AI UNDERDOGSDAILY PICK
AI UNDERDOGS
10 个 AI 同时写代码，谁来审？
10 agents code in parallel. Who reviews?
tmppr
你开了 10 个 AI agent 同时改代码
它们各写各的，最后谁来审、谁来合？
人根本跟不上这个速度
You spin up 10 coding agents working in
parallel. They each write code independently — but
who reviews it, who decides what merges?
You can't keep up
PR、CI、审查，全在本地跑
PRs, CI, reviews — all local
tmppr 把 GitHub 的协作流程搬到了本地——PR
CI、代码审查、issue、自动化
合并规则
这些开发者熟悉的协作原语全部在本地运行
专门给 agent 之间用来互相协调分工
tmppr brings GitHub workflows locally — PRs, CI
code review, issues, automations, merge rules. All the
collaboration primitives devs already know, running locally so
agents coordinate with each other. Not another CLI
wrapper
★ SIGNAL 1
不是包装，是重新定义分工
Not a wrapper — a redefinition
agent 提交 PR
另一个 agent 来审代码
审过了自动合并——这整套流程在本地闭环
作者不是简单套个 CLI
而是重新思考了 agent 之间怎么划分工作
怎么对彼此的代码负责
One agent submits a PR, another reviews the
code, and if it passes, it merges automatically
— the whole loop runs locally. The author
rethought how agents divide work and hold each
other accountable
★ SIGNAL 2
插件 SDK 是真的能用的那种
A plugin SDK that actually works
插件 SDK 不是搞个 YAML 配置文件就叫扩展——你可以加 agent provider
自动化触发器、合并规则
hook 执行器、事件订阅
作者认真想过扩展性
而不是把所有东西硬编码完了就交差
The plugin SDK lets you add agent providers
automation triggers, merge rules, hook executors, event subscribers
The author actually thought about extensibility instead of
hardcoding everything and calling it a day
多人同时改代码的冲突和协调
本来就是工程难题
tmppr 的思路是：既然 agent 越来越多
越来越能自己写代码
那就给它们一套专门的协作层
而不是让人类在中间手动当裁判
Coordinating multiple developers changing code simultaneously has always
been hard. tmppr's bet: as agents get more
capable, give them their own coordination layer instead
of making humans the bottleneck
AI UNDERDOGS
Dev tools · Multi-agent coordination
Dev tools · Multi-agent coordination
tmppr
关注 · 每天发现更多 AI 神作
tmppr.com