jev CLI

# Jev CLI

[中文](README.zh-CN.md)

A stateless CLI for typed semantic judgments with the Jev model (TypeSafe System One), plus an Agent Skill for using those judgments in scripts, CI, and agent workflows. Jev does not generate text: you provide a state and typed questions, it returns calibrated probabilities and typed answers. The CLI turns judgments into shell primitives — stable JSON on stdout, diagnostics on stderr, and exit codes you can branch on.

The CLI pins the model to `jev-latest`.

## Install

The skill and CLI are installed separately.

```sh
npx skills add model-clis/jev
```

Windows (preferred):

```powershell
scoop bucket add model-clis https://github.com/model-clis/homebrew-packages
scoop install model-clis/jev
```

Windows without Scoop:

```powershell
irm https://raw.githubusercontent.com/model-clis/jev/main/scripts/install.ps1 | iex
```

Apple Silicon macOS (preferred):

```sh
brew tap model-clis/packages
brew install jev
```

Linux x64 or macOS Apple Silicon without Homebrew:

```sh
curl -fsSL https://raw.githubusercontent.com/model-clis/jev/main/scripts/install.sh | sh
```

Installers fetch only GitHub Release assets, verify SHA-256, and default to `~/.local/bin`. Set `JEV_INSTALL_DIR` to change the destination or a complete tag such as `JEV_VERSION=v2026.918.0` to pin a release. Explicit installer runs may replace an existing binary; the skill never installs or upgrades silently.

## Login and usage

Configure the TypeSafe API key privately in your own terminal—never paste it into a chat:

```sh
jev login
```

The key is stored at `~/model-clis/jev/credentials.json` (mode `600` on Unix). For CI, containers, and other headless automation, set the `JEV_API_KEY` environment variable instead — it takes precedence over the stored credentials. `JEV_API_B