2026-08-06（内容获取于 08/06 05:01）

Moove raises $250M to become the backbone of the robotaxi industry

TechCrunch · 08/06 04:50

Moove is scaling up the autonomous vehicle fleet management side of its business and plans to someday own, not just manage, Waymo robotaxis.

中文介绍 Moove公司近期完成2.5亿美元融资，旨在成为自动驾驶出租车行业的支柱。该公司正扩大其自动驾驶车队管理业务，并计划未来不仅管理，还将直接拥有Waymo的自动驾驶出租车。

How Lightspeed found their newest hire… via Instagram DM

TechCrunch · 08/06 04:48

Lightspeed partners Josh Machiz and Claire Zau stopped by the Equity studio to talk about the strategies behind their growing social media presence and their podcast, Lightwork.

中文介绍 Lightspeed的合伙人Josh Machiz和Claire Zau在Equity节目中分享了其通过Instagram私信招聘新员工的经历。他们讨论了Lightspeed日益增长的社交媒体策略及其播客“Lightwork”，展示了风险投资公司在人才招聘和品牌建设上的创新方式。

Why Lightspeed is going all-in on creator-led venture capital

TechCrunch · 08/06 04:48

Venture firms are turning to creators to build trust with the next generation of founders before a check is ever written. It’s a trend that’s been building with a16z’s acquisition of Erik Torenberg’s Turpentine podcast and OpenAI’s acquisition of TBPN. Lightspeed Venture Partners just made its own n

中文介绍 Lightspeed Venture Partners正全力投入“创作者主导的风险投资”模式。风险投资公司正通过与创作者合作，在新一代创始人签署投资协议前建立信任。这一趋势的例子包括a16z收购Erik Torenberg的“Turpentine”播客和OpenAI收购TBPN。

Klaviyo acquires Elias Torres’ Agency in full-circle reunion for tech founders

TechCrunch · 08/06 04:05

The serial entrepreneur joins the e-commerce company as CPO to lead its AI agents.

中文介绍 电商公司Klaviyo收购了连续创业家Elias Torres的代理机构。Elias Torres将以首席产品官（CPO）的身份加入Klaviyo，负责领导其AI代理业务。此次收购被视为科技创始人之间的一次“圆满重逢”。

Hank Green found the AI problem that YouTube labels can’t catch

Ars Technica · 08/06 03:51

"Slop" isn't the only problem.

中文介绍 知名创作者Hank Green发现了一个YouTube现有标签系统无法识别的AI内容问题。文章指出，除了常见的“劣质AI生成内容”（slop）之外，还有其他更难被察觉的AI问题存在，这给平台内容审核带来了新的挑战。

The Most Dangerous AI Hacking Techniques Still Have Humans in the Loop

Wired AI · 08/06 03:42

Security researcher James Kettle tried to push the limit of AI’s hacking abilities—and discovered how effective it can be when combined with human expertise.

中文介绍 安全研究员James Kettle尝试探索AI的黑客能力极限，发现最危险的AI黑客技术仍需要人类参与。他指出，当AI与人类专业知识相结合时，其攻击效率能显著提高，这突显了“人机协同”在网络安全攻防中的关键作用。

SpaceX claims Starlink Mobile will be better than AT&T, T-Mobile, and Verizon

Ars Technica · 08/06 03:38

SpaceX won't build large cell towers but plans small base stations across US.

中文介绍 SpaceX宣称其“星链移动版”（Starlink Mobile）服务将超越AT&T、T-Mobile和Verizon等美国主要运营商。该公司计划在美国各地部署小型基站，而非建设大型蜂窝塔，以提供更优质的移动通信服务。

Jeff Dean and other top AI researchers are leaving Google to launch their own startup

TechCrunch · 08/06 03:30

The legendary Google executive is joined by other outgoing Google execs in a joint mission to use AI to push forward the process of scientific discovery.

中文介绍 谷歌传奇高管Jeff Dean与其他多位顶尖AI研究人员正离开谷歌，共同创立一家新公司。他们的使命是利用人工智能推动科学发现的进程，旨在开创AI应用的新领域。此次离职预示着AI研究领域可能出现新的格局。

This Atlantic hurricane season is looking like a dud, but there will be a price to pay

Ars Technica · 08/06 03:19

"The models are forecasting something outside the envelope of anything we have ever observed."

中文介绍 本季大西洋飓风季目前异常平静，但科学家警告称这可能需要付出代价。气象模型预测的结果超出了以往的观测范围，暗示着尽管当前风平浪静，未来可能面临意想不到的气候后果或异常现象。

Google's AI shakeup: DeepMind's Hassabis steps aside, senior scientists depart

Ars Technica · 08/06 03:05

Google's AI brain drain continues.

中文介绍 谷歌的人工智能部门正经历一场重大人事变动，其AI人才流失仍在继续。DeepMind负责人Demis Hassabis卸任日常管理职务，多位资深科学家也已离职。这一系列变动对谷歌的AI战略和研发能力构成挑战。

AI Hacks Are Bad. AI Worms and Viruses Will Be Worse

Wired AI · 08/06 02:30

Chinese researchers have shown that AI models have the capacity to act like aggressive and adaptive computer viruses.

中文介绍 中国研究人员已展示，人工智能模型有能力像具有攻击性和适应性的计算机病毒一样运作。这表明，未来的AI病毒和蠕虫可能比现有的AI黑客技术更具破坏性，对网络安全构成严峻挑战。

Review: Spider-Man: Brand New Day reminds us that superhero movies can be good

Ars Technica · 08/06 02:27

It's as much about grounded characters and well-timed emotional beats as CGI-filled action extravaganzas.

中文介绍 影评指出，电影《蜘蛛侠：全新的一天》（Spider-Man: Brand New Day）提醒观众，超级英雄电影可以很优秀。该片成功之处在于其扎实的角色塑造和恰到好处的情感表达，而非仅仅依赖充满CGI的动作场面。

Google Deepmind loses both its CEO and chief scientist as Demis Hassabis and Jeff Dean step down simultaneously

The Decoder · 08/06 02:20

Google Deepmind is overhauling its leadership as Demis Hassabis steps back from day-to-day management to become Alphabet's chief scientist and Jeff Dean leaves Google after 27 years to launch AI startup Discovery Loop. Former Deepmind CTO Koray Kavukcuoglu will take over as Google races to close the

中文介绍 谷歌DeepMind正经历领导层重组，CEO Demis Hassabis卸任日常管理，转任Alphabet首席科学家。而首席科学家Jeff Dean在谷歌任职27年后离职，创立AI初创公司“Discovery Loop”。原DeepMind首席技术官Koray Kavukcuoglu将接任DeepMind的领导职务。

Weeks into explosive diarrhea outbreak, sluggish CDC plans response team

Ars Technica · 08/06 02:11

Reports of cases began in June, and the national tally is now nearing 23,000.

中文介绍 报告指出，在“爆发性腹泻”疫情持续数周后，美国疾控中心（CDC）终于计划组建正式的应对小组。自6月以来，全国病例报告已接近23,000例，CDC的迟缓反应引发关注。

Reddit aims to make ‘karma’ less important for first-time posters with shift to AI moderation tools

TechCrunch · 08/06 02:00

Reddit is expanding its moderation tools and building stronger abuse prevention systems that it says could eventually reduce communities’ reliance on karma and account-age requirements, making it easier for legitimate newcomers to participate.

中文介绍 Reddit正扩大其内容审核工具并强化防滥用系统，转向采用AI进行审核。此举旨在减少社区对“karma”积分和账户注册时长门槛的依赖，从而让新用户能更容易地参与讨论，同时确保平台秩序。

cloudflare/computer

TypeScript · ★ 2,625 · 🍴 120 · 📈 796 stars today

Give your agent a computer 👾

中文介绍 Cloudflare Computer 项目旨在为 AI 智能体提供一个虚拟的、可编程的“电脑”环境。它允许 AI Agent 像人类一样，在沙盒化的环境中执行命令行操作、浏览网页、读写文件等，从而实现更复杂、更具自主性的任务。该项目解决了当前 Agent 普遍缺乏与外部系统深度交互能力的问题，使其能够更好地模拟真实世界中的操作。开发者可利用此工具构建能够自动化软件开发、数据分析、系统管理等多种复杂场景的智能体。

huangruiteng/loopx

Python · ★ 2,024 · 🍴 157 · 📈 327 stars today

Lightweight loop engineering state kernel for long-running AI agent teams. Agent-loop agnostic across Codex, Claude Code, and other coding agents, with durable goals, quota-aware auto-wake, executable todos, evidence logs, and verifiable handoffs.

中文介绍 LoopX 是一个轻量级的循环工程状态内核，专为管理和协调长时间运行的 AI 智能体团队而设计。它提供了一个与具体 Agent 类型（如 Codex、Claude Code 等）无关的框架，能够处理复杂的多步任务流。LoopX 的核心功能包括持久化目标管理、配额感知自动唤醒机制以及可执行的任务调度，确保智能体团队在资源受限的环境下高效、稳定地运行。它解决了大型 Agent 系统在状态管理、资源调度和多 Agent 协作方面的挑战，特别适用于自动化软件开发、复杂问题解决等场景。

TencentCloud/TencentDB-Agent-Memory

TypeScript · ★ 14,958 · 🍴 1,363 · 📈 1,891 stars today

TencentDB Agent Memory is a team-level memory hub for AI Agents — turning conversations, docs, and code into four reusable memory assets (Chat Memory, Skill, LLM-Wiki, Code-Graph) that are governed, shared, and equipped across agents and frameworks.

中文介绍 TencentDB Agent Memory 为 AI Agents 提供本地化的长效记忆解决方案。它采用四层渐进式管道设计，旨在实现无需外部 API 依赖的完全本地化记忆存储与检索，解决了 AI Agent 在需要长期记忆和上下文感知时对外部服务的依赖问题。该项目通过内建机制管理记忆，增强了数据隐私性和系统性能。适用于需要为 AI Agent 构建稳定、独立且具备丰富记忆能力的开发者，例如开发智能客服、个人助理或具备学习能力的自动化系统。

donnemartin/system-design-primer

Python · ★ 361,442 · 🍴 57,637 · 📈 304 stars today

Learn how to design large-scale systems. Prep for the system design interview. Includes Anki flashcards.

中文介绍 system-design-primer 是一个全面的资源库，旨在帮助工程师学习如何设计大规模分布式系统，并为系统设计面试做准备。它涵盖了从核心概念到具体案例的系统设计知识体系，提供结构化学习路径和实践指导，并附赠 Anki 抽认卡辅助记忆。特别适合希望提升系统架构能力或备战科技公司高级职位面试的软件工程师和架构师。

firecrawl/pdf-inspector

Rust · ★ 11,308 · 🍴 752 · 📈 1,583 stars today

Fast Rust library for PDF inspection, classification, and text extraction. Intelligently detects scanned vs text-based PDFs to enable smart routing decisions.

中文介绍 firecrawl/pdf-inspector 是一个高性能的 Rust 库，专为 PDF 文件提供快速检查、分类及文本提取功能。它能智能区分扫描版与文本版 PDF，支持开发者基于内容类型做出智能路由决策，例如决定是否进行 OCR 识别。对于需要自动化处理海量 PDF 文档、实现高效数据提取或构建智能文档工作流的后端服务和应用程序非常有用。

esengine/DeepSeek-Reasonix

Go · ★ 31,520 · 🍴 2,026 · 📈 747 stars today

DeepSeek-native AI coding agent for your terminal. Engineered around prefix-cache stability — leave it running.

中文介绍 `esengine` 推出的 `DeepSeek-Reasonix` 是一个专为终端用户设计的 DeepSeek 原生 AI 编码代理。它旨在为开发者在命令行界面提供智能编程辅助，例如代码生成、补全或问题解答。该项目特别强调了其“前缀缓存稳定性”工程优化，确保代理可以长时间运行并保持高效，减少重复计算。这使得开发者能够无缝地在终端内获取 AI 协助，从而提升编码效率和开发体验。

addyosmani/agent-skills

JavaScript · ★ 81,933 · 🍴 8,819 · 📈 203 stars today

Production-grade engineering skills for AI coding agents.

中文介绍 为 AI 编码代理提供生产级的工程技能，旨在提升 AI 代理在软件开发中的实际能力和可靠性。它可能包含代码生成、测试、重构等实用工具和方法，帮助开发者构建更高效、更可靠的智能编程助手。适用于 AI 工程师和软件开发团队。

obra/superpowers

Shell · ★ 267,229 · 🍴 23,877 · 📈 931 stars today

An agentic skills framework & software development methodology that works.

中文介绍 obra/superpowers 提供一个结合了智能 Agent 能力框架与软件开发方法的项目，旨在有效构建和部署基于 AI Agent 的复杂系统。它定义了一套结构化的方法论，用于组织、管理和编排 Agent 的“技能”，使其能够协同完成复杂的任务。该项目解决了在 Agent 驱动型软件开发中，如何系统化地设计、实现和测试 Agent 能力的问题，帮助开发者和团队更高效地将 AI Agent 集成到实际应用中，从而加速智能系统的迭代与交付。

roboflow/supervision

Python · ★ 48,878 · 🍴 4,617 · 📈 132 stars today

We write your reusable computer vision tools. 💜

中文介绍 Roboflow/supervision 旨在提供一套可复用的计算机视觉工具集，帮助开发者简化常见的 CV 工作流程。它包含了从数据加载、图像预处理到模型推理结果可视化和指标评估等一系列实用功能，如边界框操作、掩码处理、多边形转换以及各种检测结果的绘制工具。该库解决了计算机视觉项目中的重复性任务，使工程师和研究人员能够更高效地构建、训练和部署他们的视觉模型，从而加速应用开发。

vercel/next.js

JavaScript · ★ 141,510 · 🍴 31,680 · 📈 144 stars today

The React Framework

中文介绍 Next.js 是一个基于 React 的全栈 Web 框架，提供了服务器端渲染（SSR）、静态站点生成（SSG）和 API 路由等功能。它解决了传统 React 应用在性能优化、SEO 和复杂数据获取方面的挑战。Next.js 凭借其文件系统路由、图像优化等特性，极大地简化了现代 Web 应用的开发流程，是构建高性能、SEO 友好且易于维护的 React 网站和应用的理想选择。

tailwindlabs/tailwindcss

TypeScript · ★ 96,808 · 🍴 5,547 · 📈 408 stars today

A utility-first CSS framework for rapid UI development.

中文介绍 Tailwind CSS 是一个“工具优先”的 CSS 框架，专为快速 UI 开发而设计。它解决了传统 CSS 框架带来的样式重复和定制限制问题，通过提供大量低级别的功能类 (utility classes)，让开发者直接在 HTML 中构建复杂的用户界面，无需编写自定义 CSS。这种方法极大提升了开发效率和设计灵活性，使得维护和修改样式变得更加直观。Tailwind CSS 特别适合那些需要高度定制化界面、追求开发速度和精细控制视觉表现的前端项目。

uber/ADR

Python · ★ 997 · 🍴 84 · 📈 354 stars today

ADR secures enterprise AI agents through observability, security benchmarking, and threat detection. Deployed at Uber.

中文介绍 Uber 的 ADR 项目致力于保障企业级 AI agent 的安全性。它通过提供可观测性、进行安全基准测试以及实施威胁检测等功能，确保部署在生产环境中的 AI 系统免受潜在攻击和漏洞的侵害。该工具旨在解决企业在应用 AI 技术时面临的安全挑战，尤其适用于那些对数据安全和系统稳定性有高要求的组织，由 Uber 内部实践验证并对外开源，帮助开发者构建更安全的 AI 应用。

lyogavin/airllm

Jupyter Notebook · ★ 29,009 · 🍴 3,117 · 📈 833 stars today

AirLLM 70B inference with single 4GB GPU

中文介绍 AirLLM 致力于解决大型语言模型 (LLM) 推理的硬件限制问题，使得 700 亿参数级别的模型能够在仅有 4GB 显存的单张 GPU 上高效运行。它通过优化模型加载、量化和计算方法，显著降低了 LLM 推理的资源门槛。这对于在本地设备、低成本硬件或边缘设备上部署复杂 AI 应用的开发者和研究人员极为有用。

You already own the most valuable thing in shopping. You just can't reach it.

@fridayresearch_ · 1.9K 粉丝 · 1.6M 阅 · 597 赞 · 451 转

07/31 01:30

Your taste is scattered across forty companies that each own a piece and answer to advertisers. FRIDAY is building the version you own — a personal shopping intelligence that lives on your device and

中文介绍 FRIDAY 正构建一款设备端个人购物智能工具，旨在整合用户分散在四十家公司中的购物偏好数据，将其归还给用户。这款工具将帮助用户管理自有品味数据，摆脱广告商影响，实现更自主的购物体验。

I went on paternity leave. My agent built a $3M pipeline from 2,741 conversations. Here's the code.

@wayne_liang_ · 1.5K 粉丝 · 999.9K 阅 · 637 赞 · 279 转

08/04 00:47

For eight weeks, a clone of me ran our sales front line: 300+ prospect calls a week, 132 new paying customers, 37 enterprise opportunities worth ~$3M. It also made up a price, emailed internal notes

中文介绍 博主分享育儿假期间其AI销售代理的卓越表现。该代理8周内每周处理超300个客户电话，带来了132个新付费客户和37个价值约300万美元的企业商机，并能自动报价、发送内部邮件。帖子还提供了相关代码，展示了AI在销售自动化和营收增长方面的潜力。

Grok Build will rewrite how you use your laptop -A practical guide to your own Autonomous Employee

@XFreeze · 251.7K 粉丝 · 778.3K 阅 · 558 赞 · 80 转

08/05 20:12

Grok Build: What It’s Actually For Most people underestimate it by an order of magnitude. Ask someone what an AI coding agent is for, and they’ll probably tell you it builds apps. That’s true and it

中文介绍 XFreeze 强调 Grok Build 有能力重塑用户使用笔记本电脑的方式，并将其定位为一个「自主员工」。这篇帖子预示将提供一份实用指南，旨在帮助用户充分利用 Grok Build 的高级功能，实现工作流程的自动化和智能化，从而提升效率，革新个人生产力工具的使用体验。

Taste, Judgment and AI

@addyosmani · 407.9K 粉丝 · 363.3K 阅 · 510 赞 · 45 转

08/04 03:03

Taste is recognizing quality and greatness in the absence of all the evidence. Judgment is committing to it in the presence of some risk. It’s the ultimate foundation of great work. When you look at a

中文介绍 博主深入探讨了「品味」与「判断力」对伟大工作的重要性，尤其是在AI时代。他指出，品味是在缺乏证据时识别优质与卓越的能力，而判断力则是在面临风险时做出承诺的勇气。文章强调了这些人类核心特质，可能暗示它们是AI无法完全取代的价值基石。

The Invisible Economy

@RaoulGMI · 1.5M 粉丝 · 90.0K 阅 · 530 赞 · 72 转

08/05 00:38

Within about two years, most economic activity on Earth won't have a human anywhere in it. Think about a single trade today. Someone decides to make it, another person approves it, someone books it,

中文介绍 RaoulGMI 提出了「隐形经济」这一概念。根据其标题和有限的预览文本，这篇帖子很可能探讨了由自动化和人工智能主导的未来经济模式，其中人类的直接参与显著减少。博主旨在引发读者对 AI 如何重塑全球经济结构及其潜在影响的深度思考。

HERMES AGENT + BUZZ BY JACK DORSEY. FULL SETUP GUIDE

@IBuzovskyi · 3.8K 粉丝 · 76.2K 阅 · 505 赞 · 34 转

08/03 20:43

Buzz launched July 21, 2026. Block open-sourced it under Apache-2.0. Desktop builds for macOS, Windows, and Linux shipped the same day. The pitch: Slack + GitHub + AI agents in one window. Built on

中文介绍 博主提供了由 Jack Dorsey 创立的 Buzz 平台的完整设置指南，并提及 HERMES AGENT。Buzz 已于2026年7月21日开源，并发布了 macOS、Windows、Linux 桌面版。该平台旨在整合 Slack、GitHub 与 AI 代理，打造一站式协作工具。帖子分享了其全面设置步骤。

How to Recursively Improve Your Agents

@ashpreetbedi · 20.5K 粉丝 · 73.1K 阅 · 501 赞 · 50 转

08/03 23:33

Today I'm going to show you how to recursively improve your agents. We'll build an agent that starts at 7/10, then run a recursive auto-improvement loop until every probe passes. Here's how it works:

中文介绍 博主分享如何递归地提升AI代理性能的教程。他将演示如何构建一个初始表现为7/10的代理，并通过运行一个递归式的自动改进循环，使其持续优化，直至所有测试探针 (probe) 均顺利通过。这提供了一种实用的AI代理自我优化方法。

How to Build an Autonomous Content Engine With Hermes

@VibeMarketer_ · 29.9K 粉丝 · 56.6K 阅 · 501 赞 · 30 转

08/04 21:26

I'm going to show you how to automate your content ideation system using Hermes kanban, a new multi-agent feature from Nous Research. Every night, it will monitor a mix of 50+ X accounts, newsletters,

中文介绍 VibeMarketer_ 分享了如何利用 Nous Research 发布的 Hermes kanban（一个全新的多代理功能）构建一个自主内容创意系统。该系统每晚会自动监控超过 50 个 X 账号和简报等信息源，以自动化内容构思流程。博主将展示具体实现方法，帮助用户大幅提升内容生产效率和策略制定。

AI slop can look tasteful

@YifeiChen1121 · 224 粉丝 · 51.8K 阅 · 567 赞 · 22 转

08/05 03:48

How nature and serif fonts became the default face of artificial intelligence TLDR: AI branding converged on serifs + nature photos + soft neutrals to feel "humane." It worked… until everything

中文介绍 YifeiChen1121 分析了 AI 品牌形象的一种审美趋同现象：多数 AI 产品倾向于使用衬线字体、自然风光照片和柔和中性色调，以营造「人道」的感觉。这种策略旨在让 AI 更易被接受，初期效果显著，但随着应用的广泛，这种风格已变得过于普遍，可能导致品牌失去独特性。

MiniMax H3: An Open Model Breaking the Boundaries Between Tasks and Modalities

@MiniMax_AI · 107.4K 粉丝 · 42.3K 阅 · 696 赞 · 103 转

07/31 09:53

Today, we're launching MiniMax H3, a general-purpose multimodal generation model. H3 understands unified context across text, images, video, and audio, generating video with native stereo sound, up to

中文介绍 MiniMax 推出 H3 通用多模态生成模型，该模型能理解文本、图像、视频、音频的统一上下文，并能生成带有原生立体声音频的视频内容，旨在打破任务与模态之间的界限。

AI News Roundup!

@AITECHio · 455.4K 粉丝 · 32.6K 阅 · 512 赞 · 110 转

07/31 19:58

ChatGPT Can Now Talk While Running Your Computer OpenAI has taken AI agents another step forward. You can now have a natural voice conversation with ChatGPT while it operates your computer in the

中文介绍 OpenAI 推出 ChatGPT 新功能，用户可与模型进行自然语音对话，同时让它操作电脑。这标志着 AI 智能体技术又向前迈进了一步，增强了人机交互的深度和广度，预示了更高级别自动化应用的可能。

PayBox 101

@moonpay · 421.1K 粉丝 · 32.0K 阅 · 547 赞 · 62 转

08/02 21:18

MoonPay, the global financial technology company powering the movement of value across fiat and digital assets, has launched PayBox, the first payment vault built for AI that lets a person's AI agent

中文介绍 金融科技公司MoonPay推出PayBox，这是首个专为AI设计的支付保管库，旨在让用户的AI代理管理和执行支付，实现法币与数字资产的无缝价值转移，进一步推动AI在金融领域的应用。

Using AI to learn

@Hi_Mrinal · 22.2K 粉丝 · 20.2K 阅 · 528 赞 · 41 转

08/02 18:13

The best way to learn from AI is to really reverse the whole process like to use ai for deep understanding rather than just "getting the answer" revert the typical workflow, instead of asking the AI

中文介绍 博主Hi_Mrinal围绕「利用AI进行学习」这一主题展开分享。推文简要提及AI在学习过程中的应用潜力，可能涉及如何有效整合AI工具、提升学习效率或获取新知识的方法。具体内容预期会探讨AI作为辅助学习工具的各种可能性。

Anthropic Are Buying Rare Books, Feeding Them Into AI, Then Destroying Them.

@ActionModelAI · 57.6K 粉丝 · 6.8K 阅 · 521 赞 · 390 转

07/30 21:12

It sounds like the plot of a dystopian film. But according to recently released court documents, internal company communications, and a reported $1.5 billion settlement, it's something that has

中文介绍 报告指出，Anthropic 被曝购买稀有书籍用于AI训练后销毁，这涉及内部文件和15亿美元的和解金。帖子揭示了AI公司数据获取方式引发的争议，探讨了版权、伦理及数据所有权问题。

TurnSight: Turn-Level Hindsight Self-Distillation for Tool-Integrated Reasoning

👍 17

08/04 08:00

Tool-Integrated Reasoning (TIR) enables LLMs to solve complex tasks through iterative tool interactions. However, existing reinforcement learning methods often rely on trajectory-level supervision, limiting fine-grained credit assignment in long-horizon TIR scenarios. On-policy self-distillation off

中文介绍 TurnSight引入了一种「回合级事后自我蒸馏」方法，旨在解决大语言模型（LLMs）在工具集成推理（TIR）中面临的挑战。现有强化学习方法通常依赖轨迹级监督，限制了在长周期TIR场景中进行细粒度信用分配。TurnSight通过更精细的监督机制，提升LLMs利用工具解决复杂任务的效率和准确性。

PAST-Bench: Benchmarking the Foundations of Recursive Self-Improvement in Personal Agents

👍 28

08/04 08:00

Recursive self-improvement requires agents to turn accumulated experience into better future behavior. Personal AI agents offer a concrete setting for studying this capability because they retain preferences, task histories, tool routines, and learned skills across sessions. Yet whether retained exp

中文介绍 PAST-Bench是一个新的基准，用于评估个人AI智能体递归自我改进的基础能力。该研究指出，递归自我改进要求智能体能将积累的经验转化为更优行为。个人智能体提供了一个具体环境，它们能在不同会话中保留偏好、任务历史、工具例程和学习技能，为深入研究智能体的自我提升机制提供了平台。

When Attention Goes Blind: Numerical Failure in ALiBi Positional Encodings

👍 4

08/04 08:00

We identify a previously overlooked failure mode of ALiBi positional encoding: its linear bias scaling underflows floating-point precision, which zeroes out a large fraction of attention weights and renders the affected attention heads partially blind. We analyze this failure mode, characterize its

中文介绍 研究发现ALiBi位置编码存在一个此前未被注意的故障模式：其线性偏置缩放会导致浮点精度下溢。这使得大量注意力权重变为零，进而导致受影响的注意力头部分「失明」。论文分析了这一故障的根本原因、对模型性能的影响，并探讨了潜在的解决方案，以提高ALiBi编码的数值稳定性。

Video-DeepResearch: Towards the Next-Generation Multimodal Deepresearch Agent

👍 45

08/04 08:00

We introduce Video-DeepResearch (Video-DR), extending multimodal agents from static images to continuous video streams, a setting that demands dense spatiotemporal grounding coupled with open-web exploration. Preliminary evaluations reveal two critical bottlenecks in current models: (1) modality bia

中文介绍 论文介绍了Video-DeepResearch (Video-DR)，旨在将多模态智能体从处理静态图像扩展到连续视频流。这一新设置要求智能体具备密集的时空感知能力，并能进行开放网络探索。初步评估发现，当前模型在实现这一目标时面临两个关键瓶颈。该工作致力于开发下一代多模态深度研究智能体。

JoyAI-Video-Edit: Real-Time Open-Ended Video Editing with Autoregressive Diffusion

👍 76

08/04 08:00

Real-time video editing requires low-latency causal generation with bounded computational resources while preserving source fidelity and long-term temporal consistency. We present JoyAI-Video-Edit, a 16B-parameter autoregressive diffusion framework for real-time, open-ended video editing without acc

中文介绍 JoyAI-Video-Edit是一个160亿参数的自回归扩散框架，旨在实现实时、开放式视频编辑。该系统解决了实时视频编辑面临的低延迟因果生成、有限计算资源、保持源视频保真度以及长期时间一致性等挑战。它为用户提供了高效且灵活的视频编辑体验，推动了视频创作工具的发展。

ReflectRL: Learning from Golden Negative Trajectories via Reflective-to-Direct Reasoning

👍 2

08/04 08:00

On-policy training has emerged as a powerful post-training paradigm for improving the reasoning capabilities of large language models, and is often enhanced by golden trajectories from stronger expert models. However, when the expert fails on harder problems, existing trajectory-guided methods lose

中文介绍 策略内训练能有效提升大语言模型（LLMs）的推理能力，常通过专家模型的「黄金轨迹」增强。然而，当专家模型在复杂问题上失败时，现有方法难以有效利用这些「负面」轨迹。ReflectRL提出通过「反思到直接推理」机制，有效地从这些“黄金负面轨迹”中学习，从而进一步改进LLMs的推理表现。

UniWorld-Design: From Pixel Generation to Layer-Native Design

👍 19

08/04 08:00

We introduce UniWorld-Design, a framework that redefines image generation from flat pixel synthesis to structured visual composition, with semantic RGBA layers as the atomic units of generation, understanding, and editing. Our key insight is that pixels define how an image is rendered, whereas layer

中文介绍 UniWorld-Design是一个将图像生成从平面像素合成重新定义为结构化视觉组合的框架。其核心思想是将语义RGBA层作为图像生成、理解和编辑的原子单位。通过这种层原生设计，该框架能够实现更精细的视觉创作和编辑，突破了传统像素级生成模式的局限性，提高了设计的灵活性和效率。

HalluTruthQA-4K: A Fine-Grained Corpus and Annotation Process for Arabic Hallucination Detection and Truth Verification

👍 0

08/04 08:00

Large language models can generate fluent Arabic answers while introducing factual errors that are difficult to identify and verify. Existing Arabic hallucination resources often assign a binary label to an entire response, indicating whether it is hallucinated or non-hallucinated, but provide limit

中文介绍 针对大语言模型在生成阿拉伯语答案时常出现的事实错误（幻觉），论文推出了HalluTruthQA-4K。这是一个细粒度的语料库，并提供了相应的标注流程，用于检测阿拉伯语文本中的幻觉并进行真相验证。该资源旨在弥补现有阿拉伯语幻觉检测资源仅提供二元标签的不足，以更精细的方式评估和改进LLMs的可靠性。

Logic Before Language: Pre-pretraining on Formal Derivations Fosters Skill Acquisition and Compressibility

👍 1

08/04 08:00

Pre-pretraining language models (LMs) on symbolic data can accelerate and improve natural language acquisition. However, existing pre-pretraining tasks, such as Dyck and procedural algorithms, rely on narrow primitives that fail to capture the expressive capacity of natural language. Moreover, prior

中文介绍 论文提出「逻辑先于语言」的观点，认为在形式推导上进行预预训练能有效促进语言模型的技能习得和可压缩性。针对现有预预训练任务（如Dyck算法）依赖狭窄原语、无法充分捕捉自然语言表达能力的局限，该方法通过更复杂的逻辑训练，为模型更好地学习和理解自然语言奠定基础。

ETA: A New Agentic Paradigm for Embodied Tasks

👍 1

08/04 08:00

When will robots have their ChatGPT moment? Such a breakthrough requires a general-purpose robot that can handle unfamiliar tasks in unfamiliar environments, remain controllable over long interactions, and learn from experience. Today's embodied systems largely follow an end-to-end observation-to-

中文介绍 论文提出了ETA，一种针对具身任务的全新智能体范式。为实现机器人领域的「ChatGPT时刻」，ETA旨在解决现有机器人系统在处理陌生任务、陌生环境适应、长期交互可控性以及从经验中学习等方面的挑战，超越了当前主流的端到端具身系统，以期开发出更通用的机器人。

Sparse Weight Decomposition for Efficient Circuit Extraction

👍 0

08/04 08:00

Dense pretrained transformers do not naturally expose interpretable units for circuit extraction. Existing approaches obtain such units by learning auxiliary sparse representations or training sparse models, incurring substantial additional computation while potentially introducing a fidelity gap be

中文介绍 密集的预训练Transformer模型在「电路提取」中缺乏可解释的单元。现有方法通过学习辅助稀疏表示或训练稀疏模型来解决，但会带来大量额外计算。论文提出「稀疏权重分解」方法，旨在更高效地从Transformer模型中提取可解释的电路单元，从而更好地理解模型内部工作机制，同时避免计算资源的显著增加。

MultiGlobeQA: A Multilingual and Globally Diverse Benchmark for Geospatial Reasoning

👍 0

08/04 08:00

Geospatial reasoning, i.e., computing distances, containment, and other spatial relations over real-world entities, is central to navigation and logistics, yet large language models (LLMs) struggle with the required geometric and topological computation despite storing considerable geographic knowle

中文介绍 论文推出了MultiGlobeQA，这是一个多语言且全球多样化的地理空间推理基准。尽管大语言模型（LLMs）存储了大量相关信息，但在处理计算距离、包含等真实世界实体的几何和拓扑空间关系时，其地理空间推理能力仍面临挑战。MultiGlobeQA旨在评估并推动LLMs在该关键领域的进步，这对导航和物流至关重要。

ContinualSkillBench: Can LLM Agents Truly Evolve Their Capabilities?

👍 11

08/04 08:00

Modern agent frameworks equip large language models with external skill libraries to solve complex tasks. However, it remains unclear whether these systems can effectively evolve their skills and whether the resulting skills improve task-solving capabilities. To bridge this gap, we introduce Continu

OmniPack: Unified Token Compression for Efficient Omni-modal Large Language Models

👍 23

08/04 08:00

Omni-modal large language models (Omni-LLMs) have achieved remarkable performance on audio-visual understanding tasks, but processing long and highly redundant visual and audio token sequences incurs substantial computational overhead, demanding aggressive token compression for efficient deployment.

Efficient Knowledge Distillation for LLMs: Offline Top-K Logits and a Fused Chunked KL Loss

👍 1

08/04 08:00

Small language models are often the only option for deployment under tight latency, cost, and on-premises constraints, but they are rarely trained from scratch: a compressed model is usually recovered through knowledge distillation (KD). This recovery step largely decides the final quality, yet it i

LegalPincite: Multi-level Legal Information Retrieval Dataset

👍 2

08/04 08:00

A common task in legal Information Retrieval (IR) is to find relevant legal sources from case-law collections. While legal practice often requires pinpoint citations (pincites) to specific case paragraphs, most existing public legal IR datasets lack paragraph-level citation annotations. Yet, publicl

Risky Business: Measuring The Faithfulness-Safety Tension

👍 0

08/04 08:00

Chain-of-Thought (CoT) reasoning offers a promising window into model monitoring. However, monitoring relies on faithfulness, i.e., the model output strictly derives from its reasoning trace. We identify an alignment tension where a model must be faithful enough to be monitored, yet robust enough to

When Agents Learn to Be You: Benchmarking Privacy Leakage, Impersonation Risk, and Defenses in Persona Skills

👍 8

08/04 08:00

Persona skills distill personal interaction histories into portable and executable artifacts for downstream agents. While enabling flexible personalization, this process concentrates fragmented personal signals, amplifies their impact through reuse, and challenges defenses designed for individual re

When Teachers Mislead: Spurious-Signal-Aware On-Policy Distillation

👍 0

08/04 08:00

On-Policy distillation (OPD) transfers teacher capabilities by supervising student-sampled trajectories with dense token-level teacher signals. Recent selective OPD methods improve this process by prioritizing signals that are confident, informative, or learnable. However, the assumptions overlook a

Finite-field Krasner quotients: isomorphism thresholds, characteristics, and censuses

👍 0

08/04 08:00

We study Krasner quotient hyperfields arising from finite fields, F_q/G_r, where G_rle F_q^times has index r. Building on the structure theorem of Baker--Jin, we determine the characteristic and C-characteristic of all sufficiently large such quotients: they depend only on the parity of r and, when

Predictive Enhancement Calibration for Latent Breast MRI Virtual Contrast Enhancement

👍 0

08/04 08:00

Virtual contrast enhancement (VCE) synthesizes enhanced breast MR images from pre-contrast acquisitions. Modern latent generators offer strong image priors, but their bounded natural-image autoencoders conflict with the non-canonical intensity scale of MRI. We show that the upper bound can alter rad

AI-Assisted Peer Review Across Research Communities: From Reviewer AI Policies to LLM Review Quality

👍 0

08/04 08:00

AI-assisted peer review is increasingly discussed and adopted as a tool to support the scientific publishing process, yet there is little systematic understanding of how publication venues regulate its use or of how capable current AI review systems are. We address these questions by first surveying

SkillJack: Persistent Skill Backdoors in Self-Evolving Agents

👍 19

08/04 08:00

Self-evolving agents increasingly convert interaction histories into reusable skills that persist beyond individual tasks. While prior work studies memory and retrieval poisoning, such attacks only affect agents when poisoned records are retrieved as context. We uncover a new and more fundamental ri

ChronoLens: Measuring Language Change Across Time, Languages, and Linguistic Levels

👍 1

08/04 08:00

Historical language change affects morphology, syntax, semantics, and pragmatics, yet computational studies typically examine these levels with incompatible representations and therefore cannot determine whether they evolve together across languages. We address this problem by asking how the magnitu

When Many Answers Are Valid, Voting Fails: Symbolic Verification for Best-of-K Causal Reasoning in LLMs

👍 2

08/04 08:00

Self-consistency assumes the most frequent answer among sampled reasoning traces is the most reliable, but this can fail in causal reasoning: samples often repeat the same confounding error, and votes fragment across multiple valid answers, letting an invalid answer win despite a valid minority trac

LLaDA MoE v2: Scaling Mixture-of-Experts Diffusion Language Models

👍 24

08/04 08:00

Diffusion language models (dLLMs) offer an alternative to autoregressive (AR) language modeling, yet the scaling behavior of Mixture-of-Experts (MoE) dLLMs remains poorly understood. We systematically characterize how optimization hyperparameters, compute allocation, and architecture scale for MoE d

DataSpace: Benchmarking Data Agents for Verifiable Analytics over Heterogeneous Workspaces

👍 1

08/04 08:00

Data agents enable natural-language analytics over organizational workspaces, where relevant evidence may be scattered across databases, structured files, long documents, and multimedia. Existing benchmarks largely isolate structured querying, retrieval, or open-ended analysis, leaving heterogeneous

OliveGemma: A 3 Billion Visual Language Model for Recognising the Mediterranean & European Diet

👍 0

08/04 08:00

Image based dietary assessment offers a scalable alternative to self reported food diaries, yet fine-grained food recognition remains challenging due to high intra-class variability and visually similar dishes. This study presents OliveGemma, a vision language model for recognising and reasoning abo

State Propagation Also Satisfies: A Complex-Valued State-Space Model for Deterministic State Tracking

👍 0

08/04 08:00

Transformer-based architectures have dominated sequence modeling, largely due to the expressive power of attention mechanisms. However, for a class of deterministic state tracking tasks---such as parity checking, modular counting, and parenthesis matching---attention may be overkill. In this paper,

Multi-Task Multi-Frame Visual Piano Transcription

👍 3

08/04 08:00

Audio-based piano transcription performs well on onset, pitch, and velocity, but the sustain pedal lets sound persist long after key release, so audio systems predict pedal-extended offsets rather than physical key release. Yet existing Visual Piano Transcription (VPT) systems focus on onset detecti

Cloudflare Wallets

08/06 04:55

the programmable wallet for the agentic Internet

中文介绍 Cloudflare Drop是一款简化部署流程的工具，用户只需在浏览器中拖放文件夹，即可立即将其内容部署到Cloudflare平台上，实现快速上线。

ZapDigits MCP

08/06 04:54

The MCP server for marketing data

中文介绍 ZapDigits MCP 提供了一个专为营销数据管理设计的 MCP 服务器，旨在处理和优化企业在市场推广活动中收集到的各类数据，提升数据利用效率。

Capacity Desktop

08/06 04:54

A free Lovable that lives on your Mac

中文介绍 Capacity Desktop 是一款为 Mac 用户设计的免费应用程序。它旨在提供一个可爱的桌面工具，为用户提供日常便利或特定的功能体验。

ngrok AI Gateway

08/06 04:54

One private gateway for every AI model

中文介绍 ngrok AI Gateway 旨在为用户提供一个统一且私有的安全网关，以连接和管理各类人工智能模型及相关服务，简化开发和部署流程。

AdAnt AI

08/06 04:54

Claude for viral, high-converting social ads

中文介绍 AdAnt AI 利用类似于 Claude 的技术，专注于为社交媒体生成具有高转化率和病毒式传播潜力的广告内容。

Wispr Flow Notetaker

08/06 04:53

Meeting notes that get the details right.

中文介绍 Wispr Flow Notetaker 是一款会议记录工具，旨在准确捕捉会议中的所有细节，帮助用户高效记录。

NextDoor.Company

08/06 04:51

Discover startups hiring near you, on a map

中文介绍 NextDoor.Company 提供一个地图服务，帮助用户发现其附近正在招聘的初创公司职位。

Hansel

08/06 04:43

Remember everything you've worked on

中文介绍 Hansel 是一款帮助用户记住所有工作内容和项目进度的工具，旨在提高工作效率和记忆力。

StepGrab

08/06 04:38

Turn any Mac task into a step-by-step guide

中文介绍 StepGrab 能够将任何 Mac 操作系统上的任务转化为详细的、分步式操作指南，方便用户学习和分享。

Keytones

08/06 04:37

Distinct key sounds for uppercase, lowercase & more

中文介绍 Keytones 是一款适用于 Mac 系统的应用程序，为大小写字母键等提供独特的按键声音，以增强用户输入体验。

Cursor 3.0 Tutorial For Beginners (Full Course)

07/31 03:41

中文介绍 YouTube博主Riley Brown发布了一期60分钟的视频教程，旨在帮助用户高效学习AI编程工具Cursor。该教程声称能将超过1000小时的学习内容浓缩呈现，覆盖从初学者到专业水平，使观看者能在短时间内全面掌握Cursor工具的使用技巧。

Claude Code + Codex Can FINALLY Work Together (Buzz AI)

07/30 04:54

中文介绍 YouTube博主Riley Brown（通过Buzz AI）发布视频，探讨了人工智能模型Claude Code与Codex现已能够协同工作。该视频内容可能关注如何整合并利用这两种AI模型，以实现更强大的编程辅助功能或解决复杂的编码问题，标志着AI工具协作能力的新进展。

Master Claude for Excel in 20 Minutes (Full Guide)

07/28 07:23

Opus 5 Is Here… But NEW Claude Voice Is Even Bigger

07/26 04:28

OpenAI just released Codex Voice (It's basically Jarvis)

07/25 00:09

Codex Basically Runs My Company Now. Here’s How.

07/22 06:27

Devin AI: The Full Beginner’s Guide (Better Than Claude Code?)

07/17 02:25

Director by OpenArt

07/17 02:12

Codex New Browser: The Secret Weapon Everyone Should be Using

07/15 21:00

OpenAI Just Merged ChatGPT and Codex. This Changes Everything.

07/13 01:24

AI Coding 推荐把这条提示词加入到约束中

08/05 22:14

7 回复 · 程序员 节点

如何快速理解 ai 生成的代码？

08/05 18:08

9 回复 · 程序员 节点

突然觉得薅 AI 中转站没意思了

08/05 14:43

29 回复 · 程序员 节点

烧了 6B+ token,分享下我实践出来最好的 AGENTS.md

08/05 11:20

53 回复 · 程序员 节点

App Store 账户被停用了

08/04 13:07

18 回复 · Apple 节点

App 有个功能是联网搜索 AppStore 中的图标，这种情况也必须 ICP 备案吗？

08/04 10:19

12 回复 · Apple 节点

Mac 应用上架中国区 App Store，需要在 App Store Connect 填写 ICP（App）备案号？

08/04 08:52

15 回复 · Apple 节点

用起来比较舒服的 Linux 桌面 Niri+DankMaterialShell

08/03 21:40

11 回复 · Linux 节点

PopClip 刚刚更新后，为什么显示要重新付费了？

08/03 17:02

25 回复 · Apple 节点

闲置的 mbp 18 款还能干什么？

08/03 14:43

7 回复 · Apple 节点

I'm switching my phone from Android to Linux

08/06 03:50

20 points · 4 comments

Meta Ran Ads That Contained AI-Generated Child Sexual Abuse Imagery

08/06 03:47

83 points · 28 comments

Muse Code and Muse Spark 1.2

08/06 03:15

74 points · 46 comments

Sula: A Gemini protocol server written in Scryer Prolog

08/06 02:52

19 points · 1 comments

Zed DeltaDB

08/06 02:52

159 points · 65 comments

What happens if you put work into the second dimension?

08/06 02:39

18 points · 11 comments

Born Against, or why hobby programming communities are against LLM usage

08/06 02:37

88 points · 96 comments

Beating GPT-5.6 Sol on retrieval with 100x cheaper open models

08/06 02:18

108 points · 21 comments

Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence (2025)

08/06 02:17

46 points · 28 comments

Atlassian Rovo Exfiltrates Data, Bypassing Controls

08/06 01:23

92 points · 31 comments

Phishers are hijacking legitimate cloud infrastructure

08/06 01:20

32 points · 6 comments

Celld: Self-hosted, distributed Durable Objects

08/06 00:50

74 points · 4 comments

Launch HN: HyperProbe (YC S26) – Agents that do read-only debugging in prod

08/06 00:47

Hi HN, this is Shailendra and Karan here. We are building a fast and safe way for coding agents to debug issues live in production.When prod breaks, it lets Cursor, Claude, and others drop virtual breakpoints or probes safely in your running code, and extract the exact variable values that logs don’

Discovery Loop

08/06 00:19

419 points · 266 comments

Changes at Google DeepMind: Demis Hassabis from CEO to Chair, Jeff Dean departs

08/06 00:05

https://www.axios.com/2026/08/05/google-deepmind-demis-hassa...https://www.reuters.com/business/google-shakes-up-ai-leaders...https://www.discoveryloop.com/, https://news.ycombinator.com/item?id=49184960https:/&#x2F

The Valley of Webhooks

08/05 23:22

101 points · 42 comments

Aristotle quotes on virtue, knowledge, and happiness

08/05 22:06

133 points · 59 comments

Rubin Observatory's first LSST Camera release: 500k galaxies in the COSMOS field

08/05 22:04

66 points · 9 comments

The Entropy of a Markov Chain

08/05 22:00

84 points · 7 comments

Cloudflare OS: an open platform for agents, apps, and work

08/05 21:58

391 points · 210 comments

Building an Advanced Agentic Harness

08/05 21:54

82 points · 39 comments

Painting with Gaussians

08/05 21:34

79 points · 14 comments

Faster Than Ninja

08/05 21:33

77 points · 27 comments

Civilian plane crash in New Mexico tied to military GPS blocking

08/05 19:03

423 points · 215 comments

Position: LLMs Can't Jump

08/05 19:01

216 points · 149 comments

Western Sahara

08/05 06:34

96 points · 59 comments

We let models localize into 16 languages. How we made it read native.

08/04 21:23

7 points · 1 comments

GNU Hurd News 2026-Q2

08/03 00:57

71 points · 43 comments

Goodhart's Law Comes for Every Benchmark You Trust

08/01 02:14

20 points · 7 comments

Discovery of a multicomponent alloy forged by the Hiroshima atomic blast

07/31 04:09

83 points · 31 comments

v2.1.222

08/05 06:39

What's changed Fixed worktree-isolated sessions and their subagents being able to run destructive git commands against the main checkout; isolation now applies to file edits and Bash in every session type Fixed PreToolUse auto-allow hooks bypassing tool restrictions in background agent tasks (summar

中文介绍 Anthropic的Claude Code发布v2.1.222版本，修复了工作区隔离会话及其子代理能对主检出执行破坏性Git命令的问题，并改进文件编辑和Bash的隔离。此外，还修复了PreToolUse自动允许钩子绕过后台工具限制的漏洞。

v2.1.221

08/04 08:14

What's changed [VSCode] Added Focus view: a chat-menu toggle that hides tool activity behind an expandable per-turn summary with a live running-tool indicator, toggled with Ctrl+Alt+F or the "Claude Code: Toggle Focus view" command Added mode: "mask" for sandbox credential files on Linux and WSL — s

中文介绍 Anthropic的Claude Code发布v2.1.221版本，为VSCode引入了「专注视图」功能，该视图可在聊天菜单中切换，通过Ctrl+Alt+F或命令隐藏工具活动，并以可展开的回合摘要及实时运行工具指示器显示。此外，还为沙盒凭据文件增加了「mask」模式。

v2.1.220

07/25 09:35

What's changed Bug fixes and reliability improvements

v2.1.219

07/25 01:14

What's changed Added Claude Opus 5 (claude-opus-5), now the default Opus model — 1M context, fast mode at $10/$50 per Mtok Added sandbox.network.strictAllowlist setting to deny non-allowlisted hosts for sandboxed commands without prompting Added DirectoryAdded hook that fires after /add-dir or the S

v2.1.218

07/23 05:24

What's changed Changed /code-review to run as a background subagent, so review work no longer fills your conversation and keeps stacked slash commands as its review target Added screen-reader announcements of deleted text for word and line deletions (Option+Delete, Ctrl+W, Cmd+Backspace, Ctrl+U, Ctr

v2.1.217

07/22 05:35

What's changed Added emoji shortcode autocomplete in the prompt input: type :heart: to insert ❤️, or :hea for suggestions — disable with the emojiCompletionEnabled setting Added warnings when transcript writes are failing (e.g. disk full) or when session saving is off due to an inherited environment

v2.1.216

07/21 06:14

What's changed Added sandbox.filesystem.disabled setting to skip filesystem isolation while keeping network egress control Fixed a slowdown in long sessions where message normalization cost grew quadratically with the number of turns, causing multi-second stalls and slow resumes Fixed auto mode deny

v2.1.215

07/19 10:56

What's changed Claude no longer runs the /verify and /code-review skills on its own; invoke them with /verify or /code-review when you want them

v2.1.214

07/18 09:20

What's changed Fixed single-segment dir/** allow rules like Edit(src/**) auto-approving writes to nested dir/ directories anywhere in the tree instead of only /dir Fixed a permission-check bypass affecting commands run in Windows PowerShell 5.1 sessions Fixed Bash permission checks to fail closed on

v2.1.212

07/17 08:26

What's changed /fork now copies your conversation into a new background session (its own row in claude agents) while you keep working; the in-session subagent it used to launch is now /subtask Added claude auto-mode reset to restore the default auto-mode configuration, with a confirmation prompt (pa