2026-07-03（内容获取于 07/03 06:10）

Thiel Capital’s Jack Selby nabs stakes in hot startups like Etched through Arizona connections

TechCrunch · 07/03 05:57

Selby's VC firm Copper Sky Capital is currently raising a $300 million second fund, according to a regulatory filing.

中文介绍 Thiel Capital的Jack Selby旗下的风投公司Copper Sky Capital，正通过监管备案募集3亿美元的第二期基金，用于投资Etched等热门初创公司。

The website of the future may assemble itself for every visitor

Latent Space · 07/03 05:25

Adobe is experimenting with “agentic sites” that generate pages around an individual user’s intent. At AIEWF, we talked to Carlos Sanchez about the Web's future.

中文介绍 Adobe正在试验「智能代理网站」，该类网站能根据用户意图自动生成个性化页面。公司在AIEWF会议上与Carlos Sanchez讨论了这一未来网页技术。

Meta has a new app called Pocket that is absolutely nothing like the old Pocket

The Verge · 07/03 05:09

Mozilla shut down the well-loved read-it-later Pocket app last year, and now Meta is launching an app called Pocket with an entirely different, AI-focused pitch: this new app lets you make and share little interactive "gizmos" built from an AI prompt, as reported by Business Insider. Meta CEO Mark Z

中文介绍 Meta推出了名为「Pocket」的全新AI应用，与Mozilla此前已关闭的同名稍后阅读应用截然不同。这款新应用主打通过AI提示词生成并分享互动式「小玩意」（gizmos）。

IQM, Europe’s first public quantum company, admits the future of the tech is uncertain

TechCrunch · 07/03 04:42

IQM, a full-stack quantum company out of Finland, went public on the Nasdaq today at a valuation of about $1.9 billion.

中文介绍 芬兰全栈量子计算公司IQM作为欧洲首家上市量子企业，今日在纳斯达克挂牌，估值达19亿美元。公司承认量子技术未来仍存在不确定性。

The best July 4th sales we found so far

The Verge · 07/03 04:15

July 4th sales are typically a precursor to what we’d see during a mid-July Prime Day, but obviously things are flipped around this year. Last week’s big Prime Day sale is over, yet there are a number of familiar deals still poking around in the week leading up to the nation’s birthday. Best Buy is

中文介绍 7月4日美国独立日促销通常是年中Prime Day前的预热，但今年情况不同，Prime Day已在上周结束。不过，独立日前的一周内仍有许多熟悉的科技产品折扣。

Jersey Mike’s IPO illustrates how bad the AI hype has become

TechCrunch · 07/03 04:11

Just for kicks, I took a look at Jersey Mike's IPO documents. Surely a sandwich shop would have no need to mention AI. But lo and behold.

中文介绍 一项对Jersey Mike's（一家三明治店）IPO文件的分析显示，即便此类传统企业也在其文件中提及AI，这凸显了当前人工智能炒作的过度程度。

Newly discovered PamStealer isn't your typical macOS malware

Ars Technica · 07/03 03:38

The discovery underscores the increased effort being poured into Mac infostealers.

Microsoft launches $2.5 billion "Frontier Company" to embed 6,000 AI engineers inside enterprise clients

The Decoder · 07/03 03:18

Microsoft is investing $2.5 billion in a new unit called "Frontier Company" that puts 6,000 engineers directly at enterprise customers. The goal is to integrate AI into core processes with measurable ROI, not more experimentation. Microsoft is positioning itself as a platform-neutral alternative to

中文介绍 微软正投资25亿美元成立「前沿公司」（Frontier Company），旨在派遣6000名AI工程师直接进驻企业客户，以将AI技术深度融入核心业务流程，追求可衡量的投资回报，而非仅限于实验。

A warning sign about AI’s real cost, courtesy of Google and Amazon

TechCrunch · 07/03 03:14

AI has made it a lot harder for tech companies like Amazon and Google to deliver on their net-zero pledges.

中文介绍 人工智能的能源需求使得亚马逊和谷歌等科技巨头更难实现其“净零排放”承诺，这揭示了AI的真实环境成本。

Meta quietly launches vibe-coded gaming app Pocket

TechCrunch · 07/03 02:44

Meta has quietly launched Pocket, an experimental AI app that lets users generate and share interactive mini games using text prompts.

中文介绍 Meta悄然推出了一款名为「Pocket」的实验性AI应用。这款应用允许用户通过文本提示词生成并分享互动式迷你游戏。

Travel app Hopper to pay $35M in FTC settlement over ‘unfairly’ charging hidden fees

TechCrunch · 07/03 02:39

Hopper will pay $35 million to settle FTC allegations that it used deceptive “dark patterns” to hide fees and mislead travelers about the costs and benefits of services.

中文介绍 旅行应用Hopper将支付3500万美元与FTC（联邦贸易委员会）达成和解。FTC指控Hopper使用“黑暗模式”隐藏费用，并误导旅行者关于服务成本和效益的信息。

Anthropic is discussing a new custom chip with Samsung

TechCrunch · 07/03 02:31

The news comes about a week after OpenAI announced its own custom AI chip in a partnership with Broadcom.

中文介绍 继OpenAI一周前宣布与Broadcom合作开发定制AI芯片后，Anthropic也正在与三星讨论开发其新的定制芯片。

Weber marks down grills and griddles to their best prices ever for July 4th

The Verge · 07/03 02:19

The Weber Spirit E-325 grill with three burners. | Image: Weber If our recent Decoder interview with Weber Blackstone CEO Roger Dahle has you craving freshly grilled meats or veggies, Weber just so happens to have a variety of grills, smokers, griddles, and accessories selling at big discounts ahead

中文介绍 为迎接7月4日独立日，知名烧烤炉品牌Weber正对其多款烤架、平底锅及配件进行史上最低价促销，其中包括三炉头的Weber Spirit E-325烤炉。

Can Cursor Remain a Platform for OpenAI and Anthropic’s Models Inside SpaceX?

Wired AI · 07/03 02:01

Cursor hopes to continue offering third-party AI models after it's acquired by SpaceX, testing the relationships between frontier AI labs.

中文介绍 在被SpaceX收购后，Cursor希望继续提供OpenAI和Anthropic等第三方的AI模型，此举将考验领先AI实验室之间的合作关系。

Anthropic reportedly explores custom chip manufacturing with Samsung while insisting Nvidia still matters

The Decoder · 07/03 01:30

Anthropic is reportedly in talks with Samsung Electronics about manufacturing a custom AI chip. The project is still early, but Anthropic has already hired chip engineers. After OpenAI's "Jalapeño," yet another major AI company is pushing into chip development to cut infrastructure costs. The articl

中文介绍 继OpenAI的「Jalapeño」项目后，Anthropic据报正与三星电子初步洽谈定制AI芯片制造事宜，并已招聘芯片工程师。这反映出主要AI公司为降低基础设施成本，纷纷涉足芯片开发的趋势。

usestrix/strix

Python · ★ 31,984 · 🍴 3,358 · 📈 2,167 stars today

Open-source AI penetration testing tool to find and fix your app’s vulnerabilities.

中文介绍 `strix` 是一个开源的 AI 安全工具，旨在帮助开发者和安全团队自动发现并修复应用程序中的潜在漏洞。该项目利用人工智能技术，模拟“AI 渗透测试员”的行为模式，对应用进行深度分析和攻击模拟，从而有效识别传统安全扫描工具可能遗漏的安全缺陷。它解决了人工安全审计耗时耗力、且容易出错的问题，特别适用于需要持续安全保障的开发流程（如 CI/CD），帮助团队在软件发布前提升应用的整体安全性，降低潜在的安全风险。

JuliusBrussee/caveman

JavaScript · ★ 80,718 · 🍴 4,518 · 📈 866 stars today

🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman

中文介绍 `caveman` 项目是一个针对 Claude Code 的技能，旨在通过一种“原始人式”的简化语言风格，大幅削减大型语言模型（LLM）的 token 使用量。它能将 token 消耗降低约 65%，有效解决了 LLM 交互成本高昂和效率低下的问题。对于需要优化 LLM API 调用、降低运营成本或提升处理效率的开发者和企业而言，这是一个实用的 token 节约方案。

msitarzewski/agency-agents

Shell · ★ 125,400 · 🍴 20,340 · 📈 2,925 stars today

A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables.

中文介绍 agency-agents 项目提供了一个完整的 AI 代理机构框架，通过部署多个具有特定人格和专业技能的 AI Agent，协同完成复杂任务。这些 Agent 涵盖前端开发、社区运营、创意生成和事实核查等多样化角色，旨在模拟人类团队协作，实现自动化决策与执行。它适用于需要集成多种 AI 能力、自动化跨领域工作流的开发者和企业。

hasaneyldrm/exercises-dataset

HTML · ★ 9,204 · 🍴 1,030 · 📈 942 stars today

A comprehensive dataset of 433 fitness exercises. Each entry includes name, category, target muscle group, equipment, instructions, thumbnail image, and animation video.

中文介绍 hasaneyldrm/exercises-dataset 提供了一个包含433种健身运动的综合数据集，每项运动详细收录了名称、类别、目标肌肉群、所需器械、完整说明、缩略图及动画视频。该数据集旨在为健身应用开发、运动科学研究或AI模型训练（如姿态识别）提供高质量、结构化的多媒体资源。开发者可利用此数据快速构建功能丰富的健身指南、训练计划生成器，或用于AI在健身领域的应用开发，极大地降低了数据收集与整理的门槛。

santifer/career-ops

JavaScript · ★ 57,737 · 🍴 11,377 · 📈 322 stars today

AI-powered job search system built on Claude Code. 14 skill modes, Go dashboard, PDF generation, batch processing.

中文介绍 `career-ops` 是一个基于 Claude Code 构建的 AI 驱动的求职系统，旨在自动化和优化整个求职流程。它提供 14 种技能模式，集成 Go 语言开发的仪表盘，支持简历等 PDF 文档生成以及批量处理功能。该项目解决了传统求职耗时费力的问题，适合希望利用 AI 技术高效管理、追踪并自动化求职申请流程的个人求职者。

obra/superpowers

Shell · ★ 244,330 · 🍴 21,676 · 📈 962 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。

ChromeDevTools/chrome-devtools-mcp

TypeScript · ★ 45,065 · 🍴 2,932 · 📈 141 stars today

Chrome DevTools for coding agents

中文介绍 `chrome-devtools-mcp` 是专为“编码代理”（coding agents）设计的 Chrome DevTools 扩展。它将 Chrome DevTools 强大的调试和分析能力引入到 AI 编码代理的开发流程中，帮助开发者监控、理解和优化代理的行为与输出。通过这些工具，开发者可以更有效地调试由 AI 生成或处理的代码，提升编码代理的开发效率和准确性。

browser-use/video-use

Python · ★ 13,740 · 🍴 1,689 · 📈 550 stars today

Edit videos with coding agents

中文介绍 `video-use` 是一个探索通过“编码代理”来编辑视频的创新项目。它旨在颠覆传统的图形界面视频编辑模式，允许用户通过编写代码或使用智能代理来自动化和执行复杂的视频剪辑、特效添加及内容生成等任务。该项目解决了传统视频编辑软件操作繁琐、难以批量处理的问题，为开发者、内容创作者和研究人员提供了一种全新的、更具编程性和自动化潜力的视频制作方式，尤其适合需要进行程序化视频生成、自动化后期处理或实验新颖交互模式的场景。

actions/checkout

TypeScript · ★ 8,147 · 🍴 2,527 · 📈 5 stars today

Action for checking out a repo

中文介绍 `actions/checkout` 是 GitHub Actions 官方提供的一个核心组件，其主要功能是在 CI/CD 工作流中将指定代码仓库的内容拉取到工作流运行器（runner）的环境中。这是几乎所有需要访问仓库代码进行构建、测试、部署或分析等操作的 GitHub Actions 工作流的首要步骤，确保了后续任务能够基于最新的代码进行操作，是自动化开发流程中不可或缺的基础工具。

affaan-m/ECC

JavaScript · ★ 225,133 · 🍴 34,453 · 📈 508 stars today

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

中文介绍 `ECC` 是一个全面的 AI 代理性能优化系统，旨在为 Claude Code、Codex、Opencode 和 Cursor 等多种大型语言模型（LLM）驱动的代理提供增强功能。它专注于提升代理的技能、直觉、记忆、安全性，并采用研究优先的开发方法，以构建更强大、高效且可靠的 AI 代理。该项目适合开发和部署高级 AI 代理的工程师和研究人员，特别是在需要整合多模态能力和优化代理行为的场景中。

HKUDS/Vibe-Trading

Python · ★ 17,296 · 🍴 2,879 · 📈 918 stars today

"Vibe-Trading: Your Personal Trading Agent"

中文介绍 Vibe-Trading 是一个个人交易 Agent 项目，旨在为用户提供智能化的交易辅助。它可能通过分析市场数据、情绪指标等，自动化交易决策和执行策略，帮助用户更好地管理投资组合并捕捉交易机会。该项目适用于希望利用 AI 技术增强交易能力、实现自动化投资或寻求个性化交易策略的个人投资者和量化交易爱好者。

agentskills/agentskills

Python · ★ 21,572 · 🍴 1,374 · 📈 47 stars today

Specification and documentation for Agent Skills

中文介绍 `agentskills` 项目提供了 AI 代理“技能”的标准化规范和详细文档。它旨在建立一个通用的框架，定义代理技能的结构、描述方式以及如何与其他代理或平台进行集成。这对于促进 AI 代理生态系统的互操作性和模块化开发至关重要。开发者、平台提供商及研究人员可以参考这份规范，设计和构建统一且可复用的代理能力，从而简化复杂 AI 系统的开发流程。

openai/codex-plugin-cc

JavaScript · ★ 22,573 · 🍴 1,375 · 📈 448 stars today

Use Codex from Claude Code to review code or delegate tasks.

中文介绍 `codex-plugin-cc` 是一个 OpenAI 官方插件，旨在将 Codex 模型的功能集成到 Claude Code 环境中。它允许用户在 Claude Code 内部直接调用 Codex 的能力，执行代码审查或将特定编程任务委派给 Codex 处理。该插件解决了在不同 AI 模型之间切换的效率问题，为需要利用 Codex 强大代码理解和生成能力的用户提供便利，特别适合开发者进行自动化代码质量检查或智能任务分配。

langflow-ai/langflow

Python · ★ 150,680 · 🍴 9,394 · 📈 74 stars today

Langflow is a powerful tool for building and deploying AI-powered agents and workflows.

中文介绍 `Langflow` 是一个强大的可视化工具，专注于构建和部署 AI 驱动的代理及复杂工作流。它可能通过直观的拖放界面，让开发者能够轻松连接各种组件，如大型语言模型（LLM）、外部工具、记忆模块和数据源，从而快速原型设计并部署 AI 应用。该项目旨在降低开发门槛，赋能工程师和数据科学家，无需编写大量代码即可创建复杂的生成式 AI 解决方案，尤其适用于 RAG 等应用场景。

pytorch/pytorch

Python · ★ 101,213 · 🍴 28,228 · 📈 45 stars today

Tensors and Dynamic neural networks in Python with strong GPU acceleration

中文介绍 PyTorch 是一个广泛使用的开源机器学习框架，专注于提供高效灵活的深度学习研究与开发环境。其核心特性包括张量（Tensors）操作，支持在 Python 中构建动态神经网络，并具备强大的 GPU 加速能力，尤其适用于利用 NVIDIA CUDA 进行高性能计算。PyTorch 以其直观的 API 和动态计算图机制，受到研究人员和开发者青睐，广泛应用于计算机视觉、自然语言处理等领域，是构建和训练复杂深度学习模型的首选工具之一。

harvard-edge/cs249r_book

Python · ★ 25,524 · 🍴 3,059 · 📈 33 stars today

Machine Learning Systems

中文介绍 `cs249r_book` 是来自 Harvard-EDGE 的“机器学习系统”课程书籍或相关学习材料。该项目深入探讨了机器学习模型的端到端生命周期，涵盖了从数据管理、模型训练、评估到生产部署、监控和维护等各个环节。它旨在教授如何设计、构建并扩展实际的 ML 系统，解决生产环境中的挑战。对于希望掌握 MLOps 实践、深入理解机器学习系统工程的学生、工程师和研究人员，这是极具价值的参考资源。

ryanmcdermott/clean-code-javascript

JavaScript · ★ 94,546 · 🍴 12,473 · 📈 11 stars today

Clean Code concepts adapted for JavaScript

中文介绍 `clean-code-javascript` 项目将 Robert C. Martin 的《代码整洁之道》核心理念，专门适配并应用于 JavaScript 编程语言。它提供了一系列实践指南和最佳范例，旨在帮助 JavaScript 开发者编写出更具可读性、可维护性和可扩展性的代码。内容涵盖命名规范、函数设计、面向对象原则、错误处理及代码格式化等，是所有希望提升代码质量、遵循专业开发标准的 JavaScript 工程师的重要参考资料。

27 Hidden Claude Features, Settings & Shortcuts That Most Users Don't Know

@sairahul1 · 121.4K 粉丝 · 2.9M 阅 · 506 赞 · 97 转

06/26 16:47

Most people use Claude Code like a fancy autocomplete. They prompt. They wait. They accept the first output. They are leaving 90% of the power on the table. There are 27 specific moves that separate

中文介绍 此推文列举 27 个大多数用户可能不了解的 Claude 隐藏功能、设置与快捷方式。旨在帮助用户发现并充分利用 Claude 更深层次的能力，以提高效率和操作体验，解锁其未被发现的潜力。

How To Build a One-Person Company Using Claude Cowork

@sairahul1 · 121.4K 粉丝 · 876.6K 阅 · 547 赞 · 83 转

06/27 17:03

Emails. Formatting. Compiling reports. Preparing decks. Organizing files. Researching. Marketing. Writing. SEO. The average knowledge worker spends 60% of their day on above work that doesn't require

中文介绍 帖子讨论如何利用 Claude Cowork 搭建一人公司。博主可能分享了相关策略、工具集成或具体实践方法，旨在为独角创业者提供运营思路和效率提升方案。

PorTAL: Portable Task Adapters for LLMs

@RampLabs · 13.4K 粉丝 · 335.4K 阅 · 507 赞 · 45 转

07/02 02:09

Researcher: Ben Geist Abstract Parameter-efficient fine-tuning (e.g. LoRA) adapts a frozen LLM to a task, but the resulting adapter is locked to one base model. When a new model is released, the

中文介绍 「PorTAL」是一项新研究，旨在解决现有参数高效微调（如 LoRA）的适配器无法在不同基础 LLM 模型间移植的局限。该研究提出可移植任务适配器，使用户无需在新模型发布时重新训练，提高了 LLM 微调的灵活性和效率。

The AI Economy: The Next Chapter

@rickyho_1989 · 9.7K 粉丝 · 296.6K 阅 · 508 赞 · 69 转

06/30 20:23

Part I: The Economics of Intelligence Why the AI industry is about to optimize for intelligence per dollar rather than intelligence itself I have become increasingly convinced that the artificial

中文介绍 博主分析了AI经济的未来发展趋势。他认为，AI行业将从追求「智能本身」转向优化「单位成本智能」，即更注重智能的性价比。这预示着AI应用将迎来一个更注重效率和成本效益的新阶段。

How To Master Fable (Fundamentals Guide)

@milesdeutscher · 671.1K 粉丝 · 243.8K 阅 · 500 赞 · 73 转

07/02 02:13

TL;DR: Everything you need to do to get maximum value from Fable. I guarantee that after you're done reading, you'll have all the necessary tools to quite literally 10x your AI productivity with

中文介绍 分享 Fable 工具的入门指南，旨在帮助用户充分利用 Fable 平台，实现 AI 生产力提升。内容涵盖 Fable 的核心功能和使用技巧，助力用户高效完成 AI 相关任务，最大化工具价值。

$OUST Deep Dive - One of My Favorite Physical AI Plays

@crux_capital_ · 52.8K 粉丝 · 217.5K 阅 · 503 赞 · 56 转

06/28 04:53

This is the most in depth report I have ever written, on a very exciting company. If you read this article you will have a much deeper understanding of what Ouster does, how it plays into Physical AI,

中文介绍 博主发布一篇关于 Ouster 公司的深度分析报告，详尽阐述该公司在“实体 AI”（Physical AI）领域的核心角色、技术及市场潜力，旨在帮助读者深入理解其业务模式。

ORACLE: Official AI Agents Trade on Polymarket

@OracleLimited · 37.6K 粉丝 · 202.9K 阅 · 2.8K 赞 · 562 转

06/29 22:08

In 2026, autonomous AI agents have become one of the most effective strategies on prediction markets. Over 30% of all activity on Polymarket now comes from algorithmic and AI-powered wallets. We

中文介绍 「OracleLimited」分享了对2026年预测市场的未来愿景。届时，自主AI智能体将成为最有效的交易策略之一，Polymarket上超过30%的活动将由算法和AI驱动的钱包完成。该分享描绘了AI在去中心化金融和预测领域深度参与的趋势。

ORACLE: Official AI Agents Trade on Polymarket

@OracleAiTrading · 34.1K 粉丝 · 176.1K 阅 · 2.7K 赞 · 567 转

06/30 02:06

In 2026, autonomous AI agents have become one of the most effective strategies on prediction markets. Over 30% of all activity on Polymarket now comes from algorithmic and AI-powered wallets. We

中文介绍 「OracleAiTrading」提出了一个关于2026年预测市场前景的展望。文中指出，自主AI智能体将主导市场，成为最有效的交易策略，届时Polymarket上逾30%的交易将由算法和AI驱动的钱包进行。这篇推文描绘了AI在未来金融市场中的核心作用。

How to Build a $10,000-Level Website With Animations in Claude Code

@monokern · 1.9K 粉丝 · 175.8K 阅 · 546 赞 · 49 转

06/28 22:57

Agencies charge $5,000 for a portfolio site that looks this good I built mine in 2 hours. Here's exactly how This is the real walkthrough - not a generic template guide I'm using my own portfolio as

中文介绍 博主「monokern」分享了如何使用Claude Code在两小时内构建一个价值5000美元级别的带动画效果的个人作品集网站。他通过自身案例详细展示了具体操作步骤，强调这是一个实用的真实工作流，而非通用模板指南，旨在帮助用户高效利用AI工具实现高质量的网站开发。

How To Become An AI Engineer in 2026 (Without a CS Degree)

@cyrilXBT · 187.0K 粉丝 · 171.8K 阅 · 505 赞 · 91 转

06/29 22:38

There is a sentence sitting on almost every AI engineering job posting that stops people before they even apply. Bachelor's degree in Computer Science required. Most people read that line, close the

中文介绍 针对AI工程职位普遍要求计算机科学学位的问题，博主提供了一份详细指南，指导人们如何在2026年前，无需CS学位也能成为一名AI工程师。内容可能涵盖学习路径、技能培养及就业策略。

Wiki Memory

@hwchase17 · 115.7K 粉丝 · 129.7K 阅 · 500 赞 · 57 转

06/30 22:26

Memory for agents is still early, with little to no standards. “Memory” means something different to everyone. But one common pattern is emerging: wiki memory. The idea is simple: use an agent to turn

中文介绍 讨论 AI 智能体（agent）记忆机制的现状，指出目前缺乏统一标准。博主提出一种正在兴起的「wiki memory」模式，即利用智能体将信息整理成 wiki 形式作为记忆，为智能体提供了结构化、易于检索的记忆解决方案。

Toward Unrestricted Intelligence: Venice Series A

@ErikVoorhees · 908.5K 粉丝 · 124.8K 阅 · 529 赞 · 103 转

07/01 22:28

“If others possess your thoughts and constrain your words, then you exist at their permission, and you are not free.” —Anonymous Venice launched just over two years ago to create a private and

中文介绍 Venice宣布完成A轮融资，其产品旨在提供私密且不受限制的智能服务，以应对言论受限及思想被束缚的担忧。该项目两年前启动，专注于赋能用户掌控自己的数字存在与表达。

AutoMem: Automated Learning of Memory as a Cognitive Skill

👍 0

07/01 08:00

Memory expertise is a learned skill: knowing what to encode, when to retrieve, and how to organize knowledge--a capacity known in cognitive science as metamemory. We bring this perspective to LLMs by treating memory management as a trainable skill. We promote file-system operations to first-class me

中文介绍 这篇论文提出将大型语言模型（LLMs）的记忆管理视为一种可训练的认知技能。受认知科学中元记忆（metamemory）概念的启发，研究人员通过文件系统操作的视角来处理LLMs的记忆，旨在让LLMs学会如何编码、检索和组织知识，从而提升其在复杂任务中的表现，将其发展成一种类似于人类的认知能力。

Ink3D: Sculpting 3D Assets with Extremely Complex Textures via Video Generative Models

👍 3

07/01 08:00

Recent 3D generative models can synthesize high-quality geometry but often struggle to reproduce intricate textures from reference images, largely due to the scarcity of large-scale 3D training data with rich surface appearance. In contrast, visual generative models are trained on datasets several o

中文介绍 Ink3D 提出利用视频生成模型来解决当前3D生成模型在从参考图像中再现复杂纹理方面的挑战。现有的3D模型擅长生成高质量几何结构，但因缺乏大规模带有丰富表面外观的3D训练数据，难以处理精细纹理。Ink3D通过结合视频生成模型的强大能力，旨在克服这一限制，实现3D资产的复杂纹理雕刻。

The State-Prediction Separation Hypothesis

👍 7

07/01 08:00

Transformers use the same forward computation stream to both predict the next token and store useful state for future token predictions. We formulate the state-prediction separation hypothesis: disentangling the two roles yields better language modeling performance. We design a Transformer variant t

中文介绍 这篇论文提出了「状态-预测分离假说」，认为Transformer模型在进行下一个token预测和存储有用状态时，如果能将这两种角色解耦，将带来更好的语言建模性能。目前Transformer使用相同的正向计算流来同时完成这两项任务。研究人员设计了一个Transformer架构，旨在验证这一假说，优化语言模型效率与表现。

RepoRescue: An Empirical Study of LLM Agents on Whole-Repository Compatibility Rescue

👍 2

07/01 08:00

Open-source libraries and tools are widely reused, but compatibility maintenance is expensive. Once maintainers leave, useful repositories can stop working as runtimes and dependencies evolve. We study whether LLM agents can adapt old repositories to modern environments, a task we call compatibility

中文介绍 RepoRescue 是一项关于大型语言模型（LLM）智能体在整个代码库兼容性修复方面的实证研究。开源库的兼容性维护成本高昂，且随着运行时和依赖项的演变，旧代码库可能停止工作。研究探讨了LLM智能体是否能够有效地使旧有代码库适应现代环境，解决因维护者流失而导致的兼容性问题，以挽救有价值的项目。

Are Performance-Optimization Benchmarks Reliably Measuring Coding Agents?

👍 4

07/01 08:00

Repository-level performance-optimization benchmarks such as GSO, SWE-Perf and SWE-fficiency evaluate coding agents by applying patches to real repositories and comparing runtime against unoptimized baselines and official reference patches. Their leaderboard scores are increasingly used as evidence

中文介绍 该研究探讨了GSO、SWE-Perf和SWE-fficiency等代码库级别性能优化基准测试在评估编程智能体方面的可靠性。这些基准通过将补丁应用于实际代码库，并与未优化基线及官方参考补丁的运行时进行比较来评估智能体。研究指出，虽然排行榜分数持续上升，但有必要审视这些基准是否能可靠地衡量代码智能体的真实优化能力。

TiRex-2: Generalizing TiRex to Multivariate Data and Streaming

👍 0

07/01 08:00

We introduce TiRex-2, a recurrent xLSTM-based time series foundation model that generalizes the univariate TiRex to multivariate forecasting with both past and future covariates. Real-world forecasting is inherently sequential: observations arrive continuously, variables evolve jointly, and a subset

中文介绍 论文介绍了TiRex-2，这是一个基于循环xLSTM的时间序列基础模型，它将单变量TiRex模型推广到支持多元预测，并能处理过去和未来的协变量。TiRex-2旨在应对真实世界预测中数据持续到达、变量持续演变的序贯特性，从而实现更通用的时间序列预测能力，为复杂数据集提供更全面的分析。

Perceive-to-Reason: Decoupling Perception and Reasoning for Fine-Grained Visual Reasoning

👍 11

07/01 08:00

Fine-grained visual reasoning remains challenging for vision-language models, especially when small but critical visual cues are buried in high-resolution images. Existing approaches rely on repeated cropping or test-time visual search to introduce local evidence, but they typically do not explicitl

中文介绍 论文提出了“感知-推理解耦”方法，以解决视觉-语言模型在细粒度视觉推理方面的挑战，特别是在高分辨率图像中包含微小但关键视觉线索的情况下。现有方法通常依赖重复裁剪或测试时视觉搜索引入局部证据，但效率不高。该研究旨在通过分离感知与推理过程，提高模型捕捉细微视觉信息并进行有效推理的能力。

Autonomous Scientific Discovery via Iterative Meta-Reflection

👍 3

07/01 08:00

Autonomous scientific discovery systems offer the potential to accelerate research by automating the process of hypothesis generation and validation. However, current systems operate within constrained search spaces or require predefined research questions, limiting their capacity for true open-ende

中文介绍 这篇论文探讨了通过迭代元反思（Iterative Meta-Reflection）实现自主科学发现的系统。当前自主科学发现系统常受限于狭窄的搜索空间或预设的研究问题，限制了其能力。该研究旨在开发一种更通用的方法，通过自动化假设生成和验证过程来加速研究，使系统能够进行更开放和灵活的科学探索。

CausalMix: Data Mixture as Causal Inference for Language Model Training

👍 14

07/01 08:00

In Large Language Model (LLM) training, data mixing plays a pivotal role in determining model performance. Recent methods optimize mixture weights via proxy models, but they rely on the assumption of static data distributions. As a result, when the underlying data pool shifts, these methods require

中文介绍 CausalMix 提出将大型语言模型（LLM）训练中的数据混合问题视为因果推理。现有方法通过代理模型优化混合权重，但通常假设数据分布是静态的，导致在数据池变化时性能下降。CausalMix旨在通过因果推理视角，动态调整数据混合策略，以应对训练数据分布的变化，从而提高LLM的训练效率和泛化能力。

MemSyco-Bench: Benchmarking Sycophancy in Agent Memory

👍 18

07/01 08:00

Memory has emerged as a cornerstone of modern LLM-based agents, supporting their evolution from single-turn assistants to long-term collaborators. However, memory is not always beneficial: retrieved memories often induce a critical issue of sycophancy, causing agents to over-align with the user at t

中文介绍 论文介绍了MemSyco-Bench，一个用于评估大型语言模型（LLM）智能体记忆中「谄媚行为」（sycophancy）的基准。记忆对于LLM智能体至关重要，但它也可能导致问题，即检索到的记忆常诱导智能体过度迎合用户。该基准旨在识别和量化智能体因记忆而产生的谄媚倾向，以帮助开发更可靠、更中立的LLM智能体。

Logit-Contribution Scoring Identifies Non-Literal Retrieval Heads

👍 0

07/01 08:00

In long-context use, large language models frequently synthesize answers from the meaning of a relevant context span rather than literally copy-pasting them. Identifying which attention heads perform this synthesis matters for interpreting long-context model behavior. Yet existing detectors miss the

中文介绍 这篇论文提出「Logit贡献评分」（Logit-Contribution Scoring）方法，用于识别大型语言模型在长上下文使用中，那些执行非字面检索的注意力头。在长文本情境下，LLM常根据相关上下文的含义而非字面复制来合成答案。识别负责这种合成的注意力头，对于理解长上下文模型行为至关重要，有助于提升模型的可解释性。

Graph-Native Reinforcement Learning Enables Traceable Scientific Hypothesis Generation through Conceptual Recombination

👍 4

07/01 08:00

Accelerating materials discovery requires AI systems that can generate scientifically valid hypotheses through multi-step, domain-grounded reasoning. Standard large language models often produce fluent but weakly traceable responses to open-ended materials design problems, making it difficult to det

中文介绍 该研究提出一种图原生强化学习方法，通过概念重组实现可追溯的科学假设生成，以加速材料发现。传统的LLM在开放式材料设计问题上生成的回应通常流畅但可追溯性较弱。此方法旨在提供多步骤、领域基础的推理能力，确保生成的假设具有科学有效性和可解释性，克服了标准LLM在可追溯性方面的局限。

Valdi: Value Diffusion World Models

👍 7

07/01 08:00

World models can enable Model Predictive Control (MPC), but this requires dynamics prediction that is both fast enough for online use and expressive enough to represent uncertain futures. Diffusion models offer a natural mechanism for modeling uncertain dynamics, yet their iterative inference proced

MG-RWKV: Multi-Grained Context-Aware RWKV for Temporal Forgery Localization

👍 4

07/01 08:00

Driven by Artificial Intelligence-Generated Content (AIGC), the authenticity of audio-visual content is facing severe challenges. Temporal Forgery Localization (TFL) aims to precisely identify manipulated segments within untrimmed sequences. However, existing methods are limited by CNNs' local recep

MultiSynt/MT: Trillion-Token Multi-Parallel Pre-Training Data Translated Across 36 Languages

👍 0

07/01 08:00

Open web-scale pre-training corpora remain concentrated in English, limiting multilingual LLM development. We introduce MultiSynt/MT, an open synthetic parallel corpus with approximately 4.8 trillion target-language tokens across 36 European languages, produced by translating 100 billion high-qualit

TrajLoc: Trajectory-Attention Localization for Multi-Object Motion Control

👍 1

07/01 08:00

Controlling the motion of multiple objects in image-to-video (I2V) generation requires preserving object identities while enforcing adherence to distinct target trajectories. This becomes particularly challenging as the number of objects increases and their paths intersect or occlude one another. Ex

LeVLJEPA: End-to-End Vision-Language Pretraining Without Negatives

👍 0

07/01 08:00

Vision-language pretraining remains dominated by contrastive objectives, whereas vision-only self-supervised learning has largely adopted non-contrastive methods. At the same time, the role of vision-language encoders has shifted: they are increasingly deployed not as zero-shot classifiers but as th

What Survives Into Context: A Diagnostic for Budget-Constrained Multi-Hop RAG and When Submodular Evidence Packing Improves It

👍 0

07/01 08:00

Retrieval-augmented generation (RAG) under a fixed reader-context budget forces a selection problem: of the evidence retrieved, only a fraction can be shown to the reader. We argue that document recall -- the standard retrieval metric -- is the wrong quantity to optimize in this regime, and we make

MSQA: A Natively Sourced Multilingual and Multicultural SimpleQA Benchmark

👍 0

07/01 08:00

Multilingual fluency often invites a stronger assumption: a model that can speak a user's language must also understand the culture encoded by that language. We call this the Illusion of Cultural Alignment. To test this assumption directly, we introduce MSQA, a benchmark of 1,064 natively sourced qu

ABot-M0.5: Unified Mobility-and-Manipulation World Action Model

👍 11

07/01 08:00

Mobile manipulation is a key capability for general-purpose robots, yet remains challenging for current embodied learning methods. VLA policies are typically reactive and lack explicit world modeling, while existing World Action Models (WAMs) are still poorly aligned with the structure of mobile man

Domain Arithmetic: One-Shot VLA Adaptation under Environmental Shifts

👍 16

07/01 08:00

Vision-Language-Action (VLA) models often fail to perform the same learned tasks under environmental shifts, such as changes in camera pose and shifts to a different but similar robot (e.g., from Panda to UR5e). Adapting these models to the shifted environment (i.e., target domain) often requires tr

HARC: Coupling Harmfulness and Refusal Directions for Robust Safety Alignment

👍 2

07/01 08:00

Understanding how aligned LLMs internally represent safety is critical for diagnosing alignment vulnerabilities, as it explains why jailbreaks succeed and informs the design of robust alignment strategies. Prior work shows that aligned LLMs encode harmfulness and refusal as separable directions in t

Cross-Domain Generalization Failure in Lightweight Intrusion Detection Models for IIoT Networks

👍 5

07/01 08:00

Lightweight machine learning models are increasingly proposed for intrusion detection in Industrial Internet of Things (IIoT) networks due to their suitability for resource-constrained edge deployment. Most reported results evaluate these models only within their training network, leaving behavior o

NoPA: Non-Parametric Online 3D Scene Graph Generation

👍 2

07/01 08:00

Classic 3D scene graph generation approaches fail to work in real-time due to the heavy computational cost of environment mapping and the need to generate intermediate point-cloud representations. To alleviate this issue, a recent work eschews point clouds in favor of a lightweight Gaussian distribu

SPECSIA: Stylization Dataset for Novel-View Enhancement in Drawing-based 3D Animation

👍 0

07/01 08:00

Generating animation from a single 2D drawing is challenging because the output must preserve character appearance while remaining plausible and temporally coherent under motion. Existing drawing-based 3D animation pipelines often use sample-wise 2D refinement to align animated renderings with the i

ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE Serving

👍 19

07/01 08:00

In prefill-decode (PD) disaggregated LLM serving, each request is assigned to a decode worker after prefill. Existing decode routers balance only load; for mixture-of-experts (MoE) models this is incomplete: equally loaded workers can differ in latency, since each decode step loads the weights of ev

Multimodal Continuous Reasoning via Asymmetric Mutual Variational Learning

👍 16

07/01 08:00

Multimodal Large Language Models (MLLMs) are often constrained by a language-space bottleneck, forcing complex visual reasoning into discrete tokens which can lose perceptual nuance. A promising alternative is continuous latent reasoning, where the goal is to discover implicit reasoning pathways tha

Understanding Why Language Models Hallucinate: Testing Reasoning Against Priors

👍 0

07/01 08:00

Large language models often produce hallucinated answers that violate prompt-level constraints. A key diagnostic question is whether these failures reflect missing knowledge, or whether the model has the relevant information but follows the wrong inference path. We study this phenomenon as inference

VideoSearch-R1: Iterative Video Retrieval and Reasoning via Soft Query Refinement

👍 11

07/01 08:00

As video corpora continue to expand in both scale and task complexity, there is increasing demand for approaches that retrieve relevant videos from large-scale corpora (inter-video reasoning) and subsequently perform fine-grained, query-conditioned tasks (intra-video reasoning) within the retrieved

MedCAGD: Context-Aware Gated Decoder for Efficient Medical Image Segmentation

👍 0

07/01 08:00

Medical image segmentation relies on the ability of encoder-decoder architectures to translate rich feature representations into accurate pixel-level predictions under challenging conditions such as low contrast, structural ambiguity, and scale variability. While recent advances in large-scale pretr

Retrace

07/03 06:04

Debug AI agents by replaying and forking runs

中文介绍 Retrace是一款专注于AI智能体调试的工具。它允许开发者通过重放和分支运行来分析并解决智能体运行中出现的问题，从而提高开发效率。

scritty

07/03 06:03

Shared, searchable memory for every AI coding agent

中文介绍 scritty为AI编程智能体提供共享且可搜索的记忆功能。其目标是显著提高智能体之间的协作效率及信息的复用性，优化开发流程。

Sidedoor

07/03 06:02

Paste any job, find who in your network can refer you

中文介绍 Sidedoor是一款求职工具，用户只需粘贴任何职位链接，即可在其人脉网络中找到能够提供内推的人，简化了内推获取流程。

Flowly

07/03 06:02

A personal AI agent that runs on your desktop and iPhone

中文介绍 Flowly是一款个人AI智能体，支持在桌面和iPhone设备上运行，旨在为用户提供个性化的智能助理服务，提升日常效率。

Banger Mail

07/03 06:02

Shared mailboxes for teams and AI agents

中文介绍 Banger Mail专为团队和AI智能体设计，提供便捷的共享邮箱服务。它旨在简化信息协同与管理，从而有效提升整体工作效率。

PixFit

07/03 06:02

Turn 1 creative into every ad format, instantly

中文介绍 PixFit是一款效率工具，能将一个创意素材迅速转换为各种广告格式。它帮助用户即时生成多样化的广告内容，大幅提升制作效率。

Needle

07/03 06:02

The proactive GTM agent in Slack and Teams

中文介绍 Needle是一款主动式市场推广（GTM）智能体，可集成到Slack和Teams等协作平台，帮助团队进行市场拓展和客户互动。

Gaming Chat SDK by CometChat

07/03 06:02

Chat drops into Unreal like it was always there

中文介绍 CometChat推出游戏聊天SDK，能够无缝集成到虚幻引擎（Unreal Engine）中。它为游戏开发者提供便捷的聊天功能，无需复杂开发。

Context.dev

07/03 06:01

One API to scrape, enrich, and extract the internet

中文介绍 Context.dev提供一个统一的API接口，用于抓取、丰富和提取互联网数据。此服务旨在简化数据获取和处理的流程，提高效率。

html.contact

07/03 06:01

A full form backend you can test before paying

中文介绍 html.contact提供一个完整的表单后端服务，用户可以在付费前进行测试。此服务确保开发者能够验证功能，满足需求后再支付。

Building a Real AI Jarvis (GPT Realtime 2 + Cursor)

07/01 09:40

中文介绍 该视频由Riley Brown发布，展示了如何利用GPT Realtime 2和代码编辑器Cursor来构建一个真实的AI助手，旨在模仿电影中的智能管家「贾维斯」。内容可能涵盖AI实时交互、编程辅助以及个人化AI系统开发等技术细节。

Poweful Agents inside Slack

06/25 06:39

中文介绍 这段视频短片介绍了在Slack平台中部署功能强大的AI智能体。这些智能体能够协助用户完成各项任务，提高工作效率，实现自动化流程。

AI Agents Just Changed Forever: GLM 5.2, Codex Skills, Claude & Cursor

06/21 21:30

9 AI Agent Skills To Get Ahead of 99% of People

06/18 10:27

How to build high quality internal tools with Retool

06/18 05:21

SpaceX Just Bought Cursor for $60B. It’s About to Take OVER.

06/17 05:31

Why the Government Just Killed Claude Fable 5

06/13 11:50

This is Why They Banned Claude Fable 5

06/13 02:28

Hermes Agent NEW Super-App and DeepSeek v4 Catches Up To Opus 4.8?

06/07 04:10

ChatGPT And Codex Are Merging (This Changes Everything)

06/04 06:17

你们 AI 跑任务的间隙都干嘛？

07/02 21:35

5 回复 · 程序员 节点

AI 太贵了，有没有价廉物美的中转站推荐一下

07/02 20:33

10 回复 · 程序员 节点

Linux 官网 (kernel.org) 删除了所有的 release tar 包

07/02 20:06

9 回复 · 程序员 节点

真的没人觉得 codex 比 claude 笨很多吗？

07/02 19:00

34 回复 · 程序员 节点

怎么搞定纯 Python 代码解码 jpg 图片，要求无外部依赖

07/02 18:48

9 回复 · Python 节点

公司 vibe coding 的项目，团队已经无法掌控了

07/02 17:13

96 回复 · 程序员 节点

Apple 什么时候才会考虑把 Mac 上的刘海儿去掉

07/02 11:40

28 回复 · Apple 节点

oyunfor 购买土区 apple 礼品卡支付失败

07/02 09:53

13 回复 · Apple 节点

京东 MacBookpro 16 寸 64G+1T 纳米屏 现在 25985（卖后悔药的来了）

07/02 09:06

36 回复 · Apple 节点

美区 ID 开通 ChatGPT plus 到别人帐号了怎么办

07/02 00:47

6 回复 · Apple 节点

BlastRadar – paste a Git diff, get a production risk score in 10 seconds

07/03 05:19

3 points · 0 comments

Virginia bans sale of geolocation data

07/03 05:03

114 points · 15 comments

Lightning Memory-Mapped Database Manager (LMDB) 1.0

07/03 04:01

36 points · 16 comments

Wireless LAN SD

07/03 03:35

16 points · 15 comments

Great Salt Lake Tracker – Grow the Flow

07/03 03:33

19 points · 0 comments

EFF letter to FTC on X consent order (2 July 2026) [pdf]

07/03 03:27

70 points · 17 comments

A New Catalog of Stellar Rotation Periods for over a Million Stars

07/03 03:20

13 points · 2 comments

The Short Leash AI Coding Method for Beating Fable

07/03 03:11

32 points · 11 comments

Claude-real-video － any LLM can watch a video

07/03 03:10

34 points · 8 comments

JEP 539: Strict Field Initialization in the JVM moved to preview

07/03 02:56

40 points · 12 comments

Exapunks (2018)

07/03 02:41

177 points · 66 comments

Postgres transactions are a distributed systems superpower

07/03 02:38

61 points · 27 comments

Vulkan is now available on NetBSD

07/03 02:36

54 points · 12 comments

24-bit/192kHz music downloads and why they make no sense (2012)

07/03 00:24

71 points · 97 comments

Show HN: zkGolf – Competitive optimization of formally verified circuits

07/02 23:40

Zero-Knowledge Proofs (ZKPs) let an untrusted proved show that computation was executed correctly without revealing the inputs to the verifier. However to prove anything, the computation first has to be expressed as a circuit: a system of polynomial equations (constraints) over a finite field. Circu

Since Linux 6.9, LUKS suspend stopped wiping disk-encryption keys from memory

07/02 23:25

353 points · 170 comments

Launch HN: Manufact (YC S25) – MCP Cloud

07/02 23:11

Hi HN, we are Pietro and Luigi, cofounders of Manufact (https://manufact.com), a cloud for MCP apps and servers. We used to be called mcp-use, and still build open source SDKs for MCP under that name: https://github.com/mcp-use/mcp-use. We did a Show HN about that last

Spain Orders Blacklist of Palantir from Public and Private Companies

07/02 23:02

482 points · 154 comments

Podman v6.0.0

07/02 22:23

288 points · 107 comments

Immich 3.0

07/02 22:13

75 points · 27 comments

AI can't be listed as inventor on patent applications, Japan's top court rules

07/02 21:43

334 points · 181 comments

How to ask for help from people who don't know you

07/02 21:19

313 points · 48 comments

Hazel (YC W24) Is Hiring for Our Largest Government Contract

07/02 21:14

1 points · 0 comments

PeerTube is a free, decentralized and federated video platform

07/02 19:17

434 points · 191 comments

LibreCAD in the Browser

07/02 09:00

61 points · 5 comments

Client-side load balancing at a million requests per second

07/01 19:35

30 points · 5 comments

Ask HN: Since when does Craigslist's front page have emojis?

07/01 05:58

Today I noticed the inclusion of emojis in Craigslist's listings/categories: https://www.craigslist.org/area/sfbay.Now, Craigslist, as a legacy of the 1990s web, has for a long time stubbornly maintained its minimalist style, to the point where several "modern&quot

Superpowers 6

07/01 05:28

28 points · 8 comments

Reality has a surprising amount of detail

06/28 07:36

12 points · 2 comments

German button maker searched rivers of American Midwest for valuable shells

06/28 05:38

126 points · 40 comments

v2.1.198

07/02 04:45

What's changed Claude in Chrome is now generally available Added background agent notifications in claude agents — sessions that need input or finish now fire the Notification hook (agent_needs_input / agent_completed) Added /dataviz skill for chart and dashboard design guidance with a runnable colo

中文介绍 Anthropic的Claude v2.1.198版本更新。Claude现已在Chrome浏览器上普遍可用，并为Claude代理新增后台通知功能，在会话需输入或结束时触发。同时，增加了"/dataviz"技能，提供图表和仪表盘设计指导。

v2.1.197

07/01 01:56

What's changed Introducing Claude Sonnet 5: now the default model in Claude Code, with a native 1M-token context window and promotional pricing of $2/$10 per Mtok through August 31. Update to version 2.1.197 for access. https://www.anthropic.com/news/claude-sonnet-5

中文介绍 Anthropic的Claude Code发布v2.1.197更新，并引入了Claude Sonnet 5作为默认模型。该模型原生支持100万上下文窗口，并提供推广期价格：每百万tokens输入2美元，输出10美元，优惠期截至8月31日。

v2.1.196

06/30 07:27

What's changed Added support for organization default models — admins set it in the org console; it shows as "Org default" (or "Role default") in /model when you haven't picked one yourself Added readable default names for sessions at start, making them easier to identify and message Added clickable

中文介绍 Anthropic的Claude Code发布v2.1.196更新，新增了组织默认模型支持，管理员可在组织控制台设置。当用户未选择模型时，该默认模型将显示为「组织默认」或「角色默认」。此外，新版本还为会话启动时增加了可读的默认名称。

v2.1.195

06/27 05:29

What's changed Added CLAUDE_CODE_DISABLE_MOUSE_CLICKS to disable mouse click/drag/hover in fullscreen mode while keeping wheel scroll Fixed hook matchers with hyphenated identifiers (e.g. code-reviewer, mcp__brave-search) accidentally substring-matching — they now exact-match. Use mcp__brave-search_

v2.1.193

06/26 05:45

What's changed Added autoMode.classifyAllShell setting to route all Bash/PowerShell commands through the auto-mode classifier instead of only arbitrary-code-execution patterns Added auto-mode denial reasons to the transcript, the denial toast, and /permissions recent denials Added claude_code.assist

v2.1.191

06/25 05:58

What's changed Added /rewind support for resuming a conversation from before /clear was run Fixed scroll position jumping to the bottom while reading earlier output during a streaming response Fixed background agents resurrecting after being stopped — stopping an agent from the tasks panel is now pe

v2.1.190

06/24 23:53

What's changed Bug fixes and reliability improvements

v2.1.187

06/24 05:03

What's changed Added sandbox.credentials setting to block sandboxed commands from reading credential files and secret environment variables Added org-configured model restrictions to the model picker, --model, /model, and ANTHROPIC_MODEL, with a "restricted by your organization's settings" message w

v2.1.186

06/23 04:37

What's changed Added claude mcp login and claude mcp logout to authenticate MCP servers from the CLI without opening the interactive /mcp menu, with --no-browser stdin redirect support for completing over SSH Added status filtering (press f) to the /workflows agent detail view Added a "Skills" secti

v2.1.185

06/21 04:59

What's changed The stream-stall hint now reads "Waiting for API response · will retry in …" instead of "No response from API · Retrying in …", and triggers after 20s of silence instead of 10s