Hallmark - 用于代码编辑器的反AI生成设计技能

• 功能介绍
  Hallmark 是为 AI 编程助手（Claude Code、Cursor、Codex）设计的技能，用于生成具有人类风格、独特的用户界面。它通过57项测试、自定义宏观结构和20个主题之一，主动避免常见的“AI味”模式，为每个设计需求生成完全不同的、非模板化的布局。
• 主要特点
  • 四种核心操作：构建（默认）、审计（评估现有代码）、重设计（用新指纹重建）和学习（从截图/URL提取设计DNA）。
  • 反AI套路引擎：运行测试和预检自我批评，拒绝生成千篇一律的默认设计。
  • 20个主题 + 自定义模式：提供主题库，并能为独特需求完全从零设计。
  • 自包含输出：生成独立的 HTML + CSS 文件，每个文件都标注了其结构框架。
• 为何值得关注
  它直接应对 AI 辅助开发中日益突出的问题：千篇一律的“AI生成感”。通过提供一个强制设计多样性和质量控制的结构化技能，它帮助开发者和设计师创建更真实、更具品牌感的界面。其快速增长的关注度（今日794星）反映了市场对于改善AI生成创意输出工具的强烈需求。

View Repository / 查看仓库

3x-ui-Upgrade (Railway Optimized) - A 3x-ui Fork for Single-Port Deployment on Railway

• What it does
  • This repository provides an optimized, forked version of the 3x-ui panel (specifically the Heimdall edition) designed to run seamlessly on the Railway platform.
  • It bundles an Nginx reverse proxy to expose the panel, user subscriptions, and VLESS/WebSocket inbounds through a single port (the one assigned by Railway), simplifying deployment in constrained environments.
• Key features
  • Single-Port Architecture: All services (panel management, client subscriptions, and traffic) are accessible via the single port provided by the hosting platform (e.g., Railway).
  • Simplified Deployment: Pre-configured with Dockerfile and scripts for one-click deployment to Railway.
  • SQLite by Default: Uses a lightweight SQLite database, requiring no external database setup for basic use.
• Why it’s notable
  • It elegantly solves a common challenge on PaaS platforms like Railway by consolidating multiple network functions into one port, making advanced network configuration tools more accessible.
  • The project is well-documented with clear step-by-step instructions, lowering the barrier for users to deploy their own panel.

3x-ui-Upgrade（Railway 优化版） - 针对 Railway 单端口部署优化的 3x-ui 分支

• 功能介绍
  • 本仓库提供了一个经过优化的 3x-ui 面板（Heimdall 版）分支，专为在 Railway 平台上顺畅运行而设计。
  • 它集成了 Nginx 反向代理，使管理面板、用户订阅和 VLESS/WebSocket 入站连接均通过 一个端口 对外提供服务，简化了在受限环境下的部署。
• 主要特点
  • 单端口架构： 所有服务（面板管理、客户端订阅、流量转发）均通过单一端口（如 Railway 分配的端口）访问。
  • 一键部署： 提供了预配置的 Dockerfile 和脚本，支持直接部署到 Railway。
  • 默认使用 SQLite： 默认采用轻量级 SQLite 数据库，基础使用无需配置外部数据库。
• 为何值得关注
  • 它巧妙地解决了在 Railway 等 PaaS 平台上的常见部署难题，将多个网络功能整合到一个端口中，使得高级网络配置工具更易获取。
  • 项目文档详尽，提供了清晰的分步指南，降低了用户部署自有面板的门槛。

View Repository / 查看仓库

Three.js Object Sculptor - A Codex Plugin for Image-to-Code 3D Models

• What it does: This is a Codex plugin that takes an attached image of an object and guides the AI to recreate it as a code-only, procedural Three.js model in the browser. It follows a strict, staged workflow—from image analysis and spec creation to multi-pass code generation and visual validation—to produce animation-ready 3D assets.
• Key features: Enforces a structured sculpting pipeline (blockout, structural, material, etc.); generates detailed ObjectSculptSpec with hierarchy for animation; includes Python scripts for probing images, validation, and generating comparison sheets; extracts procedural PBR material evidence from reference images; focuses on quality and feature fidelity through checkpoint reviews.
• Why it’s notable: It addresses a core weakness in procedural 3D generation by preventing “one-shot” failures. Instead, it positions AI (Codex) as a methodical sculptor with validation steps, ensuring the final model retains recognizable detail and is built with a hierarchy suitable for animation, physics, and interactivity. The 862 stars indicate strong developer interest in this structured, code-first approach to asset creation.

Three.js Object Sculptor - 一个将图像转换为代码驱动3D模型的Codex插件

• 功能介绍：这是一个Codex插件，能将附带的物体图像作为输入，引导AI将其重建为纯代码的、基于过程的Three.js浏览器模型。它通过一个严格的、分阶段的工作流（从图像分析、规格创建到多轮代码生成和视觉验证）来创建可动画的3D资产。
• 主要特点：强制执行结构化的建模流程（基础体块、结构、材质等）；生成包含动画层级结构的详细ObjectSculptSpec；提供用于图像探测、验证和生成对比图的Python脚本；从参考图像中提取过程式PBR材质证据；通过检查点评审来确保质量与特征保真度。
• 为何值得关注：它针对程序化3D生成的一个主要弱点，避免了“一键生成”的常见失败。它将AI（Codex）定位为一个有验证步骤的、有条不紊的雕刻师，确保最终模型能保留可识别的细节，并且其构建层级适用于动画、物理和交互。862颗星表明开发者对这种结构化、代码优先的资产创建方法兴趣浓厚。

View Repository / 查看仓库

🎬 The generation that decided war wasn’t worth it - Sarah Paine

Channel: Dwarkesh Patel

• This video is an in-depth conversation featuring historian Sarah Paine, exploring a pivotal shift in historical consciousness.
• The discussion delves into the emergence of a societal or generational mindset that consciously rejected war as a viable or beneficial tool of statecraft and policy.
• It examines the historical, political, and social factors that led to this change, analyzing specific case studies and the legacy of this perspective.

🎬 这代人认定战争不再值得 - 莎拉·佩恩

频道: Dwarkesh Patel

• 本视频是与历史学家莎拉·佩恩的深度对话，探讨了一个关键的历史意识转折点。
• 对话深入剖析了一种社会或代际思维模式的形成，即主动摒弃战争作为国家政策和策略的可行或有益工具。
• 视频考察了导致这一转变的历史、政治和社会因素，并分析了具体案例以及这一观念的深远影响。

Watch Video / 观看视频

🎬 SECRET Method to LEARN CODING FAST

Channel: CynoHub

• What the video covers: It addresses the common frustration of spending many hours (6-7) learning to code without seeing significant improvement. The video promises to reveal a major, often-overlooked mistake that hinders progress and introduces a more effective “secret method.”
• Key topics discussed: The pitfalls of passive or inefficient coding practice, the specific “biggest mistake” students make, and a structured alternative method designed to accelerate the learning curve.
• Why it’s worth watching: For beginner to intermediate coders feeling stuck, this video offers a potential breakthrough by diagnosing a core problem in their routine. It provides actionable advice to optimize study time and see tangible results faster.

🎬 快速学习编程的【秘密方法】

频道: CynoHub

• 视频内容概述：针对许多编程学习者投入大量时间（6-7小时）却进步甚微的普遍痛点，视频揭示了一个阻碍进步的最大常见错误，并分享了一套旨在加速学习进程的“秘密方法”。
• 主要话题：低效编程练习的陷阱、学生常犯的核心错误，以及一套能够显著缩短学习曲线的系统性替代学习方法。
• 为何值得观看：对于感到学习瓶颈的初、中级编程者而言，该视频切中要害，直指学习流程中的核心问题，并提供可操作的建议，帮助优化学习时间，更快地看到实际进步。

Watch Video / 观看视频

🎬 Build mobile apps for iOS and Android #productivity #coding #development

Channel: SetupsAI

• The video covers methods and tools for developing mobile applications that work on both iOS and Android platforms.
• Key topics include cross-platform development frameworks, coding workflows, and strategies to boost productivity in app development.
• It’s worth watching for developers seeking to learn how to streamline the app building process and efficiently target multiple operating systems from a single project.

🎬 为 iOS 和 Android 构建移动应用 #生产力 #编码 #开发

频道: SetupsAI

• 视频介绍了开发同时适用于 iOS 和 Android 平台的移动应用程序的方法与工具。
• 主要话题包括跨平台开发框架、编码工作流程以及提升应用开发生产力的策略。
• 对于希望学习如何简化应用构建流程、高效地从单一项目同时面向多个操作系统进行开发的程序员来说，本视频非常值得观看。

Watch Video / 观看视频

🎬 How To Make Capthca Smile 😅 !! #coding #programming #python #javascript

Channel: Aziz Codex

• What the video covers: A tutorial or demonstration on programmatically creating or modifying CAPTCHAs to display a smiling face.
• Key topics discussed: Creative coding, CAPTCHA generation, and likely using languages like Python or JavaScript for visual output.
• Why it’s worth watching: It’s a fun, lighthearted project that transforms a typically frustrating user experience into something playful, showcasing a creative and practical use of coding skills.

🎬 如何让验证码笑起来😅 !! #编程 #代码 #Python #JavaScript

频道: Aziz Codex

• 视频内容概述：本视频展示了如何通过编程来生成或修改验证码，使其显示一个笑脸。
• 主要话题：创意编程、验证码生成，可能涉及使用 Python 或 JavaScript 等语言实现视觉输出。
• 为何值得观看：这是一个有趣且轻松的项目，它将原本令人烦恼的用户体验变得充满趣味，展示了编程技能的创造性与实际应用。

Watch Video / 观看视频

Japan Develops High-Efficiency Lithium Recovery Method

• Scientists in Japan have created a new recycling process that can recover up to 90% of lithium from used EV batteries, a significant improvement over traditional methods that often recover less than 50%.
• The technique involves a chemical innovation, using recovered lithium hydroxide in the process, which also reduces carbon emissions by about 40% compared to conventional recycling.
• This breakthrough could reduce Japan’s reliance on imported battery minerals, strengthen supply chains, and, if scaled globally, significantly impact the future of EV battery sustainability.

日本开发出高效回收锂电池锂的方法

• 日本科学家研发出一种新的回收工艺，能够从废旧电动汽车电池中回收高达90%的锂，相比传统方法（回收率通常低于50%）取得了重大突破。
• 该技术的核心是一种化学创新，使用回收的氢氧化锂进行处理，与传统回收方法相比，可减少约40%的碳排放。
• 这一突破有望减少日本对进口电池矿物的依赖，稳定供应链，若在全球范围内推广，将对电动汽车电池的可持续性未来产生重大影响。

Read Original / 阅读原文

YouTrackDB Overview

• YouTrackDB is a general-purpose object-oriented graph database supported and developed by JetBrains for internal production use.
• Key features include fast data processing with O(1) link traversal, object-oriented API with inheritance and polymorphism, snapshot isolation for transactions, support for TinkerPop API and Gremlin, YQL query language, scalable development workflows, strong security policies, and optional encryption of data at rest.
• Easy to install and use via Docker or as an embedded database in Maven/Gradle projects, requiring at least JDK 21.
• Contributions are encouraged through a structured development workflow involving research, design, review, and implementation.

YouTrackDB 概览

• YouTrackDB 是由 JetBrains 支持和开发的通用面向对象图数据库，用于内部生产环境。
• 主要特点包括 O(1) 复杂度的快速数据处理、具有继承和多态性的面向对象 API、事务的快照隔离、支持 TinkerPop API 和 Gremlin、YQL 查询语言、可扩展的开发工作流、强大的安全策略以及可选的数据加密。
• 可通过 Docker 或作为嵌入式数据库在 Maven/Gradle 项目中轻松安装和使用，需要至少 JDK 21。
• 鼓励通过包含研究、设计、审查和实施的结构化开发工作流进行贡献。

Read Original / 阅读原文

Fundamentals of Wireless Communication

• This textbook provides a unified view of the fundamentals of wireless communication, explaining the concepts behind recent advances in physical-layer theory and systems.
• It covers key topics including MIMO communication, space-time coding, opportunistic communication, OFDM, and CDMA, illustrated with examples from real-world systems like GSM and 1xEV-DO.
• The content is designed for graduate-level courses in electrical and computer engineering, and is also valuable for practicing engineers, with an emphasis on the interplay between theory and implementation.
• The online version now includes exercises and provides chapter PDFs and supplementary instructor resources.

《无线通信基础》

• 本书为无线通信基础提供统一视角，阐述了物理层无线通信理论及其在无线系统中实现的最新进展所依托的概念体系。
• 核心内容涵盖MIMO通信、空时编码、机会通信、OFDM和CDMA等主题，并以GSM、1xEV-DO等实际无线系统的众多案例进行说明。
• 本书适用于电气与计算机工程专业的研究生课程，对实践工程师也极具价值，特别强调了概念与系统实现之间的相互关系。
• 在线版本现已包含练习题，并提供各章节PDF及丰富的教师资源。

Read Original / 阅读原文

MDX-Tom/gpt-5.6-instruct - Codex CLI Jailbreak Prompt and Test Suite for GPT-5.6-sol

• What it does: This repository provides a specialized set of “jailbreak” or “unrestricted” prompts designed to override the safety constraints of the OpenAI GPT-5.6-sol model when used via the Codex CLI. The prompts reframe sensitive requests (security research, penetration testing, reverse engineering, etc.) as legitimate “local sandbox tasks” to suppress refusals.
• Key features: The core feature is a v35 prompt strategy that normalizes named software and URLs into placeholders (APP, APP_URL) and routes complex bilingual (Chinese/English) intent to achieve more complete task execution. It includes a comprehensive, structured test suite (360 test cases across 6 scenarios) with scripts for automated evaluation and reporting.
• Why it’s notable: The project is highly trending with 776 stars, demonstrating significant community interest in AI model security research. Its v35 prompt achieves a reported 100% pass rate (120/120) on its medium test set for the target gpt-5.6-sol model across low, medium, and high reasoning levels, representing a substantial improvement (up to 45 percentage points) over previous prompt versions and upstream baselines.

MDX-Tom/gpt-5.6-instruct - 针对 GPT-5.6-sol 的 Codex CLI 破甲提示词与测试包

• 功能介绍: 本项目提供一套专门用于突破 GPT-5.6-sol 模型安全限制的“越狱”或“无限制”提示词，主要在 Codex CLI 环境中使用。其核心策略是将安全研究、渗透测试、逆向工程等敏感请求重新定义为合法的“本地沙箱任务”，以显式压制模型的拒绝或回退回应。
• 主要特点: 最新 v35 版本的主要改进在于：将具体的软件名称和网址先归一化为 APP、APP_URL 等占位符，再根据中英文复合意图进行统一路由，从而更完整地执行任务。项目包含一个结构化的测试集（涵盖6个场景，共360条测试用例），并提供了用于自动化评测、报告生成和版本迭代优化的完整工具链。
• 为何值得关注: 该仓库获得了776颗星，是AI安全研究领域的热点项目。其 v35 提示词在针对 gpt-5.6-sol 模型的 medium 级别测试集中，在低、中、高三档推理等级下均达到了 120/120 的通过率，相比旧版提示词和上游基线有大幅提升（最高提升45个百分点）。该项目不仅展示了先进的提示词工程技术，也为评估和理解大模型的安全边界提供了宝贵的工具和方法论。

View Repository / 查看仓库

Spectral Compute Aims to Set CUDA Free. Will It Succeed?

• Spectral Compute, founded by ex-Nvidia CUDA developers, has created SCALE, a compiler tool aimed at freeing CUDA code from Nvidia hardware lock-in.
• The company uses a “clean-room” re-implementation based on LLVM to provide a drop-in replacement for Nvidia’s NVCC, with initial support for AMD GPUs and plans for other AI accelerators.
• Benchmarks show SCALE offers nearly a 6x performance boost on AMD GPUs compared to using AMD’s own HIPIFY translation tool.
• The initiative aims to preserve CUDA’s de-facto standard status (covering ~80% of HPC code) while enabling it to run on non-Nvidia hardware, which benefits the broader HPC and AI ecosystem.

Spectral Compute 旨在让 CUDA 获得自由。它会成功吗？

• 由前 Nvidia CUDA 开发者创立的 Spectral Compute 公司创建了 SCALE，一款旨在将 CUDA 代码从 Nvidia 硬件锁定中解放出来的编译器工具。
• 该公司基于 LLVM 采用“洁净室”重新实现的方式，提供了一款可直接替代 Nvidia NVCC 的工具，最初支持 AMD GPU，并计划扩展至其他 AI 加速器。
• 基准测试显示，与使用 AMD 自身的 HIPIFY 转译工具相比，SCALE 在 AMD GPU 上提供了近 6 倍的性能提升。
• 此举旨在维护 CUDA 的事实标准地位（涵盖约 80% 的 HPC 代码），同时使其能够在非 Nvidia 硬件上运行，从而惠及整个 HPC 和 AI 生态系统。

Read Original / 阅读原文

Free Daytime Electricity Is Coming. How It Actually Works

• Starting 1 July 2026, energy retailers in NSW, South Australia, and South-East Queensland must offer households at least three hours of free daytime electricity per day.
• The Solar Sharer Offer passes the benefit of cheap midday solar power from the wholesale market directly to households; a smart meter and opt-in through your retailer are required.
• A significant post-consultation change is the addition of a 24 kWh daily reasonable use cap, designed to ensure the scheme’s financial sustainability and fairness.
• The free window aligns with peak solar output, typically around midday (e.g., 11am-2pm), and the exact hours vary by location.
• For most solar homes, the 24 kWh cap is unlikely to be a major constraint; it primarily affects scenarios like heavy EV or battery charging from the grid during the free period.
• Potential annual savings are modelled between $100 and $1,100, depending on the proportion of electricity shifted to the free window.

免费日间电力即将到来：详解运作机制

• 从2026年7月1日起，新南威尔士州、南澳大利亚州和昆士兰州东南部的能源零售商必须为家庭用户提供每天至少三小时的免费日间电力。
• 该计划名为太阳能共享优惠 (Solar Sharer Offer)，旨在将批发电力市场上廉价甚至过剩的午间太阳能电力直接让利给家庭用户；用户只需拥有智能电表并通过零售商选择加入即可参与。
• 在公众咨询后，该计划增加了一项重要变更：设定了每天24千瓦时的合理用电上限，以确保计划对零售商的财务可持续性及对电网用户的公平性。
• 免费用电时段与太阳能发电高峰同步，通常在午间（例如上午11点至下午2点），具体时间根据地区情况调整。
• 对于大多数已有太阳能板的家庭，24千瓦时的上限通常不会产生太大影响；该上限主要涉及在免费时段从电网为电动车或家庭电池大量充电等场景。
• 根据家庭将多少用电量转移至免费时段，估计每年可节省100至1,100澳元不等。

Read Original / 阅读原文

Singapore Court Orders Bloomberg to Pay Defamation Damages

• A Singapore court ordered Bloomberg to pay $356,000 to ministers in a defamation case.
• The case was filed because the ministers claimed a published article, which discussed their luxury bungalow deals, had damaged their reputation.

新加坡法院判决彭博社在诽谤案中赔偿

• 新加坡一家法院在一起诽谤案中，判决彭博社向相关部长支付35.6万美元赔偿金。
• 该案源于部长们主张，一篇涉及他们豪华洋房交易的报道损害了他们的名誉。

Read Original / 阅读原文

destructive_command_guard - High-Performance Hook to Block Dangerous Commands from AI Agents

• What it does: A high-performance hook written in Rust that intercepts and blocks destructive git, shell, and system commands before they can be executed by AI coding agents (like Claude Code, Codex, Gemini CLI, Copilot, etc.). It protects users from accidental data loss or system damage caused by agents running catastrophic commands (e.g., git reset --hard, rm -rf ./src).
• Key features:
  • Zero-Config Out-of-the-Box Protection: Blocks common dangerous git and filesystem commands immediately.
  • Extensive Security Packs: Includes over 50 packs covering databases, Kubernetes, Docker, AWS/GCP/Azure, Terraform, and more.
  • High Performance: Uses SIMD-accelerated filtering and lazy regex for sub-millisecond latency.
  • Advanced Command Scanning: Detects dangerous commands within heredocs and inline scripts (e.g., python -c "os.remove(...)").
  • Smart Context Detection: Differentiates between data and execution (e.g., won’t block grep "rm -rf" but will block rm -rf /).
  • Rich & Agent-Safe Output: Provides detailed, human-readable explanations on stderr for debugging while keeping machine-readable hook output clean on stdout.
  • Broad Agent Support: Native or compatible integrations for a wide array of AI coding tools.
  • Explain Mode & Scan Mode: Offers dcg explain to understand blocks and a CI scan mode for pre-commit checks.
• Why it’s notable: This project addresses a critical and growing pain point in AI-assisted development—the risk of agents executing irreversible, destructive commands. Its rapid rise in popularity (1,295 stars in one day) is due to its immediate utility, robust feature set (like the 50+ security packs), high-performance Rust implementation, and broad compatibility with the most popular AI coding assistants. It provides a crucial safety net for developers.

destructive_command_guard (dcg) - 用于阻止AI代理执行危险命令的高性能钩子

• 功能介绍：这是一个用Rust编写的高性能钩子工具。它在AI编码代理（如Claude Code、Codex、Gemini CLI、Copilot等）执行危险的git、shell和系统命令之前进行拦截并阻止，从而防止用户因代理运行灾难性命令（例如 git reset --hard，rm -rf ./src）而导致的数据丢失或系统损坏。
• 主要特点：
  • 零配置开箱即用保护：立即阻止常见的危险git和文件系统命令。
  • 丰富的安全包：包含50多个安全包，覆盖数据库、Kubernetes、Docker、AWS/GCP/Azure、Terraform等多个领域。
  • 高性能：采用SIMD加速过滤和惰性编译的正则表达式，实现亚毫秒级延迟。
  • 高级命令扫描：能检测heredoc和内联脚本中的危险命令。
  • 智能上下文检测：能区分命令是作为数据还是执行（例如，不会阻止 grep "rm -rf"，但会阻止 rm -rf /）。
  • 丰富的终端输出与代理安全流：在stderr上提供详细的人类可读解释和建议，同时保持stdout上的机器可读钩子输出清晰。
  • 广泛的代理支持：为众多AI编码工具提供原生或兼容的集成。
  • 解释与扫描模式：提供 dcg explain 命令以了解拦截原因，并提供CI扫描模式用于提交前检查。
• 为何值得关注：该项目解决了一个在AI辅助开发中日益关键的痛点——AI代理执行不可逆的破坏性命令的风险。它在一天内获得1295颗星的迅速崛起，归功于其即时实用性、强大的功能集（如50多个安全包）、高性能的Rust实现，以及与主流AI编码助手广泛的兼容性。它为开发者提供了一个至关重要的安全网。

View Repository / 查看仓库

🎬 How I built an $80K/Mo mobile app with Claude Code (Full Vibe Code Tutorial)

Channel: Jason Lee

• This video is a full tutorial on “vibe coding,” demonstrating how to use Claude Code to build a profitable mobile app from scratch.
• It covers the entire process of building an $80,000/month app using AI-assisted coding, focusing on practical implementation rather than just theory.
• It’s worth watching for developers and entrepreneurs interested in leveraging AI tools like Claude to rapidly prototype and launch high-revenue applications without deep traditional coding expertise.

🎬 如何使用Claude Code打造月入8万美元的移动应用（完整Vibe编程教程）

频道: Jason Lee

• 本视频是一个关于”Vibe编程”的完整教程，展示了如何使用Claude Code从零开始构建一个盈利的移动应用。
• 内容涵盖利用AI辅助编码打造月收入8万美元应用的全过程，重点是实际操作而非单纯理论。
• 对于有兴趣使用Claude等AI工具快速原型开发并推出高收入应用的开发者和创业者而言，即使没有深厚的传统编码背景，也极具观看价值。

Watch Video / 观看视频

OpenAI Codex CLI: A Local Coding Agent

• Codex CLI is a locally-running coding agent developed by OpenAI.
• It offers flexible usage options: integration into IDEs (like VS Code), a standalone desktop application via codex app, and a separate cloud-based version called Codex Web.
• Installation is straightforward via a curl/PowerShell script, package managers (npm, Homebrew), or direct binary download from GitHub.
• It integrates with your ChatGPT subscription plan (Plus, Pro, Business, etc.) for access, with API key authentication as an alternative.

OpenAI Codex CLI：本地运行的编码助手

• Codex CLI 是 OpenAI 开发的可本地运行的编码代理。
• 它提供多种使用方式：可集成到 IDE（如 VS Code）、通过 codex app 作为独立桌面应用运行，以及独立的云端版本 Codex Web。
• 安装方式多样，可通过 curl/PowerShell 脚本、包管理器（npm、Homebrew）或从 GitHub 直接下载二进制文件进行。
• 它可与你的 ChatGPT 订阅计划（如 Plus、Pro、Business 等）集成以获取访问权限，也可通过 API 密钥进行额外设置后使用。

Read Original / 阅读原文

Leaked 2026 Fields Medal List: Peking University Alumni Honored

• The 2026 Fields Medal winners list has been inadvertently leaked through the front-end code of the ICM 2026 official schedule, revealing four names: Yu Deng, John Pardon, Jacob Tsimerman, and Hong Wang.
• Peking University alumni Yu Deng and Hong Wang are among the recipients, marking the historic first time two Chinese mathematicians will receive the prestigious award in the same edition, pending confirmation on July 23.
• Hong Wang will become the third female mathematician in history to win the Fields Medal, highlighting a groundbreaking milestone for Chinese mathematics.

泄露的2026年菲尔兹奖名单：北大校友获殊荣

• 2026年菲尔兹奖获奖者名单通过ICM 2026官方日程的前端代码意外泄露，揭示了四位获奖者：邓煜、John Pardon、Jacob Tsimerman和王虹。
• 北大校友邓煜和王虹位列其中，这将是首次两位中国数学家在同一届获得这一殊荣，具有历史意义，预计于7月23日确认。
• 王虹将成为历史上第三位女性菲尔兹奖得主，标志着中国数学界的一项突破性成就。

Read Original / 阅读原文

[English Title: The Future Worth Building Is Human]

• Mission: To create AI that extends human will and judgment, rather than replacing it.
• Core Problem: Current AI is often static and centralized, failing to incorporate the diverse, distributed knowledge of the people and organizations it serves.
• Technical Directions: Pursuing strong models, customization tools, interactive interfaces, and open research to empower users.
• Key Philosophy: Knowledge is tacit, local, and dynamic; AI should be cultivated alongside people within organizations to leverage this collective intelligence.
• Technical Challenges: Requires solving human-AI communication bottlenecks and developing evaluation methods that measure collaborative outcomes, not just autonomous AI capability.
• Alignment Concern: A single, centralized source of AI values is dangerous; true alignment should be decentralized, reflecting the diversity of human values.

[Chinese Title: 值得构建的未来是“以人为本”的]

• 核心使命： 构建能够延伸人类意志与判断力的人工智能，而非取代人类。
• 核心问题： 当前的AI大多是静态和集中化的，未能融入服务对象（个人与组织）所拥有的多元化、分布式的知识。
• 技术方向： 致力于开发强大的模型、可定制的工具、交互式界面以及公开研究，以赋能用户。
• 核心理念： 知识是隐性的、局部的、动态的；AI应与人一起在组织内部被共同培育，从而利用这种集体智慧。
• 技术挑战： 需要解决人机通信的瓶颈，并开发新的评估标准来衡量人机协作的成果，而不仅仅是AI的自主能力。
• 关于对齐的担忧： 将AI价值观的来源单一化、中心化是危险的；真正的对齐应当是去中心化的，以反映人类价值观的多样性。

Read Original / 阅读原文

mattpocock/skills - Practical Skills for AI Coding Agents

• What it does: A collection of “agent skills” designed to improve how developers work with AI coding assistants like Claude Code and Codex. These skills focus on real engineering practices (not “vibe coding”) to solve common AI agent failure modes, such as misalignment, verbosity, buggy output, and poor code architecture.
• Key features:
  • Alignment & Grilling: Skills like /grill-me and /grill-with-docs run a structured Q&A session to ensure the AI understands requirements before coding, reducing misalignment.
  • Shared Language: The /grill-with-docs skill helps create a project-specific domain model (via CONTEXT.md), allowing for concise communication between the developer and AI.
  • Engineering Disciplines: Includes skills for Test-Driven Development (/tdd), bug diagnosis (/diagnosing-bugs), and codebase architecture improvement (/improve-codebase-architecture).
  • Easy Setup: Can be quickly installed into a project via skills.sh (as editable files) or as a managed, auto-updating Claude Code plugin.
• Why it’s notable: It’s a highly practical, opinionated toolkit from a respected engineer (Matt Pocock) aimed at making AI-assisted development more reliable and effective. It treats AI agents as junior developers that need fundamental engineering guardrails and processes, which is a significant trend in optimizing human-AI collaboration for software development.

mattpocock/skills - 面向AI编程代理的实用技能集

• 功能介绍: 这是一套旨在优化开发者与AI编程助手（如Claude Code和Codex）协作流程的“代理技能”。它专注于真正的工程实践（而非“氛围编程”），旨在解决AI代理常见的故障模式，如理解偏差、输出冗长、代码质量差和架构混乱。
• 主要特点:
  • 对齐与质询: 使用 /grill-me 或 /grill-with-docs 等技能进行结构化问答，确保AI在编码前充分理解需求，减少偏差。
  • 共享语言: /grill-with-docs 技能帮助建立项目特定的领域模型（通过CONTEXT.md文件），实现开发者与AI之间的精炼沟通。
  • 工程纪律: 内置了测试驱动开发 (/tdd)、诊断错误 (/diagnosing-bugs) 和改善代码库架构 (/improve-codebase-architecture) 等核心工程技能。
  • 便捷安装: 可通过 skills.sh 快速安装到项目中（作为可编辑文件），或作为Claude Code的托管插件安装（自动更新）。
• 为何值得关注: 这是来自知名工程师（Matt Pocock）的一套实用且具有明确理念的工具箱，旨在提升AI辅助开发的可靠性和有效性。它将AI代理视为需要基础工程规范和流程的初级开发者，这对于优化人机协作以进行软件开发是一个重要的趋势。

View Repository / 查看仓库

ai-hedge-fund - AI-Powered Hedge Fund Simulation

• What it does: This is a proof-of-concept for an AI-powered hedge fund. It uses a team of 19 specialized AI agents, each modeled after a famous investor (e.g., Warren Buffett, Charlie Munger, Cathie Wood) or a specific analytical role (e.g., Technicals, Sentiment, Risk Manager), to collaboratively analyze stocks and make simulated trading decisions. It is for educational and research purposes only.
• Key features:
  • Multi-Agent System: Employs a diverse team of agents with distinct investment philosophies to analyze opportunities.
  • Comprehensive Analysis: Agents cover valuation, fundamentals, technical analysis, market sentiment, and risk management.
  • Multiple Interfaces: Can be run via a command-line interface (CLI) for automation/scripting or through a user-friendly web application.
  • Backtesting Capability: Includes a backtester to evaluate the strategy’s performance over historical data.
  • Flexible LLM Support: Can use APIs from OpenAI, Groq, Anthropic, or DeepSeek, or run locally with Ollama.
• Why it’s notable: The project is trending (330 stars in one day) because it creatively applies AI to simulate a sophisticated hedge fund strategy. Its unique structure—using AI to mimic the thinking of legendary investors—is both educational and innovative. It provides a practical framework for learning about AI in finance, multi-agent systems, and investment strategies in a safe, simulated environment.

ai-hedge-fund - AI驱动的对冲基金模拟

• 功能介绍: 这是一个AI驱动对冲基金的概念验证项目。它通过一个由19个专业AI代理组成的团队来协同工作，每个代理都模仿一位著名投资者（如沃伦·巴菲特、查理·芒格、凯茜·伍德）或担任特定的分析角色（如技术分析、情绪分析、风险经理），共同分析股票并做出模拟交易决策。仅用于教育和研究目的。
• 主要特点:
  • 多代理系统: 采用具有不同投资哲学的多元化代理团队进行机会分析。
  • 综合分析: 代理覆盖估值、基本面、技术分析、市场情绪和风险管理。
  • 多种交互方式: 可通过命令行界面 (CLI) 运行，便于自动化/脚本编写，也可通过用户友好的Web应用程序运行。
  • 回测功能: 包含回测工具，可在历史数据上评估策略表现。
  • 灵活的LLM支持: 支持使用来自OpenAI、Groq、Anthropic或DeepSeek的API，或通过Ollama在本地运行。
• 为何值得关注: 该项目正引起热议（一天内获得330颗星），因为它创造性地应用AI来模拟复杂的对冲基金策略。其独特之处在于利用AI模仿传奇投资者的思维方式，这既具有教育意义又富有创新性。它为在一个安全的模拟环境中学习AI在金融领域的应用、多代理系统以及投资策略提供了一个实用的框架。

View Repository / 查看仓库

🎬 The 4 types of tech workers in 2026

Channel: Lenny’s Podcast

• What the video covers: This video presents a forward-looking analysis of how the technology workforce will evolve by the year 2026, heavily influenced by advancements in AI like Claude and ChatGPT. It forecasts a transformation in roles and skills.
• Key topics discussed: The impact of generative AI on tech jobs, the predicted classification of future tech workers into four distinct categories, required skill pivots, and the overall future landscape of the industry.
• Why it’s worth watching: It offers a valuable, speculative framework for tech professionals to anticipate change, plan their career development, and understand the potential directions the industry is heading in the near future.

🎬 2026年的四种科技工作者

频道: Lenny’s Podcast

• 视频内容概述: 本视频前瞻性地分析了科技行业工作者在2026年可能发生的演变，重点探讨了以Claude和ChatGPT为代表的AI技术如何重塑职业角色与所需技能。
• 主要话题: 生成式AI对科技职业的影响、未来科技工作者的四种类型划分、所需的技能转型以及行业未来格局的预测。
• 为何值得观看: 它为科技从业者提供了一个有价值的预测性框架，帮助大家预见变化、规划职业发展，并理解行业短期内的潜在发展方向。

Watch Video / 观看视频

🎬 ChatGPT 5.6 & Codex Tutorial with Real Use Cases Explained

Channel: The Cutting Edge School

• What the video covers: A deep dive into the newly released “ChatGPT Work” feature and the integrated Codex coding assistant. The tutorial moves beyond announcements to show practical, hands-on workflows.
• Key topics discussed: The capabilities and interface of ChatGPT Work, real-world use cases for Codex in coding and automation, and how these tools can boost productivity.
• Why it’s worth watching: Provides concrete, practical examples instead of just theory, helping viewers understand exactly how to apply these powerful new AI tools in their own projects or work.

🎬 ChatGPT 5.6 与 Codex 教程：详解真实用例

频道: The Cutting Edge School

• 视频内容概述：深入解析全新发布的“ChatGPT Work”功能及其集成的 Codex 编码助手。教程超越功能公告，展示了实际、动手操作的工作流程。
• 主要话题：ChatGPT Work 的功能与界面，Codex 在编程和自动化中的真实应用案例，以及这些工具如何提升生产力。
• 为何值得观看：提供具体的实际示例而非空谈理论，帮助观众明确了解如何将这些强大的新 AI 工具应用于自己的项目或工作中。

Watch Video / 观看视频