AI UNDERDOGSDAILY PICK
AI UNDERDOGS
一个终端跑 10 个 Claude
Run 10 Claude Codes, One Terminal
wastedcode/claudemux
Claude Code 一次只能跑一个任务
但这个工具让你在一台机器上同时开 10 个
输出互不干扰
Claude Code runs one session at a time
This tool lets you spin up 10 on
the same machine, output streams clean
Claude Code 的并行调度器
A Multiplexer for Claude Code
简单说就是个进程调度器
专门为 Claude Code CLI 设计的
上层是 CLI 命令行直接用
底下是 Node.js 库
你可以 require 进自己的脚本里做批量编排
Basically a process multiplexer built for Claude Code
CLI. There's a CLI on top, and underneath
it's a Node library you can import to
orchestrate batch tasks programmatically
★ SIGNAL 1
136 次提交，带 ADR
136 Commits With ADRs
翻一下 git log
136 次提交
每个架构决策都有 ADR 记录——CLI 的动词怎么封装的
输出边界怎么修复的
全都写得清清楚楚
一个人的项目做到这个文档程度
是真的想让别人接手维护
The git log shows 136 commits with Architecture
Decision Records — CLI verb design, output boundary
fixes, all documented. This isn't a weekend hack
this person actually wants others to maintain it
★ SIGNAL 2
输出串流才是最难的
Output Isolation Is the Hard Part
开多个进程不难
难的是让它们的输出互不污染
代码里有个 settle-on-exhaustion 的边界修复
说明作者自己跑的时候确实遇到了输出乱窜的问题
然后专门花功夫解决了
Launching processes is easy — keeping their outputs
from bleeding into each other is hard. There's
a settle-on-exhaustion boundary fix in the code, which
means the author hit this wall in practice
and spent real effort fixing it
没有作者访谈
但 136 次提交和完整的 ADR 文档已经说明一切
这不是那种做完 demo 就扔的项目
Git 里每一个决策记录都写清楚了为什么这么做
是有长期维护意识的
No author interview available, but 136 commits and
full ADR documentation speak for themselves. Every architectural
decision is recorded with rationale — this is
a project built to last, not abandoned after
the demo
AI UNDERDOGS
跑多个 Claude Code 的人，值得看看
If you run multiple Claude Codes, check it out
wastedcode/claudemux
关注 · 每天发现更多 AI 神作
github.com/wastedcode/claudemux