reddit · SideProject · 10-08 20:50 52 分 ← 返回列表

Spent the last 10 days shipping a code review tool. Turns out the AI part wasn't what mattered

My team reviews a ton of PRs and we kept losing time on two dumb things: figuring out which one to open next, and missing stuff the diff doesn't show you (like who else calls the function that just changed). So I built something for it. It's called Alden and it runs locally on your machine. Quick rundown: ranks the PRs waiting on you and tells you why each one is where it is (blocking someone, been sitting for days, touches migrations, no tests, etc). tiny docs/test/patch-bump PRs get t

五维评分

付费意愿

需求真实度

个人可落地

竞争空间

新鲜度

AI 分析

痛点：开发者审 PR 时浪费大量时间在"先看哪个 PR"和"diff 之外被调用的隐藏影响"上，现有工具要么只做 AI 摘要、要么无法给出可信的优先级排序。

目标用户：每天需要审大量 PR 的工程团队和个人开发者，尤其是中小型研发团队。

证据：we kept losing time on two dumb things: figuring out which one to open next, and missing stuff the diff doesn't show you (like who else calls the function that just changed). / "this changed code is called from somewhere outside the diff" was useful 9 out of 9 times. The "possible secret" flag? 0 out of 7

变现：本地工具免费 + 团队协作/云端同步/历史分析订阅制（$8-15/人/月）；或按席位的 Team 版授权。

MVP：2 周内做一个本地 CLI/桌面工具：接 GitHub API 拉取待审 PR，用规则引擎（阻塞他人/滞留天数/涉及迁移/无测试/diff 外调用点）对 PR 排序并生成简报，带反馈按钮持续调整权重，支持在工具内提交 review。

竞品：Graphite、CodeRabbit、Greptile、Codacy、Reviewable、What The Diff

摘要：作者用 10 天做出本地代码审查工具 Alden，核心价值不在 AI 而在"PR 智能排序"和"diff 外被调用点提示"——后者在其自测中 9/9 次有用，而 AI 的"疑似密钥"提示 0/7 次有用。规则式检查在 29 个真实出问题的文件里只命中 2 个，说明该方向有明确的准确率提升空间，但目前缺乏他人付费验证。