jevbooks

← All projects

jev-agent-skill

yuyang2230/jev-agent-skill

Offload agent classify, screen, score, and verify calls to Jev.

What it is

A Claude Code / ZCode skill that moves high-frequency small judgments out of the main model's context and into Jev, cutting token cost for classification, batch screening, scoring, and compliance checks. Aimed at AI coding agents and their users.

How it uses Jev

The skill triggers automatically on classify/route, batch-screen, score, and verify tasks. jev.py sends a state plus typed questions (choice / noul / score) to Jev via OpenCode Zen's free tier and returns typed schema results, so raw items stay in the script and only conclusions re-enter the agent context.

Primitives:choicescorenoul

Technique worth stealing

Keep raw batch items in the script and send only typed questions to Jev; return conclusions to context, not the source data.

Try it

Store a Zen key, clone the repo into the agent's skills dir, then pipe a JSON state and questions to jev.py.

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.94
  • Shows a System One patternyes0.87
  • Handles uncertaintyno0.07
  • Measuredyes0.95
  • Runnableyes0.90
  • Worth recommendingyes0.70
  • Model replicano0.03
  • Problem scopescore on a 0–2 scale1.63
  • About Jevyes0.98

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.