jev-codex-router-skill

Portable Codex skill that asks Jev to choose a model and reasoning tier for independent subtasks, with confidence gating.

Agent skillbeta

View on GitHub ↗

Overview

Jev Codex Router Skill packages model and reasoning-effort routing as a shareable Codex Skill, documented mainly in Chinese. It reads the models and reasoning tiers the local Codex host actually supports, sends a task summary and those candidates to Jev, and returns a structured selection with confidence and a decision ID. Codex then runs the subtask with the chosen configuration if the host allows explicit model selection.

On the desktop app it routes independent subtasks dispatched to native sub-agents; for per-turn routing on the macOS CLI it points to the upstream project. Results below a confidence threshold (0.65 in the validation notes) stop dispatch and leave the main agent in charge. The installer uses only the Python standard library and does not modify existing model config or hooks. Credentials can be TypeSafe or OpenRouter keys. The README reports a v1.1.0 validation with 7 models and 40 model-tier combinations discovered and 26 Windows tests passing.

What the decision model decides

Type	Question	Answers
Choice	Which model and reasoning tier should run this subtask?(inferred)	—

How it works

Discover available models and reasoning tiers from the local Codex host
Send the task summary and candidates to Jev
If the result is valid and above the confidence threshold, record an audit entry and dispatch
Otherwise stop dispatch and let the main agent continue; verify which model actually ran

Strengths

Candidates come from the real host catalogue, so new models need no manual config
Confidence threshold prevents low-confidence routing
Pure standard-library installer that leaves existing config untouched

Caveats

Desktop mode only routes subtasks, not the main session
macOS per-turn path relies on a separate upstream runtime still to be verified
Each live routing call sends task summaries to the provider and may incur cost

Getting started

Download the release ZIP and run `python install.py` in the extracted folder, then set JEV_API_KEY or TYPESAFE_API_KEY in the Codex environment.

At a glance

Type: Agent skill
Status: beta
Language: Python
Runtime: Codex skill
Has code: yes
Has tests: yes

Use cases and integrations

Classification & Routing306

Codex262

How curated lists describe it

Portable Codex Skill for Jev model and reasoning-effort routing, with safe installation and Chinese usage guides
logicrw/awesome-jev-projects
Portable Codex Skill for Jev model and reasoning-effort routing, with safe installation and Chinese usage guides.
hellogumbo/awesome-jev

Frequently asked questions

What does Jev choose?

A model and reasoning tier for a subtask, from the combinations the Codex host reports.

What if Jev is unsure?

Below the confidence threshold the skill stops dispatch and the main agent keeps working.

Which keys work?

TypeSafe (JEV_API_KEY or TYPESAFE_API_KEY) or OpenRouter (OPENROUTER_API_KEY).

Related projects

Helm
Orchestrator that probes which coding agents are installed and logged in, then has Jev pick the best one for a task.
Agent skillbetaPython
jev-agent-hooks
Hooks for Claude Code, Codex and pi that use Jev to suggest a skill each turn, plus a rule gate for subagent model routing.
HookbetaJavaScript
jev-conductor-router
MCP server and CLI where Jev decides if a coding task needs a Conductor cloud workspace, how hard it is, and which harness and model run it.
ProjectbetaTypeScript
jev-mcp
Standalone MCP server for T3 Code with one jev_route tool where Jev picks a model and reasoning-effort combination for a task.
MCP serverbetaTypeScript
loadout
Chooses the model tier, skills and MCP servers for each coding-agent session with two Jev requests, then launches Claude Code or Codex.
MCP serverbetaPython
cursor-jev
Cursor plugin whose hooks ask Jev whether subagent spawns, file reads, writes and shell commands are worth taking.
PluginbetaJavaScript