jevbooks

← All projects

jev-harness

ismaelsoilet/jev-harness · Homepage

Zero-dependency System One decision gate for AI coding agents.

What it is

A deterministic-first decision layer for AI coding agents (OpenCode, Claude Code, Cursor, and others), triaging test failures, breaking doom loops, vetoing completion, governing reasoning effort, and routing models. Callable as CLI, MCP server, git/CI hook, and typed SDKs in Python, TypeScript, and Rust.

How it uses Jev

Jev makes typed decisions on the failure log or plan state: test-gate classifies the failure and sets skip_llm from a confidence threshold; abort-check scores dead-end probability to abort trajectories; route picks a model tier; verify checks evidence against criteria; reasoning-effort modulates effort. Results drive exit codes, JSON output, and SDK return values.

Primitives:choicescorenoul

Technique worth stealing

Gate trivial failures with a calibrated confidence threshold so frontier LLM calls are skipped deterministically.

Try it

npm test | jev-harness test-gate

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-centricyes0.93
  • Shows a System One patternyes0.83
  • Handles uncertaintyno0.10
  • Measuredunclear0.45
  • Runnableyes0.86
  • Worth recommendingno0.35
  • Model replicano0.26
  • Problem scopescore on a 0–2 scale1.52
  • About Jevyes0.97

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.