jevlisted

# Jev Give a judgment-heavy step a small, explicit decision space. Jev returns typed answers; the host agent remains responsible for planning, executing, and checking the result. It does not browse, generate prose, or remember earlier requests. The recipe library is inspiration, **not a fixed menu of supported functions**. Customize the evidence, questions, criteria and next consumer for the user’s task. This changes the decision interface and workflow, not the model weights. ## Bulk decisions Before hundreds of repeated judgments, use the `jev-triage` smoke_test workflow: have the host write a small task-specific paired pilot, test it, then show real IO and disagreements before scaling. The same method applies to the current workflow; do not create another skill or silently change its task. Read the [pilot guide](https://github.com/wuyoscar/jev-skill/blob/main/skills/jev-triage/references/smoke-test.md) when needed. It is a workflow parameter, not a Jev API field or a CLI scheduler. ## Choose the workflow Use natural language to describe the task; these are modes, not extra installed skills. | Task | Read or use | |---|---| | Install, configure a provider, or choose simulation | [Setup](references/setup.md); handled by the user's own coding agent | | Choose a model, tool or specialist | [Routing](references/routing.md) and [template](assets/routing.json) | | Review context retention or compaction timing | [Context](references/context.md) and [template](assets/context.