Most AI prompts that actually work live in your head or a random doc nobody opens.
UnDercontrol's skills system turns them into named, reusable templates — group-scoped, markdown-based, and piped straight to Claude Code or any agent via ud prompt
Your prompt library, self-hosted
UnDercontrol uses SSE to keep your tasks and finances in sync across every browser tab, desktop app, and device. Change a task status on your phone, see it update on your laptop in milliseconds. All self-hosted, all under your control.
If you live in the terminal, why leave it just to manage tasks?
UnDercontrol's ud CLI brings full task management into your shell — kubectl-style commands, SQL-like query syntax, notes, and a TUI kanban board.
You can use neovim update the content!
今天突然发瀑布和敏捷,其实是我用 Claude Code 遇到的一个思考的问题:
大家都说要开 Plan Mode,先把计划聊清楚再动手。我今天特地试了,跟 Claude 沟通了很多,把细节都说得很清楚,做出一份很详细的计划。
但问题来了——计划太多太细,跑起来上下文可能不够,还会失真,效果可能反而更差。
然后我换了个方式:计划太大就拆成几个阶段,每个阶段干什么分开写,搞一个总的待办事项,让 Claude 依次推进。最终效果嘛,也还行,但花的时间也不少。
后来想明白了,Plan Mode 最重要的是把架构打好,太细节的东西不应该在 Plan Mode 做。
架构打好之后,实际去跑,你会发现很多问题。没关系,把问题全登记成 issues。然后搞一波优化,跑第二轮,又一波问题,继续登记 issues。多轮迭代下去,就可以了。
其实这就是一套手工版、半自动的 Auto Research。但这种事本来就只能人工做,全自动的 Auto Research 现在也做不出来。只能说是这么个思想吧。
Track spending, log expenses, link them to budgets, see trend charts over 7/30/90 days, share budgets with your household or team. No third-party cloud. No compromises.
Most finance apps make you choose: privacy or features. UnDercontrol gives you both — a full budget tracker that runs on your own server. Your data never leaves your infrastructure.
Most finance apps make you choose: privacy or features. UnDercontrol gives you both — a full budget tracker that runs on your own server. Your data never leaves your infrastructure.
It seems like everyone in my feed is building either an agent terminal multiplexer, or an agent memory store. I'm definitely not working on a multiplexer.