a-Fig

Jev Score — Development skill for Claude Code

Development community

Local-first document evaluation workspaces powered by Jev.

How to install Jev Score

This entry records only its repository, not the path inside it, so there is no exact command to give. Open a-Fig/jev-score and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Jev Score does

Local-first document evaluation workspaces powered by Jev.

Alternatives in Development

IPolloWork — Enterprise-grade, local-first Agent Workbench for people and agent teams 4.9k ★
Claude Prism — An offline-first scientific writing workspace powered by Claude 1.1k ★
Alethe Agents — A local-first desktop workspace for running, organizing, and resuming multiple coding agents and shells with r 511 ★

README

Jev Score

[![CI](https://github.com/a-Fig/jev-score/actions/workflows/ci.yml/badge.svg)](https://github.com/a-Fig/jev-score/actions/workflows/ci.yml) [![Release](https://img.shields.io/github/v/release/a-Fig/jev-score)](https://github.com/a-Fig/jev-score/releases/latest) [![Node.js](https://img.shields.io/badge/Node.js-%E2%89%A522.13-339933)](https://nodejs.org/) [![MIT License](https://img.shields.io/badge/license-MIT-blue.svg)](./LICENSE)

Give your coding agent a scoreboard

Most AI writing loops run on vibes: draft, rewrite, hope. **Jev Score is a test suite for writing.** It scores every revision against the context the document must satisfy and the questions that define "good", and keeps the history so you can see whether draft 8 actually beats draft 3.

It is a local CLI and web app for developers who already let Codex or Claude Code edit their text: a resume against a job posting, an essay against its prompt, a spec against requirements, a landing page against its positioning. [TypeSafe's Jev model](https://www.typesafe.ai/) does the scoring through OpenRouter; drafts and scores stay in local SQLite.

Numbers per question, not vibes. On the bundled resume example, one agent revision aimed at the two weakest questions moved the median score from 83.2 to 96.6 (the run).
Every run is kept, shown as medians and ranges, because Jev is probabilistic.
CLI for your agent, UI for you. Both read one database: JSON out of the CLI; in the browser, a progress chart and side-by-side or diff comparison of any two drafts.

![Jev Score workspace: progress chart, current leader, and per-question scores for three README drafts](./docs/assets/workspace.png)

**Jump to:** [Install](#install) · [First score](#your-first-score) · [Concepts](#concepts) · [Agent loop](#the-agent-loop) · [Limits](#honest-limits)