jevlint
Tree-sitter unit linter: plain-language rules judged by Jev per function, type, or statement, with packs, cache, and Release binaries.
Overview
jevlint (github.com/codegirl-007/jevlint) is a Go CLI that uses Tree-sitter to extract functions, types, comments, fields, or statements, then asks Jev whether each unit satisfies rules in jevlint.json. Commands cover init, doctor, check, eval, and plugin pack management. README documents TypeSafe as the default provider plus Clef on Cloudflare Workers AI and OpenRouter-hosted Jev. Prebuilt binaries ship on GitHub Releases for linux, macOS, and windows on amd64 and arm64.
Problem: Style and architecture rules that need human judgment still sit in review comments because regex linters cannot read intent across languages.
Built for: Teams who want CI-friendly semantic linting with plain-language rules, cached Jev calls, and optional rule packs without sending chat history.
First indexed on Jev Directory: 2026-10-04
Creator and team
- Name
- codegirl-007
- Handle
- @codegirl-007
How Jev is used
- Role in the product flow
- Per code-unit Noul-style pass or fail judgments on natural-language rule descriptions, with optional localize second pass
- Primitives
- Noul
- State in
- Extracted source for one Tree-sitter unit plus rule text, exceptions, and optional callee context (up to 12 project-local callees).
- Decision out
- pass, fail, skip, or abstain per unit with confidence; failures below minConfidence stay hidden; localize reruns focused regions on fail.
- jevlint init detects languages and writes jevlint.json
- check walks include globs, batches applicable rules per unit (default concurrency 4)
- Failed function or type rules may trigger localize passes up to 24 regions
- Results cache in the OS user cache keyed by endpoint, model, credential fingerprint, and request body
jevlint sits between classic linters and agent guardrails. Abide compiles AGENTS.md rubrics on every edit diff; jev-codes scores hunks against YAML standards packs; grev filters arbitrary line streams on the shell. jevlint instead targets repository taste with Tree-sitter precision across 12 language presets, rule packs shared like owner/name modules, and jevlint eval fixtures that require explicit pass or fail outcomes. Clef and OpenRouter paths reuse the same System One request shape, so switching providers does not rewrite rules. Use uehaj-jev-semgrep when you only need line-level meaning grep, or aurorainfra-grev when the job is Unix filters on logs and CSVs.
Sourced performance claims
- README: default concurrency 4 parallel Jev requests; network failures retry up to 2 times.Source: github.com/codegirl-007/jevlint README How it works
- Public GitHub repo codegirl-007/jevlint had about 125 stars when this listing was drafted.Source: GitHub star count October 2026
Features and stack
Features
- init, doctor, check, eval, plugin, and version commands with JSON output flags
- Rule packs with pack.json, shared rules.json, and bundled eval fixtures
- OS-level result cache with --refresh-cache and --clear-cache
- --changed flag limits checks to git-modified files
- GitHub Releases binaries plus go install from source (Go 1.26+, CGO)
Stack
- Go
- Tree-sitter grammars
- TypeSafe Jev
- Optional Clef or OpenRouter Jev
Pricing: Open source CLI; Jev, Clef, or OpenRouter usage bills per your API keys. README never stores credentials in the project tree.
Links
FAQ
- Which languages are supported?
- README lists presets c, cpp, csharp, go, java, javascript, kotlin, php, python, ruby, rust, tsx, and typescript with native grammars compiled into the binary.
- How is jevlint different from Abide?
- Abide hooks agent edits against instruction-file rubrics. jevlint is a batch linter you run with jevlint check on repository paths.
- Does check send my whole repo to Jev?
- README states Jevlint sends extracted code units and file metadata to the configured provider per rule batch, not full chat logs.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
- Jev vs LLM classification
When to gate with System One probabilities instead of asking a chat model to label things.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.