Great AI tools,
right in your hands.把好用的 AI，
直接放到你手里。

Singapore man charged with AI-generated crocodile attack on reservoir, up to 10 years in jail + S $10,000 fine新加坡一男子用 AI 生成鳄鱼侵袭水库照片被起诉，最高或面临 10 年监禁 +1 万新币罚款

Huang Jenxun and Li Zaiyong worked together, promising that Nvidia would continue its 33-year cooperation with Samsung Electronics.黄仁勋与李在镕同框，承诺英伟达将延续与三星电子长达 33 年的合作

AI office enters "contextual warfare", how does Baidu play cards?AI办公进入「上下文战争」，百度如何出牌？

DeepSeek open source operator tool package, join hands with Huawei Ascendant to tear CUDA bindings!DeepSeek 开源算子工具大礼包，联手华为昇腾，手撕 CUDA 绑定！

Panrai YO2 Ruilong AI Max + Pro 495 Mini Console on shelf: 192GB +2TB Price 49,999 yuan磐镭 YO2 锐龙 AI Max+ PRO 495 版迷你主机上架：192GB+2TB 售价 49999 元

MSI Preheat Venerable N16 Flip AI + 2026 2-in-1 flipbook: Nvidia RTX Spark chip, expected to be pre-sold on October 8微星预热尊爵 N16 Flip AI+ 2026 二合一翻转本：英伟达 RTX Spark 芯片，预计 10 月 8 日预售

China's large model enters the OpenAI enterprise payment and settlement system for the first time, and Kimi K3 accesses the Codex enterprise channel中国大模型首次进入 OpenAI 企业付费结算体系，Kimi K3 接入 Codex 企业通道

Meta was exposed to packaging AI data centers as experiments for tax incentives, saving nearly $4 billion last yearMeta 被曝将 AI 数据中心包装成实验性质以获取税收优惠，去年节省近 40 亿美元

Tutorial: Four steps to give your coding agent an answer-first style教程：四步给编程智能体装上「先说答案」技能

Four steps: install the marketplace and the skill, verify it with one prompt, make it load automatically in every session, and learn how to exit the style when you actually want a long explanation. About five minutes, and the shorter replies are obvious immediately.

四步跑通：装插件市场与技能、用一条提示词验证是否生效、设为每次会话自动加载、需要详细解释时临时退出。约五分钟，装完立刻能感觉到回复变短。

Tutorial: Give your agent long-term memory in four steps (self-hosted Hindsight)教程：四步给智能体接上长期记忆（自建 Hindsight）

Four steps end to end: start the memory service with Docker, create a memory bank and retain your first fact, plug the MCP endpoint into Claude Code, then verify that recall survives a new session. Everything runs locally, so your data never leaves the machine.

四步跑通：Docker 起记忆服务、建一个记忆库并写入第一条记忆、把 MCP 端点接进 Claude Code、最后验证跨会话是否真的记住了。全程本机运行，数据不出机器。

Tutorial: Set up your own Octop multi-agent AI assistant station in four steps教程：四步搭起自己的 Octop 多智能体 AI 助手站

From downloading the 15.7 MB source code package to start the service, to talking to the first expert agent in the browser, to accessing IM group chats and natural language scheduled tasks - a set of assistants that run entirely on your own machine.

从下载 15.7 MB 源码包启动服务，到在浏览器里和第一个专家智能体对话，再到接进 IM 群聊与自然语言定时任务——一套完全跑在自己机器上的助手。

Tutorial: Install Codex into Claude Code and let two programming agents review each other教程：把 Codex 装进 Claude Code，让两个编程智能体互相复核

Four steps: add the official plug-in market, complete the self-inspection, let Codex independently review a change, and then transfer the entire package of stuck tasks to the backend and manage the results. The whole journey takes less than ten minutes.

四步跑通：添加官方插件市场、完成自检、让 Codex 独立审查一次改动、再把卡住的任务整包转交后台并管理结果。全程不到十分钟。

OpenRig v0.6.2: Turn Claude Code and Codex into one agent squad with a single YAML fileOpenRig v0.6.2：用一份 YAML 把 Claude Code 和 Codex 编成一支智能体小队

A GitHub Trending project that shipped v0.6.2 today, around 2,700 stars. The idea is plain: a harness wraps one model, a rig wraps all of your harnesses. Define the team in YAML, boot it with one command, and run Claude Code plus Codex side by side inside the same rig. Official npm offline package, about 5.95 MB.

当天发布 v0.6.2 的 GitHub 热榜项目，约 2700 Star。它的思路很直白：harness 包的是一个模型，rig 包的是你所有的 harness——把队伍写进 YAML，一条命令启动，Claude Code 与 Codex 同处一个 rig 里按图编队。官方 npm 离线包约 5.95MB。

OpenAI official Codex plug-in is installed into Claude Code: commanding two programming agents at the same time in one terminalOpenAI 官方 Codex 插件装进 Claude Code：一个终端里同时指挥两个编程智能体

GitHub hot list project this week, about 34,000 stars. It directly connects Codex to Claude Code: you can let Claude do the work, let Codex review it independently, or you can dump a stuck task in its entirety to the Codex background for running. The plug-in itself is only about 118 KB and can be installed with three commands from Claude Code’s plug-in market.

本周 GitHub 热榜项目，约 3.4 万 Star。它把 Codex 直接接进 Claude Code：你可以让 Claude 干活、让 Codex 独立复核，也能把一个卡住的任务整包甩给 Codex 后台跑。插件本体仅约 118 KB，靠 Claude Code 的插件市场三条命令装好。

Claude Cowork official plug-in library: 22 domain plug-ins to turn AI into an expert in your positionClaude Cowork 官方插件库：22 个领域插件，把 AI 变成你岗位的专家

Anthropic is an official open source knowledge work plug-in library, about 25.7k stars. A compressed package contains 22 out-of-the-box domain plug-ins, from finance, legal affairs, human resources to design, scientific research, and data analysis. Install them and let AI work according to your job process.

Anthropic 官方开源的知识工作插件库，约 25.7k Star。一个压缩包内含 22 个开箱即用的领域插件，从财务、法务、人力到设计、科研、数据分析，装上就让 AI 按你的岗位流程干活。

Jianying Headless AI editing plug-in: allows the agent to directly generate editable drafts剪映 AI 剪辑插件 Jianying Headless：让智能体直接生成可编辑草稿

An editing automation project with 2,700+ stars launched on GitHub in two weeks: AI writes an editing plan, generates multi-track editable drafts with one click, and can also call the native editing engine to directly export MP4, making it a powerful tool for making short videos in batches.

GitHub 上线两周 2700+ 星的剪映自动化项目：AI 写好剪辑计划，一键生成多轨可编辑草稿，还能调用本机剪映引擎直接导出 MP4，批量做短视频的利器。

Hindsight v0.10.2: Long-term memory that lets your agent actually learnHindsight v0.10.2：给智能体一套会「学」的长期记忆

One of the fastest-rising memory systems today, around 43,500 stars. It stores facts, entity relationships and temporal reasoning in PostgreSQL plus pgvector, so an agent remembers people, events and time, and forms opinions under configurable disposition traits. One docker run starts the service, and an MCP endpoint comes built in.

当天涨星最猛的记忆系统之一，约 4.35 万 Star。它把事实、实体关系和时序推理存进 PostgreSQL + pgvector：智能体能记住人、事、时间，还能按可配置的性格倾向形成判断。一条 docker run 起服务，自带 MCP 端点。

Tencent Cloud open source Octop 1.0.2b4: a multi-user, multi-agent self-hosted AI assistant installed in one process腾讯云开源 Octop 1.0.2b4：一个进程装下多用户、多智能体的自托管 AI 助手

This week's GitHub hot list project, about 5.5k stars. A single process startup provides a web console, command line, eight major IM channels (Feishu, DingTalk, QQ, WeChat, Telegram, Discord, Enterprise WeChat, etc.) and natural language scheduled tasks. All data falls in the ~/.octop/ directory of the local machine.

本周 GitHub 热榜项目，约 5.5k Star。单进程启动就同时提供 Web 控制台、命令行、八大 IM 渠道（飞书、钉钉、QQ、微信、Telegram、Discord、企业微信等）与自然语言定时任务，所有数据落在本机 ~/.octop/ 目录里。

OpenSpec v1.13.2: Write specifications first and then code, and put an "anti-deviation" constraint on AI programmingOpenSpec v1.13.2：先写规格再写代码，给 AI 编程上个「防跑偏」约束

A specification-driven development tool with about 70.5k stars, a GitHub weekly list project in September 2026. It allows AI to first write requirements into verifiable specification documents, and then write code according to the specifications, fundamentally solving the old problem of "changing to the fifth round is no longer the original requirement."

约 70.5k Star 的规格驱动开发工具，2026 年 9 月 GitHub 周榜项目。它让 AI 先把需求写成可验证的规格文件，再照着规格写代码，从根上解决「改到第五轮已经不是原来的需求」这个老毛病。

PR Lens open source tool: automatically generate architecture and data flow walkthrough diagrams during code reviewPR Lens 开源工具：在代码评审里自动生成架构与数据流走查图

Can’t understand the code review, and the PR backlog is unattended? PR Lens generates animated architecture diagrams and data flow walkthrough diagrams directly in Pull Requests, allowing reviewers to understand the change path at a glance. Approximately 1,300 stars on GitHub.

代码评审看不懂、PR 积压没人盯？PR Lens 直接在 Pull Request 里生成动画版架构图与数据流走查图，让评审者一眼看懂改动路径，GitHub 约 1300 星。

i-have-adhd skill pack: make your coding agent answer first, explain lateri-have-adhd 技能包：让编程智能体把答案说在最前面

A small skill with around 52,000 stars that does exactly one thing: it makes the agent lead with the command, the file path and numbered steps instead of burying the answer in the third paragraph. The whole pack is about 226 KB and works with 20+ clients including Claude Code, Codex, Cursor, Gemini CLI and opencode, with official Simplified Chinese install docs inside.

约 5.2 万 Star 的小技能，只干一件事：让智能体先给命令、路径和编号步骤，再谈解释。砍掉「好问题」「希望有帮助」这类铺垫。整包约 226KB，兼容 Claude Code、Codex、Cursor、Gemini CLI、opencode 等 20 多个客户端，自带官方简体中文安装文档。

Anthropic's official open source financial intelligence skill library: ready-made methods for investment banking, stock research, private equity, and wealth managementAnthropic 官方开源金融智能体技能库：投行、股票研究、私募、财富管理的现成打法

The official Anthropic repository with approximately 38,000 Stars. It is not a model that makes decisions for you, but breaks down the most time-consuming repetitive work of financial analysts - building models, writing memos, and organizing research notes - into a skill package that can be directly executed by an intelligent agent, and the final judgment is still signed by a human.

约 3.8 万 Star 的 Anthropic 官方仓库。它不是替你拍板的模型，而是把金融分析师最耗时的重复劳动——搭模型、写备忘录、整理研究笔记——拆成智能体能直接执行的技能包，最终判断仍然由人签字。

ECC Intelligent Agent Skill Pack: 286 production-level skills, arm your AI assistant to your teeth at onceECC 智能体技能包：286 个生产级技能，一次把 AI 助手武装到牙齿

From the GitHub weekly top project affaan-m/ECC (approximately 268,000 Stars). This site has extracted all the skills from the 50MB+ complete source code and packaged them in a slim form: 286 skills, 1.34 MB, just install it and use it.

来自 GitHub 周榜头部项目 affaan-m/ECC（约 26.8 万 Star）。本站已从 50MB+ 完整源码里抽出全部技能本体并瘦身打包：286 个技能、1.34 MB，装上就能用。

Sepia goes AI-flavored writing skill pack: 2700+ stars in three weeks, compatible with 77+ agentsSepia 去 AI 味写作技能包：三周 2700+ 星，兼容 77+ 智能体

Sepia is a writing skill that lets AI text bid farewell to the "machine-like" feel: three-level corrections from narrative structure, literary flow to word choice and sentence making. It is compatible with 77+ clients and is not bound to a single ecosystem. The package comes with simplified Chinese instructions.

让 AI 文本告别「机器味」的写作技能 Sepia：从叙事架构、文气流动到遣词造句三层修正，兼容 77+ 客户端，不绑定单一生态，包内附简体中文说明。