NekoNeko

[NekoNeko @amashiro2233]

@Aph3li0sLM 是mimo嘛

[NekoNeko @amashiro2233]

@Da7_Tech @devindesktop I’d use Devin to push my work on autonomous operator synthesis further.

I’m building agents that write operators themselves, so I’d throw real problems at it: understand the codebase, implement the operator, run tests, debug failures, and iterate until it works.

[NekoNeko @amashiro2233]

@Da7_Tech Im ready

[NekoNeko @amashiro2233]

@Da7_Tech I'm willing to try Devin.

[NekoNeko @amashiro2233]

@loopPoly 这是啥🤔

[NekoNeko @amashiro2233]

@loopPoly 以前手动合并太痛苦了，现在ai全干 告诉他边界，完全胜任

[NekoNeko @amashiro2233]

@tianyi Github ID: Mashiro2000
https://t.co/LrmqSVKZ2s…
对Coding Agent很感兴趣，本项目是对Claude Code的高度复刻，UI和使用体验还原了90%多。现也在国内某公司维��垂类CodingAgent，希望参加DeepSeek Harness的内测工作

[NekoNeko @amashiro2233]

@tianyi Github ID: Mashiro2000
https://t.co/1lvb8b9nxp
对Coding Agent很感兴趣，本项目是对Claude Code的高度复刻，UI和使用体验还原了90%多。现也在国内某公司维护垂类CodingAgent，希望参加DeepSeek Harness的内测工作

[NekoNeko @amashiro2233]

@thsottiaux The Codex CLI does not support rolling back both code and conversation; it only supports conversation rollback. Even if Codex makes a mistake, I can’t effectively roll it back.

[NekoNeko @amashiro2233]

@PandaTalk8 ai provider sdk可以选择vercel的ai sdk，agent loop就自己搓吧

[NekoNeko @amashiro2233]

@wquguru 这是相关推文 https://t.co/pSwClBifhH

[Bearly AI @bearlyai]

Cursor internal analysis shows how hard Anthropic is subsidizing Claude Code.

Last year, a $200 monthly subscription could use $2,000 in compute. Now, the same $200 monthly plan can consume $5,000 in compute (2.5x increase).

[NekoNeko @amashiro2233]

@wquguru cursor之前不是发了一篇文章，claude 20 max的实际成本是5000刀🤔

[NekoNeko @amashiro2233]

@RookieRicardoR Claude支持在claude.md里面引用agents.md的内容呢。语法是`@AGENTS.md`

[NekoNeko @amashiro2233]

@runes_leo 可以启用tool search特性，按需加载工具

[NekoNeko @amashiro2233]

@elliotchen100 cc的memory看起来就只是为了mem而mem，生产力工具做mem，意义不如长期陪伴型agent

[NekoNeko @amashiro2233]

@reach_vb Cloned already! Backup!

[NekoNeko @amashiro2233]

@IceBearMiner 学习了很多
然后这边补充下opencode和cc的做法
opencode：
4个缓存点，两个打在system message，一个是静态system,另一个是动态syxtem,另一个则是非system的最后两条
cc：
用了3个点，2个在system，一个在非system的最末尾一条，没有在tool定义打

[NekoNeko @amashiro2233]

@blackanger bwt的对话内容主agent是不知道的，这个通过抓包很容易知道的

[NekoNeko @amashiro2233]

@gregpr07 API

[NekoNeko @amashiro2233]

it look very 🆒！

[Jayden0 @jiahao_jayden]

Hello everyone! I am one of the founders of We0 AI. We are from China and are excited to announce our product here. We believe We0 can help bring your ideas to life quickly. Enjoy!
we0: https://t.co/jfD8cbugZ1

https://t.co/vlNiwiMLpT