Skip to content

jev-orc

Claude Code plugin that routes every subagent spawn through Jev to pick its model, effort and tool profile.

Projectprototype

Overview

JEV Orchestrator is a Claude Code plugin that sends every Agent spawn through Jev, which reads the task text and picks the model (haiku, sonnet or opus), the effort and the tool profile for that subagent. If Jev is unavailable, the spawn still runs with sonnet at medium effort, and /jev-orc toggles the behaviour.

Two bundled skills make the most of it. plan turns a request into a local docs/dev folder with a prompt file, a task table, execution waves and one small task file per task, each detailed enough for a cheap model. run executes the plan wave by wave, spawning one Jev-routed subagent per task in parallel, stopping dependent waves on failure, and reporting per task. The author notes most savings come from planning small mechanical tasks that Jev sends to the cheapest model. README is in English and Portuguese.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich model should this subagent use?haiku · sonnet · opus
ChoiceWhat effort level and tool profile should it get?—

How it works

  1. Optionally plan a request into small task files
  2. Each Agent spawn is intercepted by the plugin
  3. Jev reads the task and picks model, effort and tools
  4. The subagent runs; on Jev failure it defaults to sonnet at medium

Strengths

  • Cheap models handle small mechanical tasks automatically
  • Fail-open default keeps work running
  • Plan and run skills for parallel waves

Caveats

  • Savings depend on how finely tasks are planned
  • Requires a TypeSafe key
  • Personal project built for the author's needs

Getting started

/plugin marketplace add hiyoku/jev-orc then /plugin install jev-orchestrator@jev-orc, and add your Jev key.

At a glance

Type
Project
Status
prototype
Runtime
Claude Code plugin
Has code
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does Jev decide?

The model, effort and tool profile for each subagent spawn.

What if Jev is down?

The subagent still runs with sonnet at medium effort.

Where do the savings come from?

Planning work into small tasks that Jev routes to the cheapest model.

Related projects

  • Skill pack handing an agent's small decisions to Jev: model routing, skill choice, search, memory, compaction and computer use.

    Agent skillbetaPython
  • One skill and MCP server for Claude Code, Codex and Hermes where Jev gates large-model turns, routes models and guards Git steps.

    MCP serverbetaPython
  • Local gateway for coding agents that asks Jev which tool to call next and passes everything else to your usual model.

    ServicebetaTypeScript
  • Runs Claude Code on Opus 5.5 with Jev re-picking reasoning effort after every tool batch, via per-message effort so the cache survives.

    ProxybetaTypeScript
  • Fourteen opt-in Claude Code and Codex skills where Jev makes one bounded choice: which skill, model, test, file or next step.

    ProjectbetaPython
  • Claude Code plugin with three Jev tools: verbatim context compaction, per-edit model routing and per-prompt skill selection.

    ProjectbetaPython