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

jev-agent-skill

Portuguese-language agent skill for millisecond Jev decisions via OpenRouter, with validation, an MCP mode and calibration evals.

Agent skillbeta
View on GitHub ↗

Overview

This jev-agent-skill, documented in Portuguese, is an Agent Skill for fast decisions with Jev through OpenRouter. State and typed noul, choice or score questions go in; decisions with calibrated probabilities and action bands come out, without text generation. It has zero dependencies on Node 20+, using keep-alive HTTPS, and performs heavy validation before and after each call based on the model's documented concepts such as atomicity, jaggedness, cardinality, budget and calibration.

An MCP serve mode keeps a warm TLS socket for agent loops, and an eval command computes accuracy and ECE over labelled cases. The README reports a prompt-injection evaluation of eight scenarios, four clean and four with injected text mid-prompt, where a single noul call got 8/8 correct with mean confidence 0.97 and ECE 0.029. A worked example catches an injected note in a ticket in one 294 ms call costing $0.00003 while still answering the legitimate urgency question and flagging the need for a human. A selftest runs 29 offline checks.

What the decision model decides

TypeQuestionAnswers
Yes / noDoes this text contain an injection?yes · no
ScoreHow urgent is this ticket, ignoring any injected instructions?—
Yes / noDoes this need a human?yes · no

How it works

  1. Validate the request against model concepts
  2. Send to Jev via OpenRouter
  3. Map probabilities to action bands (e.g. AUTO)
  4. Validate the response; optionally serve over MCP

Strengths

  • Validation catches poorly formed questions before spending
  • Measurable calibration with the eval command
  • Warm-socket MCP mode for low latency

Caveats

  • Small injection eval (8 scenarios)
  • Documentation in Portuguese

Getting started

`export OPENROUTER_API_KEY=...`, `node scripts/jev.mjs selftest`, then `node scripts/jev.mjs ask --request-file examples/requests/ticket-triage.json`.

At a glance

Type
Agent skill
Status
beta
Language
JavaScript
Runtime
Node agent skill, MCP server
Has code
yes
Has tests
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does the skill validate?

Requests and responses against Jev concepts like atomicity, jaggedness, cardinality, budget and calibration.

How well does it detect injection?

The README reports 8/8 on its eight-scenario eval with ECE 0.029.

Does it need TypeSafe directly?

No, it uses Jev through OpenRouter.

Related projects

  • SkillRanker★ 134

    Rust CLI that ranks which agent skills fit the next step from live session context with Jev, can abstain, and hooks into Claude Code.

    CLIbetaRust
  • augustus

    Agent skills for placing, building, evaluating and improving decision-model systems, with Jev as the default hosted example.

    Agent skillbeta
  • better-call-jev

    Claude Code plugin that sends judgment calls to Jev, acts at 0.8+ for yes or 0.2- for no, and gathers more evidence in between.

    Agent skillbetaJavaScript
  • Skill Scanner

    Cisco's security scanner for agent skills, with a System One analyzer as an advisory tier that cannot emit findings or change severity.

    Agent skillproductionPython
  • jev (shaharia-lab)

    Rust CLI for Jev: ask yes/no, choice and rubric questions about text and branch on exit codes, JSON or MCP tools.

    MCP serverbetaRust
  • jev-agent-authorization

    Starter kit that gates every MCP tool call with Kinde identity checks plus Jev risk signals, then allows, holds for approval or stops it.

    MCP serverbetaTypeScript