📡 Idea Radar 每天 8:00 / 18:00 推送 · 18:00 生成精选
reddit · SideProject · 57 分 ← 返回列表

I got tired of broken merges when running Claude Code and Codex in parallel, so I built an open-source tool

I've been using multiple AI coding agents in parallel, usually in separate Git worktrees. The problem wasn't getting agents to work simultaneously. It was merging everything afterward. Each agent would finish its task, but once I combined their changes, something would break. Then I'd have to figure out which change caused it, send the errors back to an agent, wait for another fix, and run the tests again. It was wasting time and tokens, especially when several sessions were working on

五维评分

付费意愿
35
需求真实度
55
个人可落地
80
竞争空间
65
新鲜度
85

AI 分析

痛点:并行跑多个 AI 编码 agent(Claude Code / Codex)后,合并各自改动时冲突频发、难以定位责任会话,反复把错误回传再跑测试,浪费时间和 token

目标用户:同时使用多个 AI 编码 agent、在多个 Git worktree 上并行开发的独立开发者与小团队

证据:Each agent would finish its task, but once I combined their changes, something would break. Then I'd have to figure out which change caused it, send the errors back to an agent, wait for another fix, and run the tests again. It was wasting time and tokens, especially when several sessions were working on the same codebase.

变现:开源核心免费引流,团队版/托管式编排服务订阅(按席位或按仓库收费);或提供 CI 集成的合并前冲突与集成测试云服务,按用量计费

MVP:两周内做出 CLI:监听多个 Git worktree 的提交,检测文件级重叠与语义重复任务,合并前自动跑测试,并把失败日志精准回传给对应 agent 会话;先支持 Claude Code + Codex 两个后端

竞品:git worktree / git rerere 等原生 Git 工具、Cursor、Devin 等自带多 agent 编排的商用产品、Fanout(作者本人的开源项目,MIT 免费)

摘要:作者并行运行 Claude Code 和 Codex 多个会话,合并改动时频繁出错,定位问题要反复把错误回传 agent 再跑测试,浪费大量时间和 token。他因此自建 MIT 开源工具 Fanout,可检测重复任务、文件重叠,并在合并前做集成测试。项目目前免费且处于早期,仍在寻找独立用户反馈,尚无任何付费验证。