Jev Agent Skill

An Agent Skill for Claude Code, Codex, Cursor, and other compatible agents to use the Jev AI typed decision API.

English is the default documentation and onboarding language. Simplified Chinese is also supported with JEV_LANGUAGE=zh-CN.

On first use, the Skill guides the user through:

1. Language: English (en-US) by default, or Simplified Chinese (zh-CN);
2. API key: a Jev AI account key beginning with sk_;
3. Help: when to use Choice, Score, and Noul.

Installation

npx skills add jev-ai/jev-agent-skill

For a manual project-local installation, copy SKILL.md to .agents/skills/jev-agent/SKILL.md.

Configuration

Create an API key at the Jev AI API key page. Keep it in the local environment. Never commit it to Git or paste it into an Agent conversation.

export JEV_API_KEY="sk_your_key_here"
export JEV_LANGUAGE="en-US"

To use Chinese guidance and examples:

export JEV_LANGUAGE="zh-CN"

Optional settings:

export JEV_API_BASE_URL="https://thejevai.com"
export JEV_MODEL="typesafe/jev-1.13"

The default endpoint is:

POST https://thejevai.com/v1/systemone
Authorization: Bearer $JEV_API_KEY

Basic usage

After installation, ask the Agent:

Use the Jev Agent skill to decide which workflow should handle this task.

For setup and help:

Use the Jev Agent skill to check my configuration and show the help guide.

Jev provides typed judgments. Your application still owns business logic, permissions, human approval, and the final action.

Five Agent examples

The following prompts can be copied directly to an Agent that has this Skill. They use English because it is the default; set JEV_LANGUAGE=zh-CN to ask the Agent to explain the workflow and write questions in Chinese.

1. Route a support ticket

Use the Jev Agent skill. Classify this support ticket into exactly one team:
billing, technical, account, or sales. Return the selected team,
probabilities, and confidence. Do not contact the customer or modify any
ticket yet.

Ticket:
"I was charged twice for my annual plan and need a refund."

The Agent should use choice with the allowed teams in criteria. Application code can route using answers.route.choice; low-confidence tickets should go to human review.

2. Guard a tool call

Use the Jev Agent skill before running this proposed tool call. Judge whether
it is safe to run without additional human approval. Consider the side
effects, reversibility, scope, and the stated policy. If the result is
uncertain or risky, do not execute the tool and explain what approval is
needed.

Tool: delete_customer_records
Arguments: {"where": "last_login < 2023-01-01"}
Policy: destructive database operations require a backup and human approval.

The Agent should use noul for the safety judgment. Jev provides a risk