jev-skill-router
Claude Code plugin that scores every installed skill against every prompt with one Jev Noul each and invokes or suggests the winners.
Overview
jev-skill-router is a Claude Code plugin and an experiment in using a judgment model as the control plane for skill selection. Instead of letting the session model decide from a possibly truncated skill listing whether to use a skill, the plugin asks Jev one yes/no question per skill on every prompt and routes using thresholds set in code.
It is aimed at Claude Code users with large skill catalogs who want visible, reproducible routing. The README is candid that it is not a cost or latency saver: in a paired A/B on opus it loaded more skill text and spent more. What it offers is coverage (every description is read, at any catalog size), per-prompt control via invoke and suggest bands, exclusions and an enforcement gate, and logs of every skill's probability.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Should this skill be used for the current prompt? | — |
How it works
- User prompt is intercepted before the session model sees it
- One Noul per routed skill is sent to Jev in parallel over the full skill descriptions
- Probabilities above the invoke band are invoked, those in the suggest band are suggested
- Excluded skills are never judged; a gate can hold the model to the verdict and scores can be logged
Strengths
- Routing decision does not depend on which session model is running; the README reports identical invoke lists across models
- Every skill description is judged, avoiding the listing budget that truncates descriptions
- README reports 51 skills judged in 121-270 ms and 1,064 in about half a second
- Per-prompt probabilities are inspectable via a route CLI command and session logs
Caveats
- Costs one Noul per skill per prompt, so spend grows with catalog size
- README states it spent more time and money than stock selection in its A/B
- Scores vary by about 0.01 between identical requests, so skills near a cut can flicker
Getting started
claude /plugin marketplace add abgregs/jev-skill-router && claude /plugin install jev-skill-router@abgregs (requires Node 20.12+ and a TypeSafe API key).At a glance
- Type
- Agent skill
- Status
- beta
- Language
- TypeScript
- Runtime
- Claude Code plugin
- Has code
- yes
Use cases and integrations
How curated lists describe it
Jev-powered skill router for Claude Code — one Noul per skill, sharded in…
wh000wh000/awesome-jev-live
Frequently asked questions
What does jev-skill-router decide?
For each installed skill it asks Jev whether that skill should be used for the current prompt and gets a probability back.
Does it save tokens or money?
No. The author reports it loaded more skill text and cost more than stock selection; its value is coverage, visibility and control.
What do I need to run it?
Claude Code, Node 20.12+ on your PATH and a TypeSafe API key, which is stored in the system credential store.
Related projects
MIT plugin and local MCP server that hands an agent's closed questions to Jev or local Laya, with policy gates and receipts.
Agent skillproductionPythonClaude Code and Codex skill that routes a task to one of 27 specialist roles, with an optional Jev engine for catalog selection.
ProjectbetaPythonClaude Code hook that decides which agent or skill should handle a request using YAML rulebooks plus one Jev Choice with confidence bands.
HookbetaPythonOrchestrator that probes which coding agents are installed and logged in, then has Jev pick the best one for a task.
Agent skillbetaPythonHooks for Claude Code, Codex and pi that use Jev to suggest a skill each turn, plus a rule gate for subagent model routing.
HookbetaJavaScriptClaude Code plugin that asks Jev before each turn for reasoning effort, subagent model, strategy and skill, with a decision log.
Agent skillbetaTypeScript