domains / families / software-engineering / jev-use

jev-use

authentic

Route enumerable judgment steps - did it work, which option, how risky, is this safe to run - to the Jev judgment model through the jev_judge and jev_gate MCP tools, batched into one call per state.

When to use

• Use when the next step is a judgment over context you already have ("did X succeed?", "which option?", "how bad is this?") rather than something to write.
• Use when you are about to check the same state several times in a row, so the questions can be batched into one call.
• Use when you want a risk check on a proposed tool call before running it.
• Do not use when the step must produce new content - text, code, free-form tool arguments - or when the options cannot be enumerated. Those are structurally the model's own work.

When not to use

• Jev judges only what is in `state`; it has no view of your conversation, repository or tool history, so a thin `state` produces a thin judgment.
• State has a size ceiling (roughly 30k tokens). Beyond it the request escalates as `oversized` instead of being judged.
• Confidence is not uniform across backends: the default escalation threshold is `0.75`, but through the Vercel gateway no confidence field is returned and a top-minus-runner-up margin is used instead, with a default threshold of `0.4`.
• The advantage is narrower than an unconfigured comparison suggests. The project publishes its losing runs in [bench/RESULTS.md](https://github.com/shitianfang/jev-use/blob/main/bench/RESULTS.md): against enum-constrained baselines the late…
• This skill does not replace environment-specific validation, testing, or expert review.

Skill page

Source