@timerringX

Citius, Altius, Fortius @GeorgiaTech Once GitHub Trending #1 https://nitter.cf/t.co/ONsBqSfQLD

Hong Kong
Joined December 2010
Whenever feel impetuous, ask yourself, “What are you really doing?”
2
1
4
2,319
I’ve found a coding tool that may be the best alternative to Codex for my workflow so far. Codex’s device-based account restrictions are quite strict. If you have a second ChatGPT Pro account, switching between Codex Desktop instances can be difficult, although the CLI seems less restrictive. In my case, I was logged out after sending my first prompt. I tried resetting device identifiers, reinstalling the app, and removing related local files, but none of that made much difference. Due to Tibo @thsottiaux 's explicit ban on using tools like sub2api to route Codex subscriptions, I began looking for coding tools that support ChatGPT subscriptions natively. I tried a large number of coding agents and model-backed tools, including many built around OpenRouter, but nearly all of them had their own limitations. This also exposed one of my biggest frustrations with Codex: it shows model changes mainly as diffs, but it does not give me a strong enough view of the project as a whole. I often cannot easily see how a batch of changes relates to the wider repository structure or to functionality already implemented elsewhere. As a result, the model may generate code for something that already exists in another module. Sometimes that is because I forgot that part of the project existed and did not explicitly tell the model to reuse or integrate it. But the agent may not discover it on its own either. Over time, that can lead to duplicated functionality, bloated code, and architectural decisions that drift away from my original design. Another frustration is that I cannot conveniently edit the model’s proposed changes in place. I often have to switch back to VS Code or another editor to make manual adjustments, which interrupts the workflow. So I started looking for a tool that could offer both a capable editor for hands-on work and a dedicated agent interface, while also supporting ChatGPT subscriptions directly. At first, I considered using VS Code with Kilo Code. A traditional editor solves many of these problems, but I gradually realized that an editor-first workflow becomes cumbersome when I am working across multiple projects. Each project needs its own window, and I end up constantly switching between editors. This reminded me of Cursor's AB test last year, where users opened Cursor and the default homepage was the Agent panel, not the editor interface. This approach seems very forward-thinking. So, is there a coding tool that meets my needs for certain editor scenarios, includes an associated Agent interface, and can directly use my ChatGPT subscriptions? After trying various tools, I finally found my answer: @zeddotdev , a high-performance, multiplayer code editor. Your last next editor. If one ChatGPT Pro account is not enough for your development, and you use multiple accounts with similar needs, Zed may be worth trying.
1
127
What year is this, anyway?
We're adding support for AGENTS.md to Claude Code. Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md. You can toggle this behavior in /config.
28
People who know me well know that I've been a huge fan of Perplexity from 2023. Recently, I didn't realize that Comet had somehow ended up on my phone. Even though I'd been using Perplexity on mobile all along, having Comet show up outta nowhere was really a pleasant surprise, and it actually works really well. When everyone's talking down on AI browsers, saying they're relics from the last era and that people have been using outdated products for the past couple of years, Perplexity always manages to surprise you. It's probably the only browser that's still making waves these days.
21
Absolutely amazing work
rebuilt the X algorithm with Jev - uses real weights - simulates virality of your post - has a global feed (you see everyone) it's insanely accurate
22
来自前前司的审美和设计真的是独一份的,共绩真是一个很好的团队,一群非常优秀的伙伴 常去 suanli.cn 看看 : )
1
1
2
193
Good skills often do not require a long winded explanation to fill in the details. No need for too many conflicting or even contradictory rules. The best skills are those that are simple but works. github.com/hardikpandya/stop…
1
79
发现一个适用于中文的 Skills 版本: 用来减少 AI 的那些“不是 而是、钉死、撞上、落到、你定、补一个实证、而不是猜、先把规则看清楚、方向反了、这三件事先做以及强行排比” 😬 github.com/KKKKhazix/human-w…
58
From my own experiences and after seeing some things others have done, I have further corrected my viewpoint: saying things directly is always the best way. Never deliberately beat around the bush for the sake of so-called emotional intelligence. In the end, it will only lead to arrogance or deceit. Sometimes people don't realize it themselves, but only when they see others doing it do they suddenly realize: Did I do that too in the past? The you in your eyes is not you, and the you in others' eyes is not you either; the true you is the one in your eyes of others. I apologize for the consequences of all my self-righteous emotional intelligence remarks in the past. 🙏
48
The foldable screen makes me more inclined to browse text and check X on my phone. The experience is just amazing, once you try it you won't go back.
45
Since I got a foldable screen, I feel like my whole way of using my phone has changed. Plus with the highly customizable Android ecosystem, I'm seriously considering switching my main phone from an iPhone to an Android 🤔.
1
67
Apart from some passkeys, there doesn't seem to be many lags when switching between them.
35
有些时候脚本 p 太快也不是好事 😅
This Post is from a suspended account. Learn more
37
12 hours, 20x pro 42%, GPT-5.6 Sol Light, 77 PRs, A memorable work except for ... ?
47
Definitely : ) 全球实际上最终会归拢到五座城市/地区 NYC, London, Tokyo, 湾区, 北京。如果限定 AI,可能只有最后两者。
这次在湾区逐渐领悟了 SF Downtown = 国贸三里屯 Palo Alto = 五道口 Redwood = 知春路 Sunnyvale = 回龙观 Santa Clara = 上地产业园 Mountain View = 西二旗 San Jose = 昌平丰台
93
Interesting Github Copilot product experience, though it would be even better if it supported switching model capabilities or BYOK.
41