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

typesafe-jev-plugin

Claude Code plugin that builds reusable Jev signatures for classification or scoring and runs CSV, Excel or text datasets through them.

Pluginbeta
View on GitHub ↗

Overview

typesafe-jev-plugin is a Claude Code (and Cowork) plugin with two skills. create-signature walks you interactively through defining a reusable solution signature: input mode (CSV/Excel columns or text files), the state schema, the Choice, Score or Noul questions and an output template, saved as a .sig.json file. run-dataset then calls Jev for every row or file and writes the answers back into an Excel workbook.

It is aimed at analysts who want batch labeling or scoring of tabular or text data without writing code. It needs Python 3.9+ with requests and openpyxl and a TypeSafe key. A Codex version is planned. Signatures can describe column-based inputs from CSV or Excel, or text-file analysis over a directory, and the output template controls which answer fields become columns.

What the decision model decides

TypeQuestionAnswers
ChoiceUser-defined classification question per row—
ScoreUser-defined score per row—
Yes / noUser-defined verification per rowyes · no

How it works

  1. Run /typesafe-jev-plugin:create-signature to define state, questions and outputs.
  2. Point run-dataset at a CSV, Excel file or text folder.
  3. Jev answers per row or file.
  4. Results are exported to Excel.

Strengths

  • No-code batch labeling inside Claude Code.
  • Reusable signature files.
  • Excel output for analysts.

Caveats

  • One API call per row can add up on large datasets.
  • Codex support not yet available.

Getting started

/plugin install /path/to/typesafe-jev-plugin, export TYPESAFE_API_KEY, then /typesafe-jev-plugin:create-signature

At a glance

Type
Plugin
Status
beta
Language
Python
Runtime
Claude Code plugin
Has code
yes

Use cases and integrations

How curated lists describe it

  • Jev from Typesafe.ai is a "System One" AI model that returns \*\*typed, calibrated judgments\*\* instead of generating text. You define what to classify (a Choice), score (a Score), or verify (a Noul), and Jev returns a structured answer with a probability distribution — fast, cheap, and directly consumable by code.

    logicrw/awesome-jev-projects

Frequently asked questions

What is a signature?

A .sig.json file describing the state schema, questions and output template for a task.

Which inputs are supported?

CSV, Excel and text files or directories.

Does it work in Codex?

Not yet; Codex support is planned.

Related projects

  • jev-in-codex

    Codex plugin for batch labelling JSONL records with Jev under custom policies, with Codex reviewing uncertain labels.

    MCP serverbetaTypeScript
  • btx-skill-jev-judge

    Claude Code plugin and Python CLI that hands judgments repeated over many items to Jev, with redaction, rate limits and a pilot run.

    Agent skillbetaPython
  • jevpipe

    Unix filter CLI that pipes files, lines or records to Jev for bulk yes/no, choice or score judgments so agents need not read everything.

    ProjectbetaPython
  • `classifier.dev`：建在 Jev 上的免费零样本分类服务（含 agent skill）

    classifier.dev: free zero-shot classification HTTP service built on Jev, with no API key or account and an agent skill.

    Article
  • dbt\_jev

    dbt package whose macros run Jev Choice, Noul and Score over SQL values in DuckDB and ClickHouse models.

    ProjectbetaSQL (Jinja)
  • duckdb-jev

    DuckDB extension that asks Jev about every row in SQL and returns answers as ENUM, DOUBLE or STRUCT columns.

    PluginbetaC++