jevgrep narrows a repo level by level before handing it to coding agents
jevgrep asks whether session expiry is handled in each folder, file and function, then passes only the surviving lines to agents like Claude Code and Codex.
JEVGREP MAKES JEV PLAY 20 QUESTIONS WITH YOUR REPO. YOUR CODING AGENT ONLY OPENS WHAT SURVIVES grep needs the right word. when you don't know the function name, that's exactly the word you don't have. jevgrep works the other way. you ask what the code does, "where is session expiry handled?", and jev judges relevance level by level. question → folders → files → declarations → verbatim excerpts → coding agent is it in this folder? this file? this function? branches that fail get dropped, the ones that pass get opened further. the agent gets a few files and exact lines instead of reading its way through the repo. works as a cli and a skill for claude code, codex and opencode. the readme's own benchmark: ~30% lower agent cost, 8/10 tasks solved with and without it. search sends eligible source to jev's api, so check that before pointing it at a private repo. that's the part i care about: the expensive model starts reading only after the cheap questions have narrowed the repo down to a few lines.
This post quotes another post: x.com/gippp69/status/2103394701063737504