AI UNDERDOGSDAILY PICK
AI UNDERDOGS
AI 写崩代码?还能撤回
AI broke it? Undo that
arrowassassin/kintsugi
用 AI 写代码最怕什么?
agent 手一滑
rm -rf 直接把项目干没了
Kintsugi 的思路不是拦着它
而是让破坏变得可逆
The scariest thing about AI coding agents?
One rogue rm -rf and your project vanishes
Kintsugi doesn't block the agent — it makes
destruction reversible
不封锁,自动备份
Auto-backup before destruction
它在 agent 跑破坏性命令之前
自动把相关文件备份一份
真搞砸了?
直接回滚
agent 继续干活,你也不用慌
Before any destructive command runs, Kintsugi snapshots the
affected files. Something breaks?
Roll back. The agent keeps working, you keep
your sanity
★ SIGNAL 1
每条命令都有记录
Full audit trail
每个 agent 跑的命令
改的文件,全部记下来
你要是同时让好几个 AI 写同一个项目
这个审计日志就是你的定心丸——出了问题能精准定位
Every command, every file change — all logged
When you're juggling multiple AI agents on the
same codebase, this audit trail tells you exactly
who did what
★ SIGNAL 2
纯本地,零云依赖
100% local, zero cloud
所有操作都在你本机完成
不调外部服务,数据不出仓库
听起来应该是标配对吧?
但市面上大部分 AI 安全工具居然都过不了这关
Everything runs locally — no external services, no
data leaving your repo. Sounds like a given?
Somehow most AI safety tools don't clear this bar
项目叫 Kintsugi——金缮
日本用金子修补破碎陶器的工艺
作者觉得代码被搞坏并不可怕
修好了反而更结实
这个命名品味我是服的
The name comes from Kintsugi — the Japanese
art of repairing broken pottery with gold. The
idea: breaking code isn't scary if you can
make it whole again. Chef's kiss on the
naming
AI UNDERDOGS
开源免费,给你的 AI agent 装个保险
Open source — insure your AI agent
arrowassassin/kintsugi
关注 · 每天发现更多 AI 神作
github.com/arrowassassin/kintsugi