What is jev-skill and what is included?

A skill for Claude Code, Codex and other coding agents, plus a standard-library Python CLI, 5 tested question sets and a guide built from 19,367 successful Jev responses. Install by cloning into ~/.claude/skills/jev or ~/.agents/skills/jev.

What did the research find?

The README opens with 'I tried to make my AI writer cheaper with Jev. The writing got worse.' The tally covers 19,367 successful Jev API responses and 90.4M metered input tokens in 4 days, including retries and reruns, as the author mapped where the model helped.

How do I make the first API call?

Run python3 "$J/tools/jev.py" ask --via typesafe --state '{"message":"I was charged twice and need help."}' --questions "$J/examples/message-routing.json". It asks which team should see a customer message first; read answers.route.choice and the probabilities for billing, account and other.

Which API routes does the CLI support?

--via typesafe uses the official endpoint with a TypeSafe key and was live tested on 28 September (HTTP 200, jev-1.13.0). Without a route flag it uses the LLM API gateway used for most experiments, which needs LLMAPI_API_KEY; --via openrouter selects a version-pinned route.

What jobs is Jev good for here?

Good candidates: route a ticket, classify a review, score a defined quality dimension, check whether a source supports a claim. Keep elsewhere: writing copy, explaining a judgment, calculating an exact value, granting permissions.

What is the operating shape it recommends?

Collect the evidence a person would inspect, ask narrow questions about one item, let code apply the answer, and send uncertain or consequential cases to review. The repo includes patterns and pitfalls references and a workflow audit prompt.

Is it TypeSafe's official skill?

No — the README says it is an independent community project, not TypeSafe's official skill, and points to typesafe-ai/skills for the official one and current provider behavior. Python 3 and an API key are enough for the CLI; there are no package dependencies.

Jev directory

CommunityJev Agents & MCP

Jev Skill for Claude Code and Codex

An agent skill for Claude Code, Codex and other coding agents, with a standard-library Python CLI, 5 tested question sets and findings from 19,367 Jev calls and 90.4M metered input tokens over 4 days.

See highlights Open github.com

Stars: 1
Language: Python
License: MIT
Reading time: 3 min
Published by: Community
Author: marcodicesare-dev
Last verified: Oct 2026

marcodicesare-dev/jev-skill

User repository on GitHub

View on GitHub

Jev skill for Claude Code and Codex: TypeSafe System One guide, Python CLI, 5 tested recipes, and findings from 19,367 Jev calls.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

Tagscommunity catalog-import logicrw agent-skill claude-code codex

Works forJev for Routing Jev for Classification Jev for Verification

Used inJev for Software & DevTools Jev for Customer Support

Languages: Python
Platforms: CLI
Capabilities: Open sourceRequires API key

Highlights

The guide is titled The Ultimate Guide to Jev (After 19,367 Calls and 90M Tokens): 19,367 successful responses and 90.4M metered input tokens in 4 days.
The README opens with 'I tried to make my AI writer cheaper with Jev. The writing got worse.' Then it kept testing to learn where the model helped.
The first-call example routes a customer message and was live tested via --via typesafe: HTTP 200 on jev-1.13.0, 28 September, per the README.
It is an independent community project, explicitly not TypeSafe's official skill; the CLI is standard-library Python with no package dependencies.
Its job table separates good candidates (routing, classification, defined scoring, source checks) from work better kept elsewhere.

Quickstart

bash

git clone https://github.com/marcodicesare-dev/jev-skill.git ~/.claude/skills/jev
export TYPESAFE_API_KEY=...

Watch out

Copy reviewed against the truncated README on 2026-10-15; the 5 field guides section cuts off mid-sentence, so only the summary of the recipes is covered. The 19,367-call tally is the author's own log rather than an independent benchmark.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

YouTube: Ray Amjad puts Jev inside an agentic coding loop and looks at what it costs to run.

YouTube: RoboNuggets puts Jev to work inside Claude Code.

Reddit: A Claude Code mod replaces the auto-mode classifier with Jev, 2x faster.

From the Jev guides

Original write-ups on what this entry is for.

MCP tool routing: Jev in front of the tool call

Oct 16, 2026

Tool menus can outgrow the context that reads them. The catalog's Jev-based routers and gates narrow the list: benchmarked tool search, per-tool scoring, fail-closed approvals, and WebMCP in the browser.

Jev Agents & MCP#mcp#tool-routing#agents

Read article

Approval policies for coding agents: hooks, guards, and gateways