jev-review

devagrawal09MIT活跃维护

基于 TypeSafe Jev 搭建的分阶段代码审查工作流与本地看板,把 AI 代码审查拆成多个可编排阶段,方便在本地跟踪和查看结果。

A staged code-review workflow and local dashboard built with TypeSafe Jev.

暂无版本698star46fork同步于 2026-11-07 07:32 UTC

怎么上手

由 README 提炼,执行前请确认
npm install
cp .env.example .env
# 在 .env 中填入 TYPESAFE_API_KEY
npm run review:changes:save -- /path/to/git/repository
npm run dashboard
硬件要求按 README 说明

README 未说明;只要求 Node.js 24+ 和 Git,模型调用走 TypeSafe API,不需要本地 GPU

下载:源码 zip·GitHub Releases(安装包)本站不镜像安装包

项目说明

由 README 自动翻译提炼

用 TypeSafe Jev 做分阶段代码审查,支持 Git diff 和全库扫描,并带本地仪表盘

基于 TypeSafe Jev 搭建的分阶段代码审查工作流与本地看板,把 AI 代码审查拆成多个可编排阶段,方便在本地跟踪和查看结果。

核心能力

  • 分阶段编排:风险矩阵、证据选择、机制分类、严重度评分、条件路由
  • 审查正确性、安全、可靠性、兼容性和测试覆盖五类问题
  • 判断测试缺口时把改动或相关测试作为上下文
  • 先挑出具体 diff hunk 或源码区域再评影响
  • 仪表盘只绑定 127.0.0.1,不提供环境文件

容易踩的坑

需要自备 TypeSafe API key 并写入 .env,否则无法调用模型;仪表盘固定监听 127.0.0.1:4317;审查结果只是提示,不构成缺陷证明,且未接入编译器诊断或静态分析器

内容由模型根据 README 生成,可能不完整或不准确,以官方文档为准。看原文 README

最近发版

0 个版本

暂无 release 数据(可能没有发版,或尚未同步)。

用户点评

登录用户的真实使用记录,可按版本和硬件筛选
共 0 条

    还没有人点评。第一个说出真实体验的人最有价值。

    README 原文

    未翻译

    Jev Review

    A small code-review workflow built with TypeSafe Jev. It can review a Git diff or scan a complete codebase, follows the strongest structured signals through focused model calls, and presents the result in a quiet local dashboard.

    Jev Review dashboard

    How It Works

    The reviewer keeps orchestration in code and uses Jev for bounded judgments:

    Noul risk matrix
      -> Choice + Score file profiles
      -> Choice evidence selection
      -> Choice mechanism classification
      -> Score severity
      -> conditional Choice reviewer routing
    
    • Exposes separate change-review and complete-codebase entry points.
    • Uses changed or related tests as context when judging test gaps.
    • Screens correctness, security, reliability, compatibility, and test coverage.
    • Selects concrete diff hunks or source regions before scoring impact.
    • Uses structured hints, counterexamples, and explicit decision boundaries.
    • Applies thresholds and workflow policy in code.
    • Shows large reports in collapsible dashboard sections.
    • Binds the dashboard to 127.0.0.1 and never serves environment files.

    Quick Start

    Requires Node.js 24+, Git, and a TypeSafe API key.

    npm install
    cp .env.example .env
    # Add TYPESAFE_API_KEY to .env
    
    # Review the current Git diff
    npm run review:changes:save -- /path/to/git/repository
    
    # Or scan every non-ignored source file under a scope
    npm run review:codebase:save -- /path/to/git/repository-or-package
    npm run dashboard
    

    Open http://127.0.0.1:4317.

    Commands

    CommandPurpose
    npm run review:changes -- <path>Print a current-diff review as JSON
    npm run review:changes:save -- <path>Save a current-diff review for the dashboard
    npm run review:codebase -- <path>Print a complete codebase scan as JSON
    npm run review:codebase:save -- <path>Save a complete codebase scan for the dashboard
    npm run dashboardStart the local dashboard
    npm run checkTypecheck, verify dependency flow, and syntax-check the dashboard client

    Architecture

    Everything lives under src/, arranged in layers that only depend downward:

    src/
      domain/      config.ts, types.ts, patch.ts   shared policy, report shapes, diff parsing
      adapters/    git.ts, repository-files.ts     change and complete-source discovery
                   report-store.ts                 atomic report save/load
      review/      changes.ts, codebase.ts          mode-specific workflows
                   *-judgments.ts, workflow.ts     Jev calls and shared staged orchestration
      cli/         review-*.ts, save-*.ts           explicit mode entry points
      dashboard/   server.ts, public/              local-only HTTP server and the plain client
    

    Imports point toward lower layers only:

    { cli, dashboard } -> review -> adapters -> domain
    

    scripts/check-dependencies.ts fails npm run check on any upward import, any import between cli and dashboard, or any cycle.

    Current Scope

    This is an experiment in composing fast typed judgments into a review workflow. It does not yet integrate compiler diagnostics, static analyzers, repository indexing, or generated explanations. Findings are review prompts, not proof of a defect.

    License

    MIT