jev-for-all
Experimental plugin wiring Jev into coding-agent harnesses to pick skills, tools, models and browser moves, starting with OpenCode.
Overview
jev-for-all connects Jev to the harnesses coding agents work in so they can ask which skill to load, which tools a step needs, which model to use, and how to move in a browser. OpenCode is supported now, with Claude Code and Hermes adapters planned per list summaries.
The README is explicit that it is fully experimental and not for production. It does not edit code, git history or agent settings, but it can pick models, hand work to cheaper models and add text to context, so wrong guesses may cost more tokens than they save. The author reports that on 12 small coding tasks, letting Jev pick the model cost 35 percent less than always using Sonnet with the same results, and a lean launch for simple tasks used 37 percent fewer input tokens, while stressing that the tasks and answer keys were self-written and single-run. If Jev is slow, down or unsure, the plugin does nothing. The model is chosen once from the first prompt.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which skill should the agent load? | — |
| Choice | Which tools and model does this step need? | — |
| Choice | Where should the browser go next? | — |
How it works
- Install the plugin for the harness
- On the first prompt, ask Jev for model and launch profile
- Ask which skills and tools a step needs
- Do nothing if Jev is slow, down or unsure
Strengths
- Reported cost and token savings on small tasks
- Fails safe by doing nothing
- Documents failures in WHERE-JEV-WINS.md
Caveats
- Fully experimental; thin evidence from 12 self-written tasks
- Model chosen once per session
Getting started
See docs/install.md for installation and uninstall instructions.At a glance
- Type
- Agent skill
- Status
- prototype
- Language
- TypeScript
- Runtime
- OpenCode plugin and agent skills
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Jev for every agentic development workflow — the System One decision model wired into whatever harness an agent codes in: OpenCode today, Claude Code and Hermes adapters next.
logicrw/awesome-jev-projectsJev for every agentic development workflow — the System One decision model wired into whatever harness an agent codes in: OpenCode today, Claude Code and Hermes adapters next.
hellogumbo/awesome-jevConnects Jev to the harnesses agents already code in, from OpenCode to Hermes, so they can ask which skill to load, which tools a step needs, and where to go next in a browser.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
Which skills, tools and model a coding agent should use, and browser moves.
Is it safe for production?
The author says no; it is fully experimental.
What savings were measured?
35 percent lower cost than always-Sonnet on 12 small tasks, with caveats.
Related projects
Orchestrator 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 skillbetaTypeScriptPer-turn router: one ~350 ms Jev call picks model tier, effort, tools and skill, with a hard deadline and regex fallback.
Agent skillbetaTypeScriptChooses the model tier, skills and MCP servers for each coding-agent session with two Jev requests, then launches Claude Code or Codex.
MCP serverbetaPythonRouter that puts models, subagents, skills, MCP tools and CLIs in one candidate set and asks Jev one Choice about which handles the step.
MCP serverbetaTypeScript