions: evaluate and classify.

    MCP serverbetaTypeScript
  • Jev DSH Decision Engine

    Decision plugin for agent harnesses where Jev recommends tools, skills or owners and scores output, for DeepSeek Harness and iPolloWork.

    Agent skillbetaTypeScript
  • jev-mcp

    Eval-first MCP server for Jev: prototype questions, map them over many items, and measure accuracy and thresholds on labelled data.

    MCP serverbetaTypeScript
  • decisions-judge-mcp★ 5

    MCP server exposing a judge tool for typed noul, choice and score questions over JSON state, backed by Jev.

    MCP serverbetaTypeScript