A written agent skill, not code: it walks through where a system contains fuzzy semantic judgment that should become a small Jev decision primitive, and what the deterministic code around it should keep owning.
User repository on GitHub
System-architecture skill for TypeSafe AI Jev/System One — find fuzzy semantic judgment and turn it into small Choice/Score/Noul primitives.
Repo stats from the GitHub API, cached Oct 2026.
A system-architecture skill for TypeSafe AI's Jev / System One. Continuously asks: Where does this system contain fuzzy semantic judgment that should become a small Jev decision primitive? TypeSafe already publishes an official skill that gives coding agents current API/SDK context and implementation guidance. The missing layer is architecture: keep deterministic logic, control flow, and side effects in code; use Jev for narrow structured judgments; batch independent questions; and route uncertainty explicitly.
Quoted from the project's README. Read the full README.
Watch out
There is no package, no runtime and no code here, just a skill document you copy into an agent skills folder, so nothing can be benchmarked and no claim in it has been measured. It explicitly declines to cover SDK syntax, request and response fields, models, limits or version-specific behaviour, and instructs the agent not to invent API fields and to read the current TypeSafe documentation instead. It assumes your harness can load skills in the Cursor or Claude Code convention. MIT licensed. Copy reviewed against the repository README on 2026-10-04, not from running it.
PyModel/jev-judge-mcpPython
geilt/typesafe-cliPython
rashedInt32/jev-mcpTypeScript
sumleo/prompt2jevPython
BeLazy167/typesafe-modTypeScriptPosts from builders shipping with Jev right now.
