Jev Harness

A Jev-first agent loop for DeepSeek Harness: Jev reads the same assembled session and tool history as the main model and either dispatches a saved Action, asks the main model, clarifies, or ends the turn. Requests covered by verified Actions can finish with zero main-model calls.

A developer-preview personal agent harness that turns a sequence of model calls into continuing work: understand the request, choose an action, execute, inspect evidence and recover context. You pick the general model and can optionally connect a Reflex/Jev decision service.

A small, opinionated harness setup for the Pi coding agent: scope control keeps tasks to declared files, an improvement loop turns friction into tested changes, eight pinned packages are verified before install, and LANCET Nano screens shell commands.

A standard-library-only judgement primitive and harness for Jev: ask N items × K typed questions in bounded, fail-open requests, and consume answers through a closed boundary in code that refuses anything outside the set you enumerate.

elvex's Doyle Irvin walks through an internal prototype that wired Jev into the company's agent harness, from ambient classification while typing to attention filtering and risk-aware approvals; none of it is in production.

A DeepSeek Harness auto mode hosted on GitLab: full-access automation guided by TypeSafe System One at every end prompt, MIT-licensed and created in September 2026.

Kiln points its autoresearch loop at Jev for Mandarin pinyin disambiguation. The optimized harness, built from jieba segmentation and CC-CEDICT word evidence with some chained Jev calls, cuts mistakes from 316 to 147 on 61,750 decision positions, ahead of a DeepSeek reasoning baseline at $1.60 versus $11 per pass.