reddit · SideProject · 10-14 02:50 53 分 ← 返回列表

I built a tool because my colleagues were getting sick of reviewing all my PRs

I've been using coding agents a lot recently, and after setting up a dedicated machine to run them remotely, I started sending my colleagues way too many PRs. Understandably, they weren't quite as excited about that as I was. I wanted to make reviewing them less work. Having agents attach browser recordings helped, and people liked them, but getting the agents to run through everything and record it took ages on my machine and burned loads of tokens. So I built Diff Stage. O

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：AI 编码代理批量产出 PR，代码评审成了新的交付瓶颈，评审者需要花大量时间才能判断分支相对 main 的行为变化。

目标用户：大量使用 AI coding agent 的工程团队，尤其是用 Laravel/Pest/Playwright 等做浏览器端测试、PR 量激增的中小团队。

证据："I started sending my colleagues way too many PRs. Understandably, they weren't quite as excited about that as I was." / "Has writing code faster just moved the bottleneck onto whoever reviews it?" / "having agents attach browser recordings helped, and people liked them, but getting the agents to run through everything and record it took ages on my machine and burned loads of tokens" / "I've now spent far more tokens building it than I'll ever save."

变现：GitHub App + SaaS 订阅（按仓库数或评审席位收费，如 $15-30/开发者/月），或按录制分钟数计量的用量套餐；免费版限制并发录制数。

MVP：两周 MVP：做一个 GitHub App，在 CI 中用 Playwright 录制指定浏览器测试的视频，同时在 PR 分支和 main 上跑同一用例，把两段视频并排嵌到 PR 评论里；只支持一个技术栈（如 Playwright + GitHub Actions），录制触发通过 PR 上的 /record 命令或配置清单控制，先不做自动选测。

竞品：Meticulous.ai、Argos CI、Percy (BrowserStack)、CodeRabbit、Greptile、Graphite、Playwright Trace Viewer

摘要：作者因用 AI 代理写代码后 PR 泛滥、同事不愿评审，做了 Diff Stage：在 GitHub Actions 里录制浏览器测试，把分支与 main 的同一测试并排放在 PR 里供评审者对比行为差异。目前仅支持 Laravel + Pest，作者自曝为做这个工具烧掉的 token 远超省下的成本，说明需求真实但尚未有任何付费或收入验证；同类方向已有 Meticulous、Argos、CodeRabbit 等玩家。