anjijava16

Jev Multiagent Retail — Development skill for Claude Code

Development community

A retail multi-agent system where TypeSafe's Jev acts as the System One decision layer inside a LangGraph orchestrator it will multiple agents using A2A Servers, Google ADK, Langgraph, Claude Agent sd.

How to install Jev Multiagent Retail

This entry records only its repository, not the path inside it, so there is no exact command to give. Open anjijava16/jev-multiagent-retail and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Jev Multiagent Retail does

A retail multi-agent system where TypeSafe's Jev acts as the System One decision layer inside a LangGraph orchestrator it will multiple agents using A2A Servers, Google ADK, Langgraph, Claude Agent sdk and Microsoft agent sdk

Alternatives in Development

  • Om Tidy — Self-maintenance pass — acts on every hygiene flag: archives completed work, groups loose clusters, splits ove 4.6k ★
  • Commerce Agents — Reference blueprint for building shopping and merchant agents with Claude 2.6k ★
  • Research Junshi — A Claude Code skill that acts as your daily 军师 (strategic research advisor) 90 ★

README

Jev from zero: a retail multi-agent system where Jev makes the decisions

This repo does two things. First it teaches TypeSafe's **Jev** model from scratch, with six small scripts you can run in a minute each. Then it builds **ShopSense**, a retail customer-service system where Jev is the decision layer of a LangGraph orchestrator that calls five sub-agents, each one backed by a different LLM (OpenAI, Anthropic, Gemini) or by plain code.

Everything is Python. Everything runs offline with no API keys (there's a mock Jev and template replies), and switches to the real thing when you set the keys.

Docs used throughout: https://docs.typesafe.ai/introduction


Contents


Part 1. What Jev actually is

Most models you've used (GPT, Claude, Gemini) are text generators. You ask a question, they write an answer. When your *code* needs a decision ("is this a refund request?", "how angry is this customer?"), you end up asking a text generator to write JSON, then parsing it, then hoping the label is spelled right and the number means something.

Jev is built the other way round. TypeSafe calls it a **System One** model: it doesn't generate text at all. You send it

  • a state: the thing being judged (a message, a document, a JSON blob of your app's data), and
  • a set of named, typed questions about that state,

and you get back typed answers with probabilities. Your code reads