zcode-jev
Typed judgment harness for coding agents with gates from PRD to release; runs on JSON-forced LLMs today and Jev by swapping the backend.
Overview
zcode-jev is a TypeScript judgment layer for coding agents, built for ZCode but provider-agnostic: you vendor its src folder into any harness. Its principle is that frontier models generate, judges decide and code owns side effects. It defines pipeline gates across the development lifecycle: PRD intake, acceptance-criteria coverage, planning, pre-execution of shell commands, per-step diffs, completion (one Noul per acceptance criterion), PR routing, CI failure triage and release. Each gate's failure action is either loop back to the agent or ask a human.
It works today with any OpenAI- or Anthropic-compatible chat endpoint using JSON-forced output, which the README says takes about 7 to 15 seconds per batched call; a Jev backend is meant to replace that by changing one constructor. Judgments are logged to a directory.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this acceptance criterion satisfied by the diff? (done-gate) | yes · no |
| Yes / no | Is this shell command irreversible or high blast radius? (pre-exec-gate) | yes · no |
| Choice | Is this CI failure real or a flake? | real · flake |
How it works
- Run a gate with the relevant input (PRD, plan, command, diff).
- The backend answers typed questions.
- Thresholds yield pass, loop or human.
- Log every judgment.
Strengths
- Covers the whole lifecycle with typed gates.
- Backend-agnostic; Jev is a drop-in.
- Completion checked per acceptance criterion.
Caveats
- Default backend is a slower JSON-forced LLM.
- Must be vendored into your harness.
At a glance
- Type
- Project
- Status
- prototype
- Language
- TypeScript
- Runtime
- library (vendored source)
- Has code
- yes
Use cases and integrations
How curated lists describe it
Typed judgment layer for coding agents — gates from PRD to ship. Jev-ready, provider-agnostic.
hellogumbo/awesome-jev
Frequently asked questions
Does it need Jev?
No, it works with JSON-forced LLMs today; Jev is a backend swap.
What does done-gate check?
One yes/no per acceptance criterion against the diff.
What happens on failure?
The gate either loops back to the agent or asks a human.
Related projects
Auditable Zig terminal agent pairing Codex for coding with Jev typed preflight and postflight checks under deterministic policy gates.
CLIproductionZigClaude Code plugin with rules and checks that make agents prove their work; an optional Jev judge triages review findings.
PluginbetaMarkdown + shellSoftware-factory supervisor where Jev independently judges coding agents' completion, test sufficiency, progress and need for a human.
ProjectbetaPythonGitHub Action that asks Jev three yes/no questions about a pull request and approves it only when all clear a confidence threshold.
ProjectbetaTypeScriptHermes plugin that asks Jev to review risky plans, check claims against sources, and judge whether a task is really finished.
PluginbetaCoding-agent tools on Jev with public benchmarks: a strict command gate, a code-review router Action, and use cases.
MCP serverbetaTypeScript