Why convert a working prompt instead of keeping it?

A prompt returns labels, scores, or JSON you must parse, with formatting rules mixed into judgment. The conversion splits each judgment into one atomic question, moves arithmetic and exact matching into code, and drops formatting instructions.

What does the five-part conversion package give me?

A decision contract, a split table mapping every instruction to choice, noul, score, code, keep-llm, or drop, validated request JSON, generated runnable code with thresholds in one constants block, and assumptions to confirm.

Do I need an API key just to design the decision?

No. Project-use suggestions, state and question design, code generation, lint, and cost estimates all work without a key. Paid evaluation runs only when you explicitly request it.

What logic stays in code instead of a Jev question?

Arithmetic, dates, exact matching, and formatting stay in code. Jev keeps only the narrow judgment, with every rule the code can compute moved into a constants block.

How do I install the skill into my coding agent?

Paste the install prompt from the README into Claude Code, Codex, OpenCode, Cursor or Gemini CLI; your agent copies the self-contained skills/prompt2jev folder, runs an offline dry run and reports what still needs you, usually just an API key. You can also use the Claude Code plugin or npx skills add. The optional CLI needs Python 3.10+.

What are the archetypes and how do I start from one?

There are five: classify-route, checklist-guardrail, rubric-composite, extract-select and verify-claim. Run prompt2jev template to print one, or copy the files in skills/prompt2jev/assets/. Each is a validated request to copy and edit, not a recorded model result.

Which languages can the generated script be in?

The CLI's code command writes Python with the official SDK, JavaScript with the official SDK, dependency-free Python, or a raw curl request for any other language. Pick one with --lang, and all thresholds and weights land in a single constants block with the branching filled in.

CommunityJev Agents & MCP

Prompt2jev

Agent skill plus CLI that rebuilds a plain requirement, an existing prompt, or current LLM-call code as typed Jev choice, score, and noul questions with runnable code.

See highlights

Open github.com

Stars: 6
Language: Python
License: MIT
Published by: Community
Author: sumleo
Last verified: Oct 2026

sumleo/prompt2jev

User repository on GitHub

View on GitHub

Agent skill and CLI that turn natural language, an LLM prompt, or the code that runs one into a TypeSafe Jev decision: typed state, Choice/Score/Noul questions, and a runnable script

Forks: 2
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Natural language, an LLM prompt, or the code that runs one in. A Jev decision out: typed questions and a runnable script. 23 seconds, no narration: the end-to-end example from prompt to real answers. Also in the repo as assets/prompt2jev-walkthrough.mp4. prompt2jev turns any of three inputs into a Jev request and the code around it. Natural language: a requirement stated in plain words, with no prompt written yet.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification Jev for Ranking Jev for Tool calling

Used inJev for Software & DevTools

Languages: Python

Highlights

Maps each judgment to one atomic question and moves arithmetic, dates, and exact matching into code constants.
Dependency-free CLI validates requests against the API contract and lints them before generating code.
Five copyable archetypes cover routing, guardrails, rubrics, extraction, and claim checks.
Installs project-local for Claude Code, Codex, OpenCode, Cursor, and Gemini CLI with an offline dry run.

Watch out

Generated thresholds and assumptions stay provisional until measured on labelled data, and live runs need a TypeSafe key. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-judge-mcp

MCP server over stdio that gives coding agents fourteen Jev judgment tools, turning evidence plus an enumerable question into auto, review, or escalate actions from calibrated probabilities.