jev-skills (n23eos)
Fourteen opt-in Claude Code and Codex skills where Jev makes one bounded choice: which skill, model, test, file or next step.
Overview
Jev Skills (n23eos) packages fourteen installable skills plus a Python CLI. Each skill hands Jev a small, bounded choice and lets the coding agent do the work: pick one installed skill, choose a model from up to four configured sizes, choose which file or excerpt to inspect, which existing test to run first, which component to investigate for a bug, or which of a few implementation plans to pursue. Newer workflows cover citation checking, CI triage, review-comment triage, tool picking, issue next steps, exact value picking and eval gap selection.
It is designed to be advisory and fail-open. Installation does not enable network calls, automatic use is off by default and must be enabled per workflow, and private or follow-up requests bypass the network. The README explicitly says the skills do not prove better decisions, lower cost or faster work and should be evaluated on your own tasks. Results report a recommendation or fallback, the selected id, confidence and usage.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which of your installed skills should the agent read for this request?(inferred) | — |
| Choice | Which existing test target should run first for this change?(inferred) | — |
| Choice | Which configured model size fits this task? | tiny · everyday · large · hardest |
| Choice | Does this public source support, contradict or give insufficient evidence for the claim? | support · contradiction · insufficient evidence |
How it works
- Agent prepares a finite candidate set and task request
- Offline preview by default; one live Jev decision only with explicit opt-in
- Jev returns a selected candidate id with confidence, or a fallback
- Agent reads or runs the chosen item and still checks fit
Strengths
- Opt-in and fail-open; private requests never leave the machine
- Skill bodies and paths are not uploaded
- Clear outputs with route, selected id, confidence and usage accounting
Caveats
- Author states no proof of better decisions, lower cost or speed
- Requires a TypeSafe key and Python 3.10+
- Model suggestions do not switch the running conversation's model
Getting started
uv tool install 'git+https://github.com/n23eos/jev-skills.git' && jev-skills install --agent codex && jev-skills doctor --format humanAt a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- CLI / agent skills
- Has code
- yes
Use cases and integrations
How curated lists describe it
Jev-powered decision skills for Claude Code and Codex. Opt-in, advisory, fail-open.
hellogumbo/awesome-jev
Frequently asked questions
Does it make network calls automatically?
No. Installation enables nothing; live calls need explicit opt-in and automatic use must be enabled per workflow.
Does the model router switch my agent's model?
No. It only suggests one of up to four configured model IDs; delegation is up to the host.
Where does the key go?
Set TYPESAFE_API_KEY in the environment that starts your agent; `jev-skills doctor` checks it is present without printing it.
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.
HookbetaJavaScriptPi extension for Jev-based tool routing, skill discovery, typed judgments and optional model, thinking, guard and compaction modes.
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 serverbetaPythonSkill pack handing an agent's small decisions to Jev: model routing, skill choice, search, memory, compaction and computer use.
Agent skillbetaPythonOne skill and MCP server for Claude Code, Codex and Hermes where Jev gates large-model turns, routes models and guards Git steps.
MCP serverbetaPython