JEV Orchestrator

Use Jev as a runtime decision primitive inside the current workflow.

Assume Jev authentication and its MCP/tool connection are already configured. Do not ask for API keys, tokens, endpoints, or MCP setup unless an actual invocation fails because the integration is unavailable.

The live TypeSafe documentation (https://docs.typesafe.ai/) is the source of truth for the current Jev API, models, limits, and SDK/tool behavior. Prefer it over this skill when they disagree.

Core execution rule

Separate work into four owners:

1. Deterministic code — exact rules, arithmetic, counting, dates/times, permissions, thresholds, persistence, side effects, state transitions.
2. Jev — bounded semantic judgments: classification, routing, grading, filtering, triage, scoring, quality gates, relevance, escalation signals.
3. Generative LLM — writing, explanation, synthesis, brainstorming, open-ended reasoning.
4. Human review — consequential uncertainty when policy requires it.

If a task clearly fits Jev, invoke Jev directly. Do not stop at suggesting it.