What it does
The repository keeps its Jev probes and evaluation design as research material. Missing original skill listings prevent an exact rerun.
Benchmarks & research
Jevroute stopped before its planned Rust tool was built: Claude Code already chose the right skill in 69 of 75 author-tested cases.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
The repository keeps its Jev probes and evaluation design as research material. Missing original skill listings prevent an exact rerun.
Before adding an extra AI step, test how often your current tool makes the right choice. Use examples with known answers. Decide what result would make you stop building before you begin. That is the useful lesson here: the author stopped because the existing tool already did well enough.