harnessjudge
Minimal web demo: paste an agent step trace and Jev judges whether to continue, retry, escalate or stop.
Overview
HarnessJudge is a minimal demo for agent builders. After an agent step, you paste the trace and Jev returns a typed judgment: ok, retry, escalate or stop.
It illustrates using a decision model as a step-level supervisor in an agent harness. The README is very short, giving only the run commands, so details of the questions and thresholds are not documented. It is aimed at agent builders who want a quick look at how a decision model could supervise an agent loop: instead of asking a large model to reflect on its own step, the trace becomes Jev's state and the answer is one of four fixed outcomes that harness code can branch on directly. No benchmark, configuration or thresholds are documented.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | After this agent step, should the harness continue, retry, escalate or stop? | ok · retry · escalate · stop |
How it works
- Paste an agent step trace
- Jev judges the step
- Display ok, retry, escalate or stop
Strengths
- Simple illustration of step-level supervision
- Fixed outcomes that harness code can branch on
Caveats
- Minimal demo with little documentation
Getting started
npm i && npm run dev.At a glance
- Type
- Demo
- Status
- prototype
- Language
- TypeScript
- Runtime
- web app
- Has code
- yes
Use cases and integrations
How curated lists describe it
Judge agent steps — ok / retry / escalate / stop via TypeSafe Jev.
hellogumbo/awesome-jev
Frequently asked questions
What does it decide?
Whether an agent step is ok, or should be retried, escalated or stopped.
How do I run it?
npm i && npm run dev.
Is it a library?
No, it is a small demo app.
Related projects
CLI where several coding agents draft implementation plans and Jev scores them, decides on cross-review and picks the finaliser.
ProjectbetaTypeScriptnpm library where Jev judges whether an agent's recent steps are stuck on the same assumption; code returns CONTINUE, WARN, REPLAN or HALT.
ProjectbetaTypeScriptAgent observability platform that finds and fixes agent failures, with an optional Jev preclassifier for conversation checks.
ProjectproductionTypeScriptOpen-source Claude Cowork alternative built on OpenCode; per list notes, its eval testkit uses Jev as a verification judge.
ProjectproductionTypeScriptWindows Codex hook that audits a conversation's recorded execution with Jev and answers yes/no questions like 'Did the tests pass?'.
ProjectbetaPowerShellClaude Code plugin and CLI for the Jev funnel: judge items in bulk, settle confident ones in code, send only uncertain ones onward.
PluginbetaPython