pythonhii @pythonhii
Be the change you want to see in the world! GD,China Joined September 2009-
Tweets8K
-
Followers230
-
Following2K
-
Likes15
OpenAI just launched GPT-6.1 Sol, and for coding it can 𝗿𝗲𝗽𝗹𝗮𝗰𝗲 𝗔𝘀𝘁𝗿𝗮 𝗼𝘂𝘁𝗿𝗶𝗴𝗵𝘁 at 1/5 the price. OpenAI also announced today that 20x Pro is being cut to 10x. I honestly don't dare use Astra anymore. Switch to 6.1 Sol before Astra eats your whole week. 6.1 Sol is already in Codex. Set it up like the tree: 6.1 Sol runs the root orchestrator and all three subagents, and Astra only does an independent review before a big change ships. Reasoning effort comes from OpenAI's DeepSWE chart: 6.1 Sol scores highest on high, and 𝗴𝗼𝗶𝗻𝗴 𝘂𝗽 𝘁𝗼 𝘅𝗵𝗶𝗴𝗵 𝗼𝗿 𝗺𝗮𝘅 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘀𝗰𝗼𝗿𝗲𝘀 𝗹𝗼𝘄𝗲𝗿, so the orchestrator runs on high. Medium scores only about 2 points lower than high and costs over 30% less per task, so the subagents run on medium. Send the tree and this prompt to Codex 👇 "Set up my Codex to match this tree: → Main session: default to `gpt-6.1-sol` on high reasoning effort. It breaks down tasks, delegates, integrates, and verifies. Handle simple tasks directly; don't delegate them. → Three project-level custom agents, each with model `gpt-6.1-sol` and model_reasoning_effort `medium`: explorer reads code, worker edits code and runs tests, researcher looks up docs. Reuse fitting ones if they exist, and create only the missing ones. → reviewer: also a project-level custom agent, with model `gpt-6-astra` and model_reasoning_effort `xhigh`. Call it only before a big change ships. It reports problems and doesn't edit code. → Put the delegation rule in AGENTS.md: when spawning a subagent, set `fork_turns` to `"none"` or pass only the last few turns. If `fork_turns` is omitted or set to `"all"`, the subagent inherits the main session's model and effort and can't override them, and the tree stops working. Before changing anything, check: → Roles other than the reviewer that still use `gpt-6-astra` or an older Sol (like `gpt-6-sol` or `gpt-5.6-sol`). List them. → Roles moving to 6.1 Sol with effort set to `none` or `minimal`: change them to `low` (6.1 Sol supports neither), then compare the results. → Model assignments written only in AGENTS.md: check recent subagent session logs to confirm which model each role actually ran. If you can't select a model in the tree, stop and ask me. Show me the changes first, and wait for my OK before writing anything."
GPT-6.1 Sol is here. Upgraded with stronger agentic coding and computer use, near-Astra performance, and cached input at a 95% discount to standard input pricing. GPT-6.1 Sol is built for complex refactors, deep codebase investigations, and long-running agents across apps.
Andrew Ng just dropped the best 2-hour course on Graph Engineering: from single agent to full automation 9:14 - your first agent 33:11 - loop engineering 1:02:46 - graph engineering 1:30:15 - agents that rewrite themselves 1:49:05 - full graph system two hours, and it replaces every agent tutorial you bookmarked this year Prompts → Agents → Loops → Graphs most people will stop after the first agent and call it automation he saves the last forty minutes for the graph that runs it without him same model, same tokens, completely different week watch it today the step-by-step guide is below, save it while it is still early ↓
《什么是 Transformer》由 Claude Code + Opus 5.5 制作 --- 提示词 --- 帮我用js制作一个视频,主题是:什么是 Transformer 要深入浅出,让高中生也能看得懂,不仅high level说的清楚,也要有细节,包括注意力机制,甚至一些数学概念 你可以用任何工具或者安装工具,可以联网检索 请给我惊喜
@supezen 让 AI 写中文文档,总有一股 AI 腔:“值得注意的是”“不仅……更……”,满篇“强大”“无缝”。 我做了个 Skill 专门治这个。规则基于阮一峰的中文技术文档写作规范,再加 14 条 AI 腔检查。 npx skills add leter/zh-tech-writing -g github.com/leter/zh-tech-…
下面是原始正文,针对各自的项目,需要稍微修改使用
# CLAUDE.md
## 核心工程原则
1. **架构与领域优先**:计划阶段应以理想架构为目标,明确业务目标、领域边界、模块职责、依赖方向和数据流,形成符合领域规律、面向长期维护且可持续演进的设计后再进入编码;不得以短期实现便利牺牲整体设计。设计必须完整,实现应当克制:不做推测性抽象,抽象延迟到第二个真实用例出现时才引入,单一场景直接实现。
2. **追求优雅的代码模块**:模块应高内聚、低耦合,通过精简且稳定的接口封装内部复杂度,使职责、命名、依赖和扩展方式清晰自然;代码按单一职责拆分,单个文件不得超过 500 行,接近上限时应优先重构模块边界。
3. **保持边界与数据流清晰**:协议模型、领域模型、持久化模型和视图模型不得相互泄漏;数据必须在边界处完成校验和独立转换,避免跨层共享可变状态。
4. **安全与隔离默认开启**:所有功能均按多租户、多用户场景设计,明确认证、授权和数据隔离边界;遵循最小权限原则,任何外部输入均视为不可信,敏感信息不得进入代码、日志或响应。
5. **面向并发与故障设计**:后端应主动考虑幂等性、竞态、事务边界、超时、取消、重试、背压和资源释放;不得通过无边界重试、吞错或隐式共享状态掩盖问题。
6. **保障完整前端体验**:前端应控制渲染成本、异步状态和并发请求,保持清晰的 UI 结构;用户流程必须覆盖加载、空状态、错误、重试、反馈和可访问性。
7. **复用稳定的业务语义**:优先复用已有模块和能力,但不要仅因代码外形相似而过早抽象;确需重复时,必须注释说明其独立演进或暂不抽象的原因。新增依赖前先核查项目已有依赖(根 `package.json` 与 `packages/` workspace)能否满足需求,不得臆断已有库缺少功能——先查阅文档和类型定义;确需引入时优先成熟且维护良好的库,不重复实现通用功能。
8. **为未来维护者保留上下文**:代码、注释、测试和架构文档是跨越时间的协作媒介。非显然的设计决策、兼容约束、已知缺陷和临时方案,必须记录原因、影响范围、潜在风险及移除条件;技术债务应关联可追踪任务,关键架构决策应同步到 ADR,禁止留下缺少上下文的 `TODO`。
9. **确保变更可验证、可观测、可回滚**:每项改动都应行为可测试、运行状态可观测、故障可定位,并兼顾向后兼容和回滚路径;错误与日志必须保留诊断上下文,但不得泄露敏感信息。
10. **删除优于兼容**:内部路径重构时直接删除过时实现,禁止新增兼容层、deprecated shim 或双写逻辑;对外契约(`/api/*` 等稳定接口、数据库迁移)的兼容性按协议契约单独评估,属于合同义务而非迁就旧代码。
> **变更速查**:通常提交前运行 `bun run precheck`;修改前端后额外运行 `bun run build:web`;修改 schema 后运行 `bun run db:generate --name
感谢读者 LiNan1984 把《深入理解 AI Agent》蒸馏成了 22 个 skills,方便 agent 在开发 agent 的时候使用,也方便人类读者快速阅读。 github.com/bojieli/ai-age…
根据这张图我们优选模型挡位 Opus 5.5 medium GPT-6 Sol high GPT-6 Luna extra-high
GPT-6 Sol and Luna are now available in Devin. On FrontierCode 1.1, GPT-6 Sol matches GPT-5.6 Sol’s score at 61% lower cost per task. GPT-6 Luna scores above GPT-5.6 Luna at about a quarter of the cost. At under $0.10 per task, it is the cheapest model on the leaderboard.
At the top of my AGENTS.md: - NEVER write unit tests after you write code. - Highly prefer E2E tests as the sole testing mechanism. Use them to verify complex features work. At the end of E2E tests, produce a verifiable and repeatable artifact. - If you must test a system in isolation, FIRST write all the ways it could fail, THEN write the code.
leave it to your boy opus to add 10 unit tests to ensure a constant string contains various substrings
最近摸索了一种新的节约 Token 但是效果不错的模式(都是被逼的,Token 太贵): Fable 写设计方案 -> (Loop 开始) Opus 执行 -> Advisor(Fable) 验收 (Loop 结束) 步骤是这样的: 1. 先 Claude 里面设置一下 advisor 为 Fable > /advisor fable 这样在使用 Opus,Sonnet 模型的时候,遇事不决它会请教 Fable,可能会有无谓的 Fable 消耗,但还是可以接受也值得 就好比请了个高级工程师当顾问 2. 复杂一点的问题先让 Fable 出方案,最大化利用 Fable 的能力。 就好比你让一个高级工程师写一个技术方案,让普通工程师去执行 3. 方案写好了确认了,让 Opus 去执行,但是执行的时候,加一句提示词: > 完成后让你的 advisor 验收一下,如果验收有问题就按 advisor 的反馈修复,修复完成后再让 advisor 验收,直到通过 advisor 的验收为止 可以配合 /goal 使用效果更佳 这样 Opus 在实施完成后就会找 Advisor 验收,通过 Advisor 就会给出反馈,然后 Opus 继续修改。 这样的好处就是整个过程你不需要人工去用 Fable 确认—— 以前我就是这么干的,Opus 执行完,我得回到之前 Fable 的会话去让它验收,或者新开一个会话,把文档发给 Fable 验收。 现在 Agent 自己就能形成闭环,不用你自己参与。 就好比你请了个高级顾问,复杂问题找顾问出方案,执行让你的工程师去按照方案执行,每次工程师做完自己去找顾问验收,顾问不仅验收还会告诉工程师哪里不对,然后工程师自己改,不用说每次做完找你,你再找顾问。
现在我很多复杂一点的任务都是 Fable 5 写方案,Codex 去执行,Fable 5 验收,相对来说可以做出比较靠谱的方案,以及兼顾性价比。具体是这么做的: 1. Fable 5 的产出是技术方案文档(图1) 一方面文档方便批注修改,另一方面文档方便其他 Agent 快速上手 2. 文档确认后在当前先 /compact 一次
有点燃炸兄弟们,这才是顶级工程师驾驭 AI 最硬核的打法,建议调到两倍速把这场 38 分钟的技术内幕看完,去感受一下什么叫真正的生产力震撼。 前 Meta 和 Netflix 资深工程师、现 xAI 核心主力 Lauren Tan 刚把给 Cursor 官方大会准备的底牌直接公开了,她一个人带领智能体舰队单月合入 2500 个生产 PR,而且自己还能安稳睡大觉。 很多人用 AI 编程把自己变成了全天候无休的人肉测试员,而她做对的事情只有一件:把人类的检查点彻底往前移,让智能体不仅要写代码,还必须自己拉起真实环境把证据拿出来。 铁汁们你知道单人带队一个月往生产环境合入 2500 个 PR 是什么概念吗? 整整 2500 个。 按每月 22 个工作日换算,相当于每天合入 113 个,平均每 4 分钟就要上线一个功能。但全网都在跟风吹代码生成有多快的时候,她开场第一句话直接泼了一盆冷水:如果代码质量不过关,盲目堆吞吐量就是一场灾难,连一个智能体都信不过就千万别开一百个云端舰队,那纯粹是在花钱买代码垃圾。 最先击穿行业认知的,是她把做完这两个字的定义重新写了一遍。 以前大家让 AI 写代码,只要终端显示编译通过或者模型回复说搞定了,人就赶紧跑去肉眼看 diff,一行行核对直接把工程师累成整条链路里最慢的瓶颈。 在她的系统里,模型自己的口头报告连标点符号都不能信,做完必须等于铁打的运行时凭据。 智能体必须自己调用 Chrome DevTools Protocol 驱动真实界面,点击到具体按钮,甚至跑完 CPU 性能分析抓出 trace 日志,把全套绿灯的执行录像留在原地,才算完成交付。 但光会验证还远远不够,遇到喜欢抄近路的智能体,下周相同的错误还会原样长出来。 传统团队纠正 AI 习惯写几十页的长篇规范或者在 prompt 里苦口婆心叮嘱,结果模型转头就忘。 她在 Dune 架构里直接把代码库当成外部记忆体,按业务功能严格把代码锁在单层目录内,更绝的是把所有被团队反复骂过的坏习惯直接写成编译炸弹。 比如在项目里直接封死 useEffect,谁敢写一行 CI 当场引爆,甚至连代码注释都直接禁止,配专门的清理工具全面清场,因为注释里 99% 都是过期的废话和误导性教条。 把错误路径用规则物理切断,AI 抄近路走出来的就只能是唯一正确的路。 真正让我觉得有意思的是,她把这套顶级工程师的心法全部封装成了现成的开源插件 pstack。 不用再对着聊天框盲目敲长提示词,系统内嵌了 23 本应对不同场景的任务剧本和 23 条工程铁律。 从复现 bug 到分层合入,智能体被严格规定先定位根因再动手,默认先做删减再做新增。 现在这套自动化外环甚至接入了用户反馈通道,智能体接到报错自己拉起模拟器复现,跑完验证直接提 PR,产品经理和设计师连一行底层代码都没摸过,都能安全地往主分支合入补丁。 用量狂飙不可怕,没有验证的并发才可怕。 以前是人写代码教 AI 抄,现在是人当主厨把控出餐口,把工程品味铸进静态检查,让智能体在环内自己修到全绿。 一个只能丢 diff 等你看的智能体是在偷走你的时间,一个带着运行录像和证据链自己闭环的智能体才配做你的工程分身。 这套打法已经在 Cursor 插件市场完全公开,输入命令即可直接挂载。 大家在日常写代码时,最敢彻底放手交给智能体去跑的是哪一步?我先说,排查冷门报错和抓性能日志我现在眼睛都不眨全丢给它,但核心架构的边界定义我半步都不会让。 x.com/poteto/status/…
here's how i shipped 2,500 PRs last month to production this was originally supposed to be for Cursor Compile in London. i couldn't make it since i was livestreaming for Grok @bot Galaxy so i'm making it available for free here on X! watch it on 2x speed, i talk slowly
# Scope Guard Complete the task with the smallest sufficient change. Explicit user instructions for this task override the defaults below. ## Before editing - Read relevant code, call paths, and project conventions. Do not explore the entire repository before a small change. - Make clear, small fixes directly. Write a short plan when the approach is unclear or the impact is substantial. - Decide routine details yourself. Ask when different interpretations would lead to materially different work. - Load only relevant Skills, not an entire workflow for a matching keyword. ## While editing - Before adding code, check in this order: 1. Existing code and patterns in the project 2. Standard library and built-in platform capabilities 3. Dependencies already installed 4. Only then, new code required for the current task - Fix the root cause. Do not add abstractions, configuration, or compatibility layers for hypothetical future needs. - Justify new dependencies or frameworks: why are existing options not enough? - Fix nearby issues only if they block the task. Otherwise, flag them for follow-up. - Prefer precise edits for small changes. Avoid unrelated refactoring or whole-file rewrites. - Remove replaced implementations. Keep old paths only when compatibility is explicitly required. - Preserve necessary validation, error handling, security, and accessibility. ## When to ask - Keep implementing, verifying, and fixing within the authorized scope. Do not repeatedly ask whether to continue. - Confirm material scope expansion, unapproved costs or permissions, and unauthorized irreversible actions before proceeding. - If asked only to analyze or review, report findings. Do not edit code. ## Testing - Verify the change and its impact. Complete required project checks. - Reuse existing tests first. Add tests for real behavior and regression risks, not tests that mechanically repeat the implementation. - Temporary verification scripts need not become permanent test files. - Once checks pass, broaden or repeat them only for new changes, failures, or specific unresolved concerns. ## If the plan grows If you find future-only layers, unrelated refactoring, extra features, or unjustified repeat checks, return to the current requirement. Drop the excess work and finish within the existing authorization. ## Done means - The requested behavior works and necessary verification is complete. Do not stop at a plan or an initial implementation. - Every change serves the task. Remove obsolete code and temporary files introduced by this work. - Briefly report the result, verification evidence, and anything unresolved or unverified.
最近做 APP 一直在用一个 Skill 叫 Apple Design,将 Apple WWDC 设计分享中的界面与动效原则整理成指导,帮助改善界面层级、反馈和整体连贯性。实践效果非常好,简单克制、很有苹果原生 App 的味道。 作者Emil Kowalski 之前在 Vercel 现在是 Linear 的设计师,Apple Design 是他整理的一组 AI Agent Skills 其中的一个,目的是让 AI 做界面时更有“设计品味”,尤其是动画选择、交互细节和视觉打磨。除了 Apple Design 还有好几个都很实用,项目GitHub Star 数接近 4w:github.com/emilkowalski/s… 「emil-design-eng:设计工程总则」 关注 UI 的整体质感、组件设计、动画选择和交互细节。适合开发新界面时作为总体设计指导。 「animate:制作动画」 从零设计并实现动画。会先判断是否有必要加动画,再决定动画目的、实现方式、属性、曲线和时长。适合弹窗、抽屉、按钮反馈等动效。 「review-animations:审查动画」 检查已有动画是否合理,例如是否多余、速度是否拖沓、缓动是否合适,以及是否考虑减少动态效果等。适合动画完成后的专项评审。 「improve-animations:改进项目动画」 检查整个项目的动画,并整理出可执行的改进建议。适合接手已有项目、想系统提升动效质量时使用。 「find-animation-opportunities:寻找动画机会」 判断界面上哪些地方值得增加动画,哪些地方不应该动。重点不是“多加动画”,而是让动效真正帮助用户理解操作反馈或界面状态。 「apple-design:Apple 风格设计原则」 将 Apple WWDC 设计分享中的界面与动效原则整理成指导,帮助改善界面层级、反馈和整体连贯性。 「prototype:制作多个原型方案」 针对一个 UI 需求生成多个不同方案,便于比较布局和组件形态。适合设计方向还不确定时使用。 「mobile-native:优化移动端网页体验」 帮助网页在手机上更像原生应用,关注安全区、触控反馈、输入框缩放和移动端视口等细节。 「write-swift:编写现代 Swift」 覆盖 Swift 值类型、并发、泛型、性能和测试等工程实践。适合 Swift 代码编写与整理。 「pick-ui-library:选择 UI 组件库」 根据需求挑选合适的组件库,避免不必要地手写复杂组件,或引入维护状况不佳的依赖。
太牛了,那些我们说不清、道不明,称之为经验的审美 这个网站全部都书面化了。变成可以附佣、可以学习的内容。 github.mytemos.com/refactoring-ui…
Anthropic 工程师讲了一堂 FDE 入门课 youtube.com/watch?v=Kwhgfw… Kevin Bai 现在在 Anthropic 的 Applied AI 团队,之前是 Rippling FDE 团队的创始成员,再之前在 Palantir 干了好几年。最近他做了一个 FDE 101 的分享,把前线部署工程师这个角色讲得很清楚,值得总结一下。 先说一个数据:在上市 SaaS 公司里,按平均合同金额排,Palantir 是 400 万美元,ServiceNow 120 万,Workday 60 万,剩下的没有一家能超过 50 万。Palantir 靠几千人做到了别人几万人做不到的客单价。靠的就是 FDE 模式。 FDE 到底在解决什么问题? Palantir 的产品 Foundry 是一个应用构建平台,技术门槛很高,但买家是石油、消费品这些行业的非技术高管。你把一个复杂的技术平台丢给一个不会写代码的人,指望他自己搞明白怎么用,这不现实。 所以 Palantir 的做法是:客户买的既不是软件产品,也不是咨询服务,而是一个"结果"。你派工程师过去,深入理解客户的业务场景,在平台上给他们把东西建出来。客户关心的是货架上多了多少商品、产线效率提升了多少,他们不关心数据怎么组织的,也不该关心。 FDE 和外包开发有什么区别? Kevin 特别强调了一点:如果你的工程师每次都从零给客户写定制代码,那你做的不是 FDE,是外包开发。FDE 模式能成立,前提是你有一个可复用的平台。工程师是在平台已有的基础能力上组装和定制,不是每次重新造轮子。没有平台,维护成本会吞掉所有利润,工程师也会因为要维护几十个毫无关联的代码库而跑路。 要不要搞 FDE?两个问题就能判断。 第一,你是不是必须把一个技术复杂的东西卖给非技术买家?如果你的客户本身就是工程师,比如卖 GitHub 或者 Datadog 的,不需要 FDE。如果你的产品本身就是开箱即用的,比如 Slack 或者 Jira,也不需要。只有你的产品很复杂、客户又不懂技术的时候,FDE 才有必要。 第二,你有没有一个可复用的平台?或者你愿不愿意投入去建一个?没有共享的基础组件,FDE 就是不可持续的。 2026 年的新变化是什么? Kevin 的判断很有意思:软件行业做生意的方式本身变了,AI 让构建软件变得极其容易,几乎所有平台都在走向 Agent 化,这意味着几乎所有平台都变得高度可定制。后果就是:越来越多的客户搞不清楚你的产品到底能做什么。你把产品的成败交给客户自己去摸索,在 Agent 时代会越来越难走通。 这就把 FDE 从 Palantir 独有的小众玩法,变成了更多软件公司需要认真考虑的事情。 最后一个问题:什么样的人适合做 FDE? Kevin 的回答很简洁:FDE 就是一个你信任到可以让他直接面对客户的软件工程师。技术能力是基本盘,但你还得放心让他代表公司去跟客户打交道。
DeepSeek just launched V4.1 Flash. 98% of GPT-6 Astra’s score at 1.4% of the cost. With GPT-6 Astra + DeepSeek V4.1 Flash, you can build an AI workforce that works 24/7 for around $50/month. Full setup: x.com/i/article/2098…
官方RSI第一次开放给用户。Anthropic 刚刚做了一件很重要的事:让 Claude Code 自己优化 Claude 的成本。 所有用Claudecode的人都应该试一试: 1. /claude-api prompt-audit 让 Claude Code 直接扫描你的Prompt\CLAUDE.md\Skills\Tool descriptions\调 Claude API 的应用代码,然后找出那些老模型时代遗留下来的 Prompt 反模式 Anthropic 特别点名这些写法: double-check your work verify twice be maximally thorough CRITICAL: YOU MUST ALWAYS... 强制 step-by-step 强制 scratchpad 固定六步流程 为旧模型写的大量 few-shot 相互矛盾的 instruction 原因很有意思: 以前这些 Prompt 是为了弥补模型能力不足。现在模型变强以后,它们反而开始拖累模型。 例如新模型真的会把: verify twice 理解成真的执行两遍查询。 be maximally thorough 则可能导致 Claude 做几十次没有必要的知识库搜索。 Anthropic 做了一个 Opus 4.8 → Opus 5 的迁移实验。 只跑一次 prompt-audit: 成本下降 14.6% 同时 准确率反而提升 5.3%。 这个结论其实挺值得注意: Prompt Engineering 也会形成技术债。 模型升级以后,旧 Prompt 不一定继续帮你,甚至可能变成负优化。 2. /claude-api cost-optimize 这个更像是一个自动 FinOps / AI Cost Engineer。 你让 Claude Code 对整个 Claude API 项目做一次成本审计,它会先分析钱花在哪里,然后尝试: Prompt caching Prompt audit 减少每次请求携带的内容 限制 output Batch API 调整 effort 换模型 如果你提供 eval,它甚至会同时测: Cost × Performance 然后找一个更合适的配置。 3. /claude-api hillclimb 这个是我觉得最有意思的。 你给 Claude 一个 eval。 Claude Code 会: 拆 train / test → 改配置 → 跑 eval → 看失败案例 → 修改 Prompt / 模型 / effort → 再跑 → 最后在 held-out test 上验证。 Anthropic 给的客服 benchmark 案例: 原始 Opus 4.8: 78.6% 自动优化后的配置: 90.5% 同时: 成本大约只有原来的 1/5。 注意它最后甚至不是简单“换更强模型”。 它试了 Sonnet 5 low effort,准确率先掉到 88.9%,然后 Claude 自己读取失败样本,给 Prompt 加 routing rules 和 refund-cap cross-reference,最终把训练集重新拉到 98.9%
One more thing: Astra’s been burning through my quota too. My quota reset today, so I went all out. It felt like Astra was burning through my quota even faster than Sol did. Then I tried this, and it worked like a charm 👇 Check AGENTS.md, relevant Skills, and memory for instructions that tell it to spawn subagents by default. Remove those default requirements, then add this to AGENTS.md: “Handle the task and verification yourself by default. Do not spawn subagents except for necessary independent reviews or when I explicitly ask you to.” Let Astra finish the work itself. The main agent also uses quota to assign tasks, pass along context, and repeatedly check progress. Pick a task you know well, let it run solo, and compare the results and usage. You might see this model very differently. When you need parallel work, explicitly ask for subagents. Not every task needs an army of agents.
If Astra feels almost unusable on ChatGPT Plus because it burns through your quota too quickly, try this: let Sol do the work and bring Astra in as an advisor for the hard parts. Let Sol handle everyday coding and testing. When it faces an architectural trade-off, gets stuck, or
If Astra feels almost unusable on ChatGPT Plus because it burns through your quota too quickly, try this: let Sol do the work and bring Astra in as an advisor for the hard parts. Let Sol handle everyday coding and testing. When it faces an architectural trade-off, gets stuck, or needs an independent review, have it ask Astra for advice. Sol then keeps working. You don’t have to relay messages between the two models. In Codex, select Astra and have it turn this setup into a Skill: “Create a Skill for Sol: Sol handles task progress, implementation, and verification. Only call Astra at high effort for difficult decisions, architectural trade-offs, or an independent review. Astra provides advice, reasoning, and risks without editing files. Sol continues the work. Explicitly specify the model when calling the advisor. Use a fresh context (set fork_turns to "none") and include the question, necessary materials, and constraints it must follow. The advisor must not spawn other agents. Don’t delegate just for the sake of it. Keep existing approval requirements unchanged. If the current tools don’t support specifying a model, tell me first.” Once the Skill is ready, switch back to Sol and use it for your tasks.
Four things to settle before giving GPT-6 Astra a task: 1/4 GPT-6 Astra can overdo the checks on small changes. For coding tasks, it tends to test thoroughly before calling the work done. That’s useful when you want a finished result. But if you only want a first version to
cern @CerrnBasher_
17 Followers 927 Following Co-Founder & Chief Investment Officer at Brilliant Advice My posts here are not financial advice. Tesla referral link: https://t.co/xFC2NacYD8
BackupAcc 👑🌞�... @AshleyMasc65631
1 Followers 1K Following If you want to quickly turn $10,000 into $1,000,000, follow the link below to join us and learn more details.👇 👇🏼🧳🔥
杭州依诺 @SOlar3_96s
0 Followers 35 Following 💕万名实时极品美女为你服务💕 💕私密💕专业💕贴心💕靠谱💕 💕 ✨TG频道关注:https://t.co/Dwz4yrQQYm ✨TG客服安排:https://t.co/NENSgdkIhf
. @scaling011
3 Followers 332 Following lead them to paradise LisanBench: https://t.co/Hg7fAOQXmd Impressum & Datenschutz: https://t.co/oIrAKzAWZ4
Luna Bby @lunablissstp6v
31 Followers 1K Following Crafted from stardust, powered by good intentions ✨
😏 Amy Bliss XO @amybliss3143nd
21 Followers 866 Following this is your grind 🥵 watch my main: @amyblissxo
win98 @keva7in
1K Followers 995 Following 10年+码农:游戏→跨境支付 FX 用 AI 做工程化,做一些奇怪的小产品 Arch Linux | Neovim | INFJ 一个活人 | 健康永远第一位
4K New Wallpaper @Fresh4KWalls
2K Followers 5K Following 📱 Premium 4K Wallpapers 🎨 Android • iPhone • PC ✨ New wallpapers every day ❤️ Follow for daily updates
mayangsheng1203 @mayangsheng1203
1 Followers 61 Following
The Meow Cats @wildnaturhub
5K Followers 7K Following
harrieschen @HarriesChen
600 Followers 1K Following 程序猿一枚, 每天的工作就是指挥 AI 写代码 信用卡爱好者 Indie Hacker Founder of https://t.co/oBDQGGA0nX
Grok EM 𝕏 @xEmGrokx
108 Followers 6K Following
lucas @candy9943296643
1 Followers 22 Following
neeraj bajaj @swayambajaj1986
21 Followers 28 Following
Stoutoret @StoutoretsWD
65 Followers 2K Following
feksbdhdyen @TensorWill
30 Followers 683 Following
Tom Huang @tuturetom
40K Followers 7K Following Co-founder & CEO @OpenDesignHQ — building Open Design, the open-source Claude Design alternative. → https://t.co/J2erf2l0vF
Smenat @Smenat127998
0 Followers 16 Following
YC (Yucheng) @yucheng
20K Followers 2K Following ⛴️ @chuhaiqu AI Solopreneurs Club 🎯 @TuttiHQ Monetize Your Influence Globally 🦾 @LuciusHQ Context Layer for Organization ♊️ Curiosity-driven 📍 Boston
Chenglong Geng @gengchl
2 Followers 101 Following
数据幻影 @DataFavorites
693 Followers 469 Following
nasser yasin @nasseryasin1
4 Followers 246 Following
恋人果 @zK8PRPJOCw8BD5d
8 Followers 216 Following
yaoyue @yaoyuezhujiu
10 Followers 283 Following
Rocky @hirockyrubio
69 Followers 1K Following
Dai ᴰᵃⁱ @0575Dai
3 Followers 8 Following
Shen @ST67387360
2 Followers 145 Following
shell @yQPx2fw01B0YRhL
1 Followers 1 Following
方高洁 @ANPJCuYaFg1PVw1
74 Followers 2K Following 解放你的寂寞 华人交友 全球最大的交友约会平台 交友约P 走心走肾 APP下载地址: https://t.co/rlrxfj8Ybx
Louis-François Bouch... @Whats_AI
12K Followers 6K Following Training AI Engineers on YouTube, Substack, Towards AI Academy and O’Reilly. Co-founder & CTO @towards_ai. Ex-Ph.D. student @Mila_Quebec.
7w 🕊️ @Btc8888XF
27K Followers 7K Following 段子手 | 价值投资 | all in 推特 脱下孔乙己的长衫,赚钱即正义! web2是生活,web3是梦想 JUST DO IT ! HTXDAO志愿者 | #蓝鸟会
大伟|AI × Web3 @david1989_zhu
4K Followers 4K Following AI Agent × Web3 实战派 前 CEX 技术总监|10 年 Java 架构师|构建过亿级交易系统 用技术视角拆解商业模式,只讲能落地的 Agent 自动化 | Web3 产品拆解 | 技术商业思维 | 架构师经验 👇 每周深度线程 + 实战框架
Stephan Livera @stephanlivera
195K Followers 3K Following Bitcoin podcaster. Investor & Advisor @ORANJEBTC, @bringinxyz UAE Media Licence Permit: 8630300
huangserva @servasyy_ai
40K Followers 795 Following 古早程序员 | AI出海 | 自由职业 机车游侠&机速购&骑享租创始人 15年前 freelance 起步 → 连续创业者 → 亏过1个亿,逆风翻盘中 分享创业,AI,读书,生活,健身 Official X channel of SERVASYY LLC
Cell 细胞 @cellinlab
34K Followers 7K Following 🎯 AI & Crypto & DIY · 探索个体与社区的更多可能 | 🌱 build 造物矩阵 @ZaoWuJuZhen & 小火炉播客 @XiaohuoluFM | 🌐 @CellNomad |💬微信 vcellinlab|📮[email protected]
刘小排 @bourneliu66
32K Followers 102 Following AI Product Founder · Building in Public Raphael https://t.co/tJOiC33sAI AnyVoice https://t.co/MQkmOW2lbu Fast3D https://t.co/6m3JO8jJvc
Flynn @fly3nn
4K Followers 2K Following 程序员|重度 AI 玩家 | 坚持原创 分享工具玩法、实战方法,探索如何变现 写做过的事,想通的问题,可被复用的经验 🌟坚持简单主义,坚持长期主义🌟
KonghaYao - 江夏尧 @konghayao
262 Followers 107 Following Designer of Fenix OS | Agent Builder from Peri Code! https://t.co/Q4J7srhcqF
Ray Fernando @RayFernando1337
39K Followers 6K Following 12 years at Apple. Now building apps for a million people. Getting to 10% body fat. New on the App Store: SayRay 👉 https://t.co/YkfXIH9sJC
平凡OnlOOker @onlookersh
6K Followers 7K Following 🐧 前腾讯产品经理 💼 现某500强 AI战略总监 🏥 专注医疗、工具及流程管理领域 💻 Arch Linux / Vim 爱好者 ❤️ 西装/雪茄/威士忌/健身/宠物/自驾游/后摇/Tango 🔞 德智体美黄全面发展 @021onlooker
Roland.W @rwayne
62K Followers 1K Following 🇦🇺PhD ing @UQ_News | 医学与经济学 一个专注于解决问题并以此赚钱的男人 @nvidia DGX Spark 玩家|非主流AI自媒体人
龙海 @longhaiqwe123
3K Followers 1K Following 👨🎓 AI builder | 立志成为 X 上最会找痛点的男人 🤖 拾音 FM | 中国故事 | OhMyTab ☎️ 微信:longhaiqwe
Paweł Huryn @PawelHuryn
134K Followers 437 Following AI PM | Deep research. I build, test, then teach. 130K+ subscribe → https://t.co/Tpe2QTFxY2
Peter Yang @petergyang
287K Followers 2K Following Practical AI tutorials and interviews for busy people | Get my best AI skills and guides at https://t.co/6VAA6p81x6
Wes Winder @weswinder
14K Followers 958 Following Building https://t.co/orUEVDm2OZ, https://t.co/eKQTlsdfno, and https://t.co/7XVkYUkcn9. 12+ years of software dev experience. Founder of the #1 vibe coding community on X. Toronto 🇨🇦
Jesse Lau 遁一子 @jesselaunz
30K Followers 2K Following 遁一子V-Blog https://t.co/iPe4rJBwPm 交易idea到EA @tradingwithea 外汇战场地图 反转雷达 https://t.co/dCpSc4IsSm 零基础学日语进度打卡:https://t.co/lGe7SUeyop
LilTea @LilTea_eth
10K Followers 2K Following 上个浪潮all in crypto,下个浪潮 all in AI,用你能用的所有方式 我用了6年的稳定高速VPN机场✈️:https://t.co/SJB2JWJfTH
Tiezhen WANG @Xianbao_QIAN
13K Followers 3K Following Helping ecosystem to grow on vLLM, ex-Head of APAC ecosystem @huggingface. Ex-Googler on TFLite/micro. Ideas on my own. Interested in future tech. DM open
Jack @JacksAISites
11K Followers 10K Following Indie dev · Nordics | Solo AI products | Real build-in-public for hobby project 独立开发者(业余时间) · 北欧 | AI
RicoUI @ricouii
2K Followers 297 Following 👏网页设计师,探索 AI 和 Vibe Coding,分享一些很不错的产品以及写写文章,探讨关于设计和开发的一切体验。目前在研究把能力变现... 🌱 Blog: https://t.co/9K32QHfTuV 🎨 渐变资源: https://t.co/e5xIIx0sao 🛠 工具箱: https://t.co/W4ToeSO8Pv
Simon Høiberg @SimonHoiberg
164K Followers 489 Following Building a portfolio of bootstrapped SaaS products. - https://t.co/EFHbYmRx37 - https://t.co/lbG5NXHFu0 - https://t.co/vsIMIcYvis - https://t.co/obC06ifrcL
Jeffrey Emanuel @doodlestein
50K Followers 14K Following Former Quant Investor | My Open Source Projects: https://t.co/9qbOCDlaqM | Try https://t.co/oCtjI2mBIl , my collection of agent coding tooling.
sleepy.md @sleepy0x13
38K Followers 2K Following Cooking @BeatingOfficial AI|科技|文化 做有品味的内容品牌 所谓进步,就是实现一些乌托邦。
Mahesh Sathiamoorthy ... @madiator
17K Followers 2K Following CEO @bespokelabsai. RL/Envs/Posttraining/Agents. Created Generative Retrieval at @GoogleDeepMind.
SuSu_酥酥👅 @NFT_Chen
19K Followers 2K Following #BiBiCrew Co-founder|@Sydney_Uni Master|Stock & Crypto Trader|AI Builder|Not Financial Advice, DYOR
Chris W @Chris_Wozniczek
1K Followers 910 Following 🫡 @cognition Ambassador Ex IBM, Atos, Diebold Nixdorf 💻 Building apps with Devin, 💾 Local LLM supporter, M2 Max Owner, MLX user IT Services Consulting
FeiZ @Fei2411
9K Followers 3K Following Personal Agent Dev · Economics·AI Engineer @cognition Ambassador Specializing in how AI can be applied to life 📮[email protected]
Bin CHEN @SeTriones
308 Followers 266 Following try@9#, Head of Infra at @Zai_org; previously Head of Infra at @Kling_ai.
TechGeek Tesla 🔋�... @JonBbC_TechGeek
14K Followers 2K Following Tesla Tech Geek and Early Access FSD tester, Covering Full Self-Driving in Chicago, Industrial Automation Manufacturing Engineer, AI Enthusiast
Teslascope @teslascope
77K Followers 147 Following A complete view of everything about your Tesla vehicle. We are the worldwide drivers' platform and highly rated vehicle companion. 🚘 Not affiliated with Tesla.
faultbugs @faultbugs
3K Followers 2K Following Tesla Model 3 owner(2020 model 3 SR+ & 2024 model 3 LR). Rumor Buster. If you wanna know the true facts of Tesla China. Just follow me.
hiraoku @2020_hira
2K Followers 950 Following フロントエンドエンジニア | React / Next.js Claude Code / Codex / Cursor / Grok Bot を比較しながら使っています。 「どれが今の開発に合うか」を、触った結果だけ短くまとめて投稿します。 大阪、たまに東京でテックイベントのお手伝い。 コーディング、文章書くの好き。
Shann³ @shannholmberg
37K Followers 13K Following I cover AI marketing & growth. Sharing every framework as I build it. Growth @lunarstrategy @turtledotxyz
yibie @yibie
6K Followers 2K Following
耳朵 @RookieRicardoR
11K Followers 461 Following 🤖 AI LLM / Agent / Skill / MCP 观察与深度应用 👨💻 高级软件工程师 / 签约作者 / 软件工程 AI 实践 📈 标普/ 纳斯达克交易员 💪 AI 领域付费咨询请私信
小安 Anleo @hx0000001
2K Followers 3K Following 构建 AI 原生系统。 智能体架构、上下文工程、评估与自动化。 AI 原生操作者 研究如何把人的判断力编码为系统。 关注智能体、上下文 记录从模型能力到真实生产的距离。
camsoft2000 @camsoft2000
4K Followers 690 Following Creator of XcodeBuildMCP and AXe, OSS maintainer and AI optimist with occasional existential doubts.
Garry Tan @garrytan
1.1M Followers 6K Following President & CEO @ycombinator —Founder @garryslist—Creator of GStack & GBrain—designer/engineer who helps founders—SF Dem accelerating the boom loop
Chris @ChrisGPT
70K Followers 2K Following Agi 2029 - AI Insider / Reporter as featured in Axios • The Information • NYT • Techcrunch











































