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

jev-pipeline-skill

Agent skill and CLI that chains Jev question design: find decision calls, check fit, draft, lint, calibrate thresholds and plan a cascade.

Agent skillprototype
View on GitHub ↗

Overview

Jev Pipeline Skill is an early v0.1 workflow for designing Jev questions. It runs find, fit, draft, lint, then (once you have responses) spread, threshold and cascade. It finds LLM calls in a codebase that are really decisions, checks whether each should become a Jev question (go, go-with-guards or no-go), drafts it from 51 field-tested patterns in awesome-jev-prompts, and lints it against known design rules.

With collected responses, it checks whether the question discriminates or is constant, fits thresholds on hard cases, checks whether the right option was offered, and produces an act, pass or escalate plan where the uncertain middle band escalates to a System Two model rather than being averaged. It orchestrates existing community tools by reference rather than copying them. It is pure standard-library Python, needs no API key since live calls happen in your own code, and installs as a skill for OpenClaw, Claude Code, Codex and OpenCode.

What the decision model decides

TypeQuestionAnswers
Choicefit: should this candidate use Jev?go · go_with_guards · no_go
Choicecascade: should this answer act, pass or escalate?act · pass · escalate

How it works

  1. find LLM calls that output decisions in a repository
  2. fit-check each candidate and draft a question from patterns
  3. lint the draft
  4. After collecting responses, check spread and fit thresholds
  5. Produce an act, pass or escalate cascade plan

Strengths

  • End-to-end guidance from finding candidates to thresholds
  • No API key needed
  • Middle band explicitly escalated, never averaged

Caveats

  • Early prototype with unstable APIs
  • Depends on upstream community tool repos
  • Quality depends on the responses you collect

Getting started

npx skills add vicfei/jev-pipeline-skill, or pip install -e . and python -m jev_pipeline run examples/demo-repo

At a glance

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

Use cases and integrations

How curated lists describe it

Frequently asked questions

Does it call Jev?

No. It helps design and calibrate questions; live calls happen in your own code.

Where do draft questions come from?

From 51 CC0 patterns in awesome-jev-prompts.

How are uncertain answers handled?

The cascade sends the middle band to a System Two model rather than averaging.

Related projects

  • Building with Jev★ 155

    Agent skill for writing and improving programs that call Jev: question design, state structure, thresholds and diagnosing wrong answers.

    Agent skillproduction
  • jev-agent-toolkit

    Portable Agent Skill teaching coding agents to use Jev as a bounded judgment layer, plus an optional MCP bridge.

    MCP serverbetaTypeScript
  • jev-anything

    Independent agent skill that makes a coding agent design, build, test and tune a bounded Jev decision layer with a contract first.

    Agent skillbeta
  • 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
  • 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
  • jev-skill (marcodicesare-dev)

    Community Jev skill for Claude Code and Codex: a guide, a dependency-free Python CLI and 5 measured question sets.

    Agent skillbetaPython