GitHub radar
Claude x Jev: A Cheap AI Skill Sorts Email Before Claude
A free Claude Code skill puts a tiny decision model in front of Claude: it sorts emails, leads and comments in a fraction of a second and a fraction of a cent, and Claude only reads what it is unsure about.
The authors describe Jev as a tiny AI that makes one decision at a time, in a third of a second, for a fraction of a cent, and tells you how sure it is. It is not a chat model: you do not write it a prompt and wait for a paragraph back. You hand it text and a fixed set of choices, and it hands back a pick, a probability for every choice, and a confidence number. Claude x Jev is a free Claude Code skill that puts this model in front of Claude: it sorts inbound emails, leads and comments, or checks a draft reply, and Claude only steps in where Jev itself is unsure. It ships with 13 slash commands, from sorting email and leads to checking Claude's drafts and gating a risky terminal command before it runs. On the author's own test of 216 real emails against Claude Opus 5.5, Jev matched Claude 95.5 percent of the time on the answers it was confident about (82 percent of the inbox), and the whole run cost one cent against almost five dollars for Claude - the author calls this '400 times cheaper'. On the emails where Jev was unsure, agreement was much worse, and Claude is still needed there. The model runs through the cloud OpenRouter API and cannot write a word, so remembering past emails, following company policy, or reconciling several answers into one is still Claude's job.
Why a vibe-coder should care
If an agent handles dozens of similar decisions a day - sorting emails, scoring leads, checking a draft before it ships - this gives you a cheap, fast first filter that handles the obvious cases and saves Claude's tokens for the emails that actually need reading. The gain is honest and limited: savings only show up where Jev is confident, and on the unclear cases nothing changes, Claude still reads and writes. The same model can also sit in front of a risky agent command as a fast brake that never auto-approves anything by default.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install the Claude x Jev skill for Claude Code from https://github.com/charlesdove977/claude-x-jev - follow the README (npx claude-x-jev install --with-commands), set up an OpenRouter key (ask me for it) and run /jev-classify on a couple of my test emails. Show me the summary line: how many items, how many seconds, and what it cost.
Light: a regular laptop with Python 3.8+, no GPU needed. But the service itself is not free - you need your own OpenRouter key, though a cheap one: the author says five dollars of credit covers hundreds of thousands of decisions.
Open on GitHub▌ More finds