It earns its place
There is a real decision loop.
- The choice repeats or sits on a critical path
- The answer is bounded before the call
- Code cannot decide it exactly
- A low-confidence path is safe
For builders using TypeSafe Jev
Where should Jev sit in your system?
Find the decisions worth making fast. Leave the rest to code, an LLM, retrieval, or a person.
A portable skill for Claude Code, Codex, and skills.sh.
State → decision → code actionJev Architect visual system
ExplainGive people the mental model first.
DiscoverFind decision loops inside real software.
DesignDefine threshold, fallback, and measurement.
The hard question first
What is Jev’s job here?
Most conversations start at the API. The useful work happens earlier. Bound the answer. Keep the action in code. Make uncertainty visible.
1. State
What would a knowledgeable person need to see?
→
2. Decision
Can it be a Noul, Choice, or Score?
→
3. Code
What happens when confidence is high, low, or missing?
First principles
Use Jev for a narrow semantic call that software can safely act on.
It is the wrong tool
The work needs something else.
- Code, math, or a lookup can decide it
- You need prose, planning, or broad research
- A wrong action cannot be reviewed or recovered
- You cannot observe whether it helped
Start with one workflow
Ask a useful first question.
Run the skill on something you already have. It should either find a safe first decision loop or tell you Jev is not the answer.
Use Jev Architect to find where Jev belongs in this workflow. Check current docs, rank the opportunities, and design a first experiment.
skills.sh
Install the portable skill.
npx skills add karanb192/jev-architect --skill jev-architect
Claude Code
Add the marketplace, then install the plugin.
/plugin marketplace add karanb192/jev-architect /plugin install jev-architect@jev-architect-marketplace