jgrep (kyu1204)
Semantic grep for code, diffs and CSV rows that packs 16 chunks per Jev request and prints file:line hits with CI-friendly exit codes.
Overview
jgrep (kyu1204), published on npm as jevgrep, searches for what code does rather than what it is called. It splits source files, git diffs or CSV rows into 5 to 60 line chunks, packs 16 chunks and 16 yes/no questions into one Jev request, and turns probabilities above a threshold into grep-style file:line hits. Exit codes follow grep, so an English sentence such as 'catches an error and silently ignores it' or 'endpoint with no auth check' can act as a CI lint rule, and a --diff mode lints only a change.
There is no index, embeddings or LLM round trip; the README says a whole src tree takes about 2 seconds for about a cent, and an --estimate flag shows cost first. A --tests mode selects tests a diff can affect: on five open-source repositories it picked 12% of test files while catching over 90% of the tests each commit's author changed, for $0.11 across 60 commits. jgrep init stores a TypeSafe or OpenRouter key and can teach Claude Code or Codex to use it. It is distinct from dzhng's jevgrep.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this code chunk match the description (e.g. silently swallows an error)? | — |
| Yes / no | Can this diff affect this test file? (--tests)(inferred) | — |
How it works
- Run jgrep with a description and paths, a diff or CSV
- Input is split into 5-60 line chunks
- 16 chunks per request are sent to Jev
- Chunks above threshold print as file:line hits
- Exit code gates CI; --tests selects affected tests
Strengths
- No index or embeddings
- Cost estimate before running
- Diff mode for PR gating
- Skill for Claude Code and Codex
Caveats
- Chunk boundaries can split relevant code
- Each run sends code to the hosted API
- Test-selection benchmark is the author's own
Getting started
npm i -g jevgrep && jgrep init, then jgrep "catches an error and silently ignores it" src/At a glance
- Type
- Agent skill
- Status
- beta
- Language
- TypeScript
- Runtime
- Node CLI and agent skill
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Developer tools: semantic grep that asks Jev one Noul per 5-60 line code chunk, diff hunk or CSV row (16 per request) and prints grep-style file:line hits above a threshold, so English sentences work as CI lint rules.
yibie/awesome-jevSemantic grep for code, git diffs and CSV rows: one Noul per 5-60 line chunk, 16 chunks per Jev request, grep-style file:line output and exit codes for CI lint rules written in English
AnotiaWang/awesome-decision-modelsgrep for what code does, not what it's called. Semantic code search powered by TypeSafe Jev.
logicrw/awesome-jev-projectsSemantic grep for code, git diffs and CSV rows: one Noul per 5-60 line chunk, 16 chunks per Jev request, grep-style file:line output and exit codes for CI lint rules written in English; ships an interactive init and a Claude Code / Codex skill.
cobanov/awesome-jevGrep for what code does, not what it's called. Semantic code search powered by TypeSafe Jev.
hellogumbo/awesome-jevSemantic grep that packs sixteen chunks and sixteen questions per request, with grep's exit codes and a `--diff` mode for linting a change against a rule written in English.
MrJev/awesome-jev
Frequently asked questions
Can I use it as a CI rule?
Yes, it uses grep's exit codes and has a --diff mode to check only changed code.
What does a scan cost?
The README says about a cent for a whole src tree; --estimate shows the cost first.
Is it the same as dzhng/jevgrep?
No, the README notes that is a separate code-discovery CLI.
Related projects
Six evidence-oriented Jev tools for pi, omp and any MCP client: ask about code, find files, locate ranges, check diffs and select tests.
ServicebetaTypeScriptCLI that scores every test against a git diff with Jev and emits filter arguments for vitest, jest, node, Playwright, cargo, go and rspec.
ProjectbetaTypeScript- Supercov★ 154
Coverage, security and quality CLI where Jev answers yes/no property questions per file so agents know what to fix first.
CLIproductionRust Claude Code plugin that adds Jev as an independent judge for review findings, debugging hypotheses, design options and search hits.
PluginbetaMCP server that finds relevant code with Jev Choice and Noul checks instead of grep sweeps, and audits diffs against task intent.
MCP serverprototypePythonCodex skill using the local Jev CLI to rank candidate code snippets before reproduction, root-cause analysis and a verified fix.
Agent skillprototypePython