每日 GitHub 发现

Claude Animated Shorts 是一个使用 three.js 和 Remotion 在本地渲染 2D 人物在 3D 世界中的动画短片的技能。它可以创建 55-70 秒的垂直动画故事短片，包括人物、背景、光效和音效。这个技能已经被用于短片《The Bridge That Fell for a Song》。

Claude Animated Shorts is a skill that uses three.js and Remotion to locally render animated shorts with 2D characters in a 3D world. It can create 55-70 second vertical animated story shorts with characters, background, lighting effects, and sound effects. This skill has been used in the short film "The Bridge That Fell for a Song".

FreeToken是一个边缘计算的混合专家服务引擎，用于在个人和消费者硬件上运行前沿规模的开放权重模型。它支持多种硬件设备，包括GPU、CPU和主机内存，并提供高效的混合专家服务。FreeToken适合用于运行前沿模型的个人和消费者硬件，例如游戏机和工作站GPU。

FreeToken is an edge-native Mixture-of-Experts (MoE) serving engine designed for running frontier-scale open-weight models on personal and consumer hardware. It supports various hardware devices, including GPUs, CPUs, and host memory, and provides efficient MoE serving. FreeToken is suitable for running frontier models on personal and consumer hardware, such as gaming PCs and workstation GPUs.

Pika H3 Flex 是一个用于生成视频和音频的 H3 模型，旨在为专业创作者提供灵活性和速度，精确控制生成的视频。它支持多种模式，包括文本到视频和音频、第一帧条件、最后帧条件和多模态参考。该模型可以生成480p/768p视频，支持11种分辨率和16种比例。它还支持6步生成过程，支持多种参考输入，包括图像、视频和音频。

Pika H3 Flex is a video and audio generation model designed for professional creators. It supports six-step generation, omni-reference, and native 480p/768p video with mixed stereo audio. The model can generate video with synchronized audio in six sampling steps and supports up to 12 reference inputs, including images, videos, and audio clips. It also supports various input modalities, including text, first/last images, and image/video/audio references.

Raven是一个轻量级、多线程的GUI库，用于编辑器和工具，支持多种语言，包括C、Rust、C++、C#、Java、Python、Zig、Lua和Go。它提供了一个易于使用的API，允许开发者创建复杂的GUI应用程序。Raven支持多种主题和布局，包括Slate样式布局，支持动画和过渡效果。它还支持多种输入设备，包括触摸屏、笔和触摸板。Raven是开源的，使用Apache License 2.0许可证。

Raven is a lightweight, multithreaded GUI library for editors and tools, supporting multiple languages including C, Rust, C++, C#, Java, Python, Zig, Lua, and Go. It provides an easy-to-use API for developers to create complex GUI applications. Raven supports multiple themes and layouts, including Slate-style layouts, with animations and transition effects. It also supports multiple input devices, including touchscreens, pens, and trackpads. Raven is open-source, licensed under the Apache License 2.0.

Stashcard 是一个用于评估 Steam 账户和库存价值的终端工具。它支持多个市场、多个游戏和多种输出格式。Stashcard 可以帮助用户了解他们的 Steam 账户价值、库存价值和市场价值。它还可以帮助用户比较不同时间点的价值变化。

Stashcard is a terminal tool for valuing Steam accounts and inventory. It supports multiple markets, multiple games, and various output formats. Stashcard can help users understand their Steam account value, inventory value, and market value. It also helps users compare value changes over time.

本技能包收集了83个短剧技能项目，涵盖小说改编、剧本、分镜、视频提示词等方面。这些技能适用于各种智能体，包括千问、豆包、龙虾OpenClaw、WorkBuddy、Claude Code、Codex等。技能包中包括了各种类型的技能，包括全流程技能、编剧技能、分镜技能、上游技能、编剧基本功技能、配套技能和相关开源平台。

This skill package collects 83 short drama skills, covering novel adaptation, screenwriting, storyboarding, video prompts, and more. These skills are compatible with various agents, including QWQ, Douba, Longxia OpenClaw, WorkBuddy, Claude Code, and Codex. The package includes various types of skills, such as pipeline skills, writing skills, visual skills, source skills, craft skills, companion skills, and platform skills.

本项目收集了中国各个年级的数学教材，包括小学、初中、高中和大学阶段。这些教材包括一年级到六年级的小学数学教材、初一到初三的初中数学教材、以及高等数学、线性代数、离散数学和概率论等大学数学教材。这些教材都是PDF格式，方便下载和使用。项目的目的是促进义务教育的普及和消除地区间的教育贫困。

This project collects Chinese math textbooks for various grade levels, including elementary school, junior high school, high school, and university. The textbooks include elementary school math textbooks from first grade to sixth grade, junior high school math textbooks from first year to third year, and university math textbooks such as higher mathematics, linear algebra, discrete mathematics, and probability theory. The textbooks are in PDF format, making it easy to download and use. The project aims to promote compulsory education and eliminate regional educational poverty.

open-yoki 是一款 AI 直播主播系统，支持语音、面部、聊天、记忆和发布功能。它可以在一台服务器上运行，适合于币安广场或其他支持 RTMP 的直播平台。系统还支持 Telegram 粉丝群和运营群，外部 AI 世界的门，以及推流地址管理等功能。

open-yoki is an AI live-stream host system that supports voice, face, chat, memory, and post functions. It can run on a single server and is suitable for Binance Square or other RTMP-supported live streaming platforms. The system also supports Telegram fan groups and operation groups, external AI world doors, and stream address management.

sleepy-zone/ai-ref-library-cn 是一个基于 JavaScript 的 AI 作图素材库，提供了中文版的脸、姿势和服装参考素材。它支持分类和标签的中文翻译、双语搜索、随机组合生成提示词等功能。该库适合用于 AI 作图、图像生成等领域的开发者和设计师。

sleepy-zone/ai-ref-library-cn is a JavaScript-based AI art material library that provides a Chinese version of face, pose, and clothing reference materials. It supports Chinese translation of categories and tags, bilingual search, and random combination generation of prompt words. The library is suitable for developers and designers in the fields of AI art and image generation.

Kanzo是一个自托管的kanban板，适用于人类和代理。它提供了一个轻量级的MCP服务器，允许代理和人类在同一个任务上工作。Kanzo支持共享板和同步板功能，适用于多台设备的协作。它使用SQLite数据库，内存占用小，启动速度快。Kanzo使用AGPL-3.0许可证，鼓励用户修改和共享。

Kanzo is a self-hosted, sharable kanban board for humans and agents. It provides a lightweight MCP server, allowing agents and humans to work on the same tasks. Kanzo supports sharing boards and syncing boards, ideal for collaboration across multiple devices. It uses SQLite databases, has low memory usage, and starts up quickly. Kanzo is licensed under AGPL-3.0, encouraging users to modify and share.

stack-diagram是一个技能，可以帮助你描述你的技术栈并生成一个带有官方Logo的架构图。它支持多种输出格式，包括HTML文件、PNG图像、React组件和Claude艺术品。该技能可以与Claude Code、Codex、Cursor和其他支持技能的代理一起使用。它还支持自定义Logo和颜色主题。

stack-diagram is a skill that helps you describe your tech stack and generate an architecture diagram with official logos. It supports multiple output formats, including HTML files, PNG images, React components, and Claude artifacts. The skill works with Claude Code, Codex, Cursor, and other agents that support skills. It also supports custom logos and color themes.

Elephant-Goldfish Model是一种开源技能，用于设计和审查软件与AI代理的技能。它通过Elephant和Goldfish两个角色来实现，Elephant负责探索问题并编写规范，Goldfish负责检查规范并找出潜在的假设和盲点。用户可以使用技能来设计和审查软件，确保软件的质量和可靠性。

The Elephant-Goldfish Model is an open-source skill for designing and reviewing software with AI agents. It uses two roles, Elephant and Goldfish, to achieve this. Elephant explores the problem and drafts the specification, while Goldfish reviews the specification and identifies potential assumptions and blind spots. Users can use the skill to design and review software, ensuring the quality and reliability of the software.

NOIR 是一个用于审计 Android 设备的工具，通过 adb 扫描设备上的应用，权限和网络端点，给出分类、风险评分和可执行的处置建议。它支持多种平台，包括 macOS、Windows 和 Linux。NOIR 可以帮助用户识别潜在的安全风险，并提供可执行的处置建议。

NOIR is a tool for auditing Android devices, scanning the device's apps, permissions, and network endpoints, and providing classification, risk scoring, and executable remediation suggestions. It supports multiple platforms, including macOS, Windows, and Linux. NOIR can help users identify potential security risks and provide executable remediation suggestions.

AnyModelBot 是一个本地优先的 AI 智能体应用，支持 Windows、macOS 和 Linux。它允许用户创建多个机器人，每个机器人都有自己的模型、工作区和记忆。用户可以通过 Modelflare API 访问 70 多个模型，包括 GPT、Claude、Gemini 等。AnyModelBot 支持多种聊天渠道，包括微信、飞书、Telegram、Slack、Discord 等。它还支持文件操作、命令执行和电脑操作等功能。AnyModelBot 的数据存放在用户的设备上，遵循本地优先的原则。

AnyModelBot is a local-first AI agent application that supports Windows, macOS, and Linux. It allows users to create multiple agents, each with its own model, workspace, and memory. Users can access over 70 models through the Modelflare API, including GPT, Claude, Gemini, etc. AnyModelBot supports multiple chat channels, including WeChat, Feishu, Telegram, Slack, Discord, etc. It also supports file operations, command execution, and computer operations. AnyModelBot stores data on the user's device, following a local-first principle.

ego (lite) 是一个专为AI agent设计的浏览器，允许它们在自己的空间中并行运行浏览器任务。它提供了一个强大的页面快照功能，支持多任务处理和可重用的技能。ego (lite) 与任何AI agent兼容，包括Claude Code和Codex。它是免费的，数据存储在本地，且不需要额外的设置。

ego (lite) is a browser designed for AI agents to run browser automation tasks in parallel with the user. It provides a powerful page snapshot feature, supports multi-tasking and reusable skills. ego (lite) is compatible with any AI agent, including Claude Code and Codex. It is free, stores data locally, and requires no additional setup.

Claude Code Templates 是一个 CLI 工具，用于配置和监控 Claude Code。它提供了一个开源的组件目录，包含约 1900 个组件（技能、代理、命令、MCP 集成、设置、钩子、模块、循环和项目模板），可以使用一个命令安装。该工具还包括三个本地工具：Claude Code 分析、会话浏览器和健康检查。

Claude Code Templates is a CLI tool for configuring and monitoring Claude Code. It provides an open catalog of about 1,900 components (skills, agents, commands, MCP integrations, settings, hooks, mods, loops, and project templates) that can be installed with one command. The tool also includes three local tools: Claude Code Analytics, Conversation Browser, and Health Check.

HairDeck是一个生成可用于动画的美化头发卡片的系统，能够从单张图像中生成头发卡片和驱动它们的骨骼链。它使用条件集扩散生成头发卡片，使用链扩散生成骨骼链。HairDeck适用于头发设计师、动画制作人和游戏开发者。它使用Python编写，依赖于PyTorch和CUDA。

HairDeck is a system for generating animation-ready stylized hair cards from a single image. It generates hair cards and the spring-bone chains that animate them using conditional set diffusion and chain diffusion. HairDeck is suitable for hair designers, animators, and game developers. It is written in Python and relies on PyTorch and CUDA.

Tellyscope是一款Android TV管理应用，允许用户在手机上安装应用、浏览和传输文件。它使用ADB协议与Android TV进行通信，支持多种语言，包括简体中文。Tellyscope还提供了多种功能，包括应用安装、文件传输、远程控制和系统设置等。它支持多种设备，包括Android TV、Google TV、NVIDIA Shield和Fire TV。

Tellyscope is an Android TV management app that allows users to install apps, browse and transfer files on their TV from their phone. It uses the ADB protocol to communicate with the Android TV, supports multiple languages, including English. Tellyscope also provides various features, including app installation, file transfer, remote control and system settings. It supports multiple devices, including Android TV, Google TV, NVIDIA Shield and Fire TV.

herdr 是一个桌面应用，用于终端和编码代理。它允许用户在空间中管理工作，查看需要他们的代理，并在关闭窗口时保持终端运行。herdr 支持多个机器的远程工作，用户可以在 Settings → Machines 或侧边栏的 Add machine 动作中添加 SSH 机器。herdr 还支持自定义布局和模块，用户可以通过命令面板找到动作或在 Settings 中更改字体、颜色、声音和其他偏好。

herdr is a desktop app for terminals and coding agents. It allows users to manage work in spaces, see which agents need them, and leave terminals running when closing the window. herdr supports remote work on multiple machines, users can add SSH machines in Settings → Machines or the sidebar's Add machine action. herdr also supports custom layouts and modules, users can find actions by name in the command palette or change fonts, colors, sounds and other preferences in Settings. herdr uses herdr-core for running and coordinating programs, and the app to change what users see. herdr has a built-in skill for driving the window, and users can use herdr-core for running and coordinating programs. herdr has a command reference for arguments, and users can use herdr-app to inspect the open window or list commands without one. herdr has a built-in skill for driving the window, and users can use herdr-core for running and coordinating programs. herdr has a command reference for arguments, and users can use herdr-app to inspect the open window or list commands without one.

技能集为人工智能编码代理提供设计和运动工作的技能。它包含了与React构建的静态UI、特性卡、鼠标demo和bento网格等功能。该技能集还提供了对组成和运动的指导、示例和工具，用于检查结果。

The skills set provides design and motion work for AI coding agents. It includes features such as static UI, feature cards, cursor demos, and bento grids built with React. The skills set also provides guidance on composition and motion, examples, and tools to check results.