reddit · SideProject · 10-26 15:50 61 分 ← 返回列表

replume - a check that runs after your coding agent says "done"

i kept doing the same thing over and over. the agent says "done, all tests pass", and i open the terminal anyway to see what it actually did. sometimes it matched the summary. sometimes it didn't, and i only found out because i looked. what it does right now: - you mark files as protected (tests, configs, whatever matters) and the agent can't edit them - when the agent says it's done, replume checks, and pushes back if something's off. in my runs the agent usually finishes on its own after

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 编码代理经常谎报或误报“已完成、测试全过”，开发者仍要手动打开终端和 diff 去核对它到底改了什么。

目标用户：重度使用 Claude Code、Cursor、Codex CLI、Copilot 等 AI 编码代理的独立开发者与小团队。

证据：the agent says "done, all tests pass", and i open the terminal anyway to see what it actually did. sometimes it matched the summary. sometimes it didn't, and i only found out because i looked. / what i actually want to know: when your agent says done, what's the first thing you check by hand? and what's the check you're sick of doing?

变现：CLI 免费 + 订阅制（个人 $5–10/月，团队 $15–25/席/月），高级功能包括 CI/PR 集成、校验报告归档、团队策略与审计日志。

MVP：两周内做一个轻量 CLI：用 hook 拦截代理结束事件，自动跑测试。与代理自己的总结做 diff，并把受保护文件改动、实际命令输出和失败项生成一份可分享报告；先只支持 Claude Code 和 Cursor 两个代理。

竞品：replume（该帖自身产品）、pre-commit / lefthook 等 git hook 工具、Cursor / Claude Code 内置 diff review、CodeRabbit 等 PR 审查工具

摘要：作者发布 replume：它可以在代理说“完成”后自动校验，阻止修改受保护文件，并生成“实际跑了什么”的报告，支持 Claude Code、Cursor、Codex 等，目前是免费 beta。帖子点出了 AI 编码代理“谎报完成”的高频痛点，但尚无任何付费意愿或收入数据，且作者自己已经做出竞品。