{ "ok": true, "projects": [ { "id": "github-com-affaan-m-ecc", "titleZh": "Claude Code配置与Agen", "name": "ECC", "repo": "https://github.com/affaan-m/ECC", "homepage": "https://ecc.tools", "summary": "Anthropic黑客松冠军作者沉淀的Agent工程配置套件，整合Skills、Agents、Hooks、命令、规则、记忆、安全与研究优先开发流程。", "details": "ECC原名Everything Claude Code，现已发展为面向Claude Code、Codex、OpenCode、Cursor等编码Agent的工程性能优化系统，覆盖规划、架构、测试、代码审查、安全、构建修复、多语言规范和可复用工作流。安装后会影响Agent行为并可能执行Hooks，应先审查配置，在项目范围内最小化启用，并避免将密钥写入规则或日志。", "why": "视频标题直接指向“Claude Code配置合集、黑客松冠军项目”；官方仓库旧路径affaan-m/everything-claude-code当前重定向至affaan-m/ECC，仓库定位与描述一致。", "tags": [ "抖音推荐", "Claude Code", "Codex", "Agent Skills", "Hooks", "工程流程", "代码审查", "安全", "ai-agents", "anthropic", "claude", "claude-code" ], "language": "JavaScript / Rust / Python", "stars": 259363, "forks": 38799, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音标题与视频ID、ECC官方仓库、frontend-design Skill、design-quality规则及GitHub官方API元数据已交叉核验", "metadataVerifiedAt": "2026-09-16T02:17:47.460Z", "source": "抖音【小满科研】《这个Skill专治AI审美失控》", "video": "https://www.douyin.com/video/7678601559491530019", "icon": "code", "color": "#f59e0b", "createdAt": "2026-07-22T07:29:10.836Z", "updatedAt": "2026-09-16T02:15:20Z" }, { "id": "github-com-justvugg-colibri", "titleZh": "Colibri：低显存运行前沿 Mo", "name": "colibri", "repo": "https://github.com/JustVugg/colibri", "homepage": "https://justvugg.github.io/colibri", "summary": "纯 C、零依赖的 MoE 推理引擎，把专家权重从磁盘按需流式读取，让低显存设备也能运行 35B 等前沿稀疏模型。", "details": "Colibri 使用 VRAM、系统内存与 NVMe 三级专家驻留，不要求一次把全部模型权重装进显存；支持 Qwen3.6-35B-A3B 等 MoE 模型。GPU 主要用于提速，实际速度受磁盘与缓存影响，8GB 显存并不等于只需 8GB 总内存。", "why": "视频强调“8G显存运行35B”的核心逻辑；官方仓库明确通过磁盘流式专家权重与分层缓存降低显存门槛，项目定位吻合。", "tags": [ "抖音推荐", "本地大模型", "MoE", "低显存", "SSD Offload", "Qwen3.6", "纯 C", "零依赖", "AI 推理", "C", "Apache-2.0" ], "language": "C / Python / Cuda", "stars": 33932, "forks": 3550, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音标题与视频ID、Colibri官方README、GitHub官方API元数据已交叉核验", "metadataVerifiedAt": "2026-09-16T02:17:47.471Z", "source": "抖音【阿甘探AI】《本地部署大模型的逻辑被颠覆，8G显存、35B大模型》", "video": "https://www.douyin.com/video/7678535761758539043", "icon": "auto", "color": "#06b6d4", "createdAt": "2026-08-27T15:50:11.899472Z", "updatedAt": "2026-09-16T02:17:46Z" }, { "id": "github-com-deepseek-ai-deepseek-harness", "titleZh": "DeepSeek Harness 开", "name": "deepseek-harness", "repo": "https://github.com/deepseek-ai/deepseek-harness", "homepage": "https://deepseek.com/harness", "summary": "DeepSeek 官方开源的插件化 AI Agent Harness，一切能力均可通过插件组合。", "details": "DeepSeek Harness（dsh）是 DeepSeek AI 开源的智能体运行框架，基于 Cordis 构建“Everything is a Plugin”架构，提供 Web UI、插件系统及面向智能体开发的文档。目前处于开发者预览阶段，版本可能出现不兼容变更。", "why": "适合研究和搭建可组合、可扩展的 AI Agent 运行环境，也可直接通过 npm 启动 Web UI 体验。", "tags": [ "DeepSeek", "AI Agent", "Agent Harness", "插件化", "Cordis", "Web UI", "开源", "ai-agents", "cordis", "dsh", "dsh-plugin", "TypeScript" ], "language": "TypeScript / CSS / Python", "stars": 225440, "forks": 26840, "license": "MIT", "install": "", "compatibility": "", "evidence": "已核验 DeepSeek AI 官方 GitHub 仓库、README、许可证、主页及 GitHub 官方 API 元数据。", "metadataVerifiedAt": "2026-09-16T02:17:47.514Z", "source": "用户推荐", "video": "", "icon": "auto", "color": "", "createdAt": "2026-08-25T15:41:38.614623Z", "updatedAt": "2026-09-16T02:16:44Z" }, { "id": "github-com-freestylefly-awesome-gpt-image-2", "titleZh": "GPT Image 2 工业级提示词", "name": "awesome-gpt-image-2", "repo": "https://github.com/freestylefly/awesome-gpt-image-2", "homepage": "https://github.com/freestylefly/awesome-gpt-image-2", "summary": "Prompt as Code 的 GPT Image 2 工业级提示词引擎与模板库，包含470+案例、20+模板和可复用Skills。", "details": "开源的 GPT Image 2 提示词工程项目，通过案例逆向整理提示词结构、模板与工作流，覆盖图片生成、编辑和自动化使用场景。", "why": "适合直接查找高质量 GPT Image 2 提示词，也可把模板和Skills接入现有AI工作流。", "tags": [ "GPT Image 2", "AI绘图", "提示词工程", "Prompt as Code", "Skills", "图像生成", "开源", "agents", "ai-image-generation", "chatgpt", "dsh-plugin", "gpt-image-2" ], "language": "JavaScript / CSS / PLpgSQL", "stars": 32116, "forks": 3089, "license": "MIT", "install": "", "compatibility": "", "evidence": "已核验抖音短链对应视频ID 7667928603090981221，并用GitHub官方API匹配同主题高热仓库；仓库描述为GPT-Image2工业级提示词引擎与模板库。", "metadataVerifiedAt": "2026-09-16T02:17:47.681Z", "source": "抖音分享", "video": "https://v.douyin.com/fMCInJ3Vrkk/", "icon": "auto", "color": "", "createdAt": "2026-08-21T08:44:55.139044Z", "updatedAt": "2026-09-16T02:13:34Z" }, { "id": "github-com-laolaoshiren-claude-code-skills-zh", "titleZh": "Claude Code Skills", "name": "claude-code-skills-zh", "repo": "https://github.com/laolaoshiren/claude-code-skills-zh", "homepage": "https://claude-skills.bt199.com/", "summary": "面向中文开发者的 Claude Code Skills、Agents 与 Plugins 精选和原创技能库，按场景分类并提供可复制安装方式。", "details": "收录406+项Claude Code Skills资源，兼顾第三方精选与可直接安装的中文原创技能，覆盖代码审查、README、API测试、重构建议、性能分析、安全与CI/CD等开发场景。", "why": "中文说明完整、分类清晰且持续更新，适合作为Claude Code技能发现入口，也可直接获取zh-code-reviewer、zh-readme、api-tester、refactor-advisor和perf-profiler等原创技能。", "tags": [ "Claude Code", "Agent Skills", "中文资源", "开发工具", "代码审查", "API测试", "性能分析", "Prompt Engineering", "MIT", "ai-agent", "awesome-list", "claude-code" ], "language": "Python / JavaScript", "stars": 848, "forks": 92, "license": "MIT", "install": "克隆仓库后，将skills目录中的目标技能复制到~/.claude/skills/；安装前应审查SKILL.md及附带脚本。", "compatibility": "主要面向Claude Code；部分通用SKILL.md可按目标Agent格式适配，具体以各技能说明为准。", "evidence": "已核验GitHub仓库、README、官网、许可证、技能数量、安装方式及最近维护状态。", "metadataVerifiedAt": "2026-09-16T02:17:47.677Z", "source": "用户推荐", "video": "", "icon": "auto", "color": "", "createdAt": "2026-08-09T01:38:41.835576Z", "updatedAt": "2026-09-15T16:32:41Z" }, { "id": "github-com-alikon-art-determinflow", "titleZh": "可恢复的AI工作流与小说生产线", "name": "DeterminFlow", "repo": "https://github.com/alikon-art/DeterminFlow", "homepage": "https://bishuxiezuo.cn/", "summary": "面向生产的AI工作流运行时，把Agent、脚本、API、数据库与人工审批编排为可校验、可重试、可恢复、可审计的流程。", "details": "DeterminFlow以节点级局部上下文、独立模型会话、结构化校验、定向返工、检查点与Token账本运行复杂AI流程。官方bishu-novel插件公开7套Workflow、84个编排节点、33组Agent/Prompt与15个脚本模块，覆盖建书、角色、故事规划、卷纲/近纲、正文、章节后验和润色。README披露的一次真实正文生产任务含11个独立模型会话、消耗176,584 Token；相较作者估算的单智能体长链场景可节省约70%–89%，该比例属于项目方基于真实账本与对照模型作出的估算，并非独立基准。", "why": "它将多Agent协作从聊天式演示推进到版本化、可恢复、可审计的生产工作流，并提供来自笔枢写作真实小说链路的开源案例，适合研究AI内容工业化和通用Workflow Runtime。", "tags": [ "AI工作流", "多智能体", "AI小说", "节点级上下文", "断点恢复", "FastAPI", "React", "Python", "AGPL-3.0", "agent-orchestration", "ai-agents", "ai-workflow" ], "language": "Python / TypeScript / CSS", "stars": 734, "forks": 109, "license": "AGPL-3.0", "install": "", "compatibility": "", "evidence": "已核验抖音原图文、GitHub官方仓库/API与README；11个模型会话、176,584 Token、7套Workflow、84节点及70%–89%节省口径均按官方README呈现，并明确节省比例为项目方估算。", "metadataVerifiedAt": "2026-09-16T02:17:47.737Z", "source": "抖音【AI.opc前沿小邹】图文作品", "video": "https://www.douyin.com/note/7669451689249948916", "icon": "auto", "color": "", "createdAt": "2026-08-03T10:15:50.502378Z", "updatedAt": "2026-09-15T14:07:46Z" }, { "id": "github-com-mrpulor-gh-nuphus-mcp", "titleZh": "Nuphus MCP：跨平台桌面与浏", "name": "nuphus-mcp", "repo": "https://github.com/mrpulor-gh/nuphus-mcp", "homepage": "https://github.com/mrpulor-gh/nuphus-mcp", "summary": "用 Rust 构建的桌面与浏览器自动化 MCP 服务器，为 AI Agent 提供 36 个工具，可控制屏幕、窗口、鼠标键盘与 Chrome。", "details": "Nuphus MCP 通过 Model Context Protocol（stdio）向任意 AI Agent 提供桌面和浏览器自动化能力，共含 36 个工具，覆盖屏幕截图、窗口管理、鼠标键盘操作、剪贴板以及 Chrome 浏览器标签页与页面交互。核心使用 Rust 实现并采用 MIT 许可证；Windows 支持完整，macOS 需授予辅助功能与屏幕录制权限，Linux 因桌面环境及 Wayland/X11 差异而仅部分受限支持。", "why": "它把原生桌面控制和 Chrome 浏览器自动化统一到标准 MCP 接口中，工具覆盖完整，适合需要 computer-use 能力的 Agent 与自动化工作流。", "tags": [ "MCP", "桌面自动化", "浏览器自动化", "36 工具", "Rust", "MIT", "AI Agent", "Chrome", "ai-agent", "browser-automation", "chrome-automation", "claude" ], "language": "Rust / JavaScript", "stars": 304, "forks": 36, "license": "MIT", "install": "npm install -g nuphus-mcp", "compatibility": "Windows：完整支持；macOS：需授予辅助功能与屏幕录制权限；Linux：部分支持，受桌面环境及 Wayland/X11 限制。", "evidence": "已核验 GitHub 仓库、README、工具清单、安装说明、平台兼容性与 MIT 许可证。", "metadataVerifiedAt": "2026-09-16T02:17:47.912Z", "source": "GitHub 官方仓库", "video": "", "icon": "bot", "color": "", "createdAt": "2026-08-03T05:14:28.239Z", "updatedAt": "2026-09-15T15:57:25Z" }, { "id": "github-com-block-buzz", "titleZh": "人类与AI Agent共建协作空间", "name": "buzz", "repo": "https://github.com/block/buzz", "homepage": "", "summary": "可自托管的人类与AI Agent共享工作空间，把沟通、代码补丁、审查、工作流和Git事件统一为可签名、可审计的事件流。", "details": "Buzz让人类与Agent进入同一社区和频道协作，支持仓库、补丁、代码审查、工作流、画布与Agent编排；底层基于Nostr relay保存签名事件，强调自托管、身份一致性和审计轨迹。当前仍是快速演进的开发者协作平台，正式使用前应核验部署、安全边界、Agent权限、备份与升级策略。", "why": "它不是单纯聊天界面，而是尝试把人类沟通、Agent执行和软件交付放进同一条可审计事件链，适合研究AI原生团队协作。", "tags": [ "AI Agent", "协作空间", "代码审查", "工作流", "Nostr", "自托管", "Rust", "Apache-2.0" ], "language": "Rust / TypeScript / Dart", "stars": 32930, "forks": 4319, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音原视频页面、GitHub仓库主页、官方README及许可证信息已核验。", "metadataVerifiedAt": "2026-09-16T02:17:48.060Z", "source": "抖音【AI腻味】《盘点本周github AI热榜》", "video": "https://www.douyin.com/video/7667879568950609139", "icon": "auto", "color": "", "createdAt": "2026-08-01T09:45:00Z", "updatedAt": "2026-09-16T02:15:19Z" }, { "id": "github-com-bojieli-ai-agent-book", "titleZh": "深入理解AI Agent设计与工程实", "name": "ai-agent-book", "repo": "https://github.com/bojieli/ai-agent-book", "homepage": "https://bojieli.github.io/ai-agent-book/", "summary": "从Agent原理到生产工程的中文开源书，包含10章正文、95个实验、PDF/EPUB及多语言版本。", "details": "全书围绕“Agent = LLM + 上下文 + 工具”，系统覆盖上下文工程、记忆、工具、评测、多Agent与生产实践，并提供95个配套实验和在线/PDF/EPUB版本。实验依赖和外部模型可能产生费用，工程结论应结合当前模型、框架版本与真实环境复核。", "why": "同时提供体系化理论、可运行实验和中文完整正文，适合作为AI Agent学习、团队培训及工程设计参考。", "tags": [ "AI Agent", "开源书", "上下文工程", "记忆", "工具调用", "评测", "多Agent", "Python", "agent", "agent-memory", "ai-agent", "book" ], "language": "Python / JavaScript / TeX", "stars": 47669, "forks": 5317, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音原视频页面、GitHub仓库主页、官方README及许可证信息已核验。", "metadataVerifiedAt": "2026-09-16T02:17:47.961Z", "source": "抖音【AI腻味】《盘点本周github AI热榜》", "video": "https://www.douyin.com/video/7667879568950609139", "icon": "auto", "color": "", "createdAt": "2026-08-01T09:45:00Z", "updatedAt": "2026-09-16T02:16:45Z" }, { "id": "github-com-koala73-worldmonitor", "titleZh": "AI全球情报与基础设施监控台", "name": "worldmonitor", "repo": "https://github.com/koala73/worldmonitor", "homepage": "https://worldmonitor.app", "summary": "聚合全球新闻、地缘冲突、市场、基础设施和风险信号的实时态势感知仪表盘。", "details": "World Monitor以地图和多数据层聚合新闻、地缘政治、市场及基础设施信号，并提供AI摘要和MCP等扩展能力。数据来自多种第三方来源，延迟、缺失、误报和AI摘要偏差不可避免；它适合态势浏览，不应作为安全、军事、投资或应急决策的唯一来源。仓库README标注AGPL-3.0，闭源网络服务集成前需核验合规义务。", "why": "把分散的全球新闻和基础设施信号集中到统一实时界面，对研究、风险观察和情报工作流具有较高参考价值。", "tags": [ "全球情报", "新闻聚合", "地缘政治", "基础设施", "实时地图", "MCP", "TypeScript", "agent", "ai", "dashboard", "geopolitics", "mcp" ], "language": "TypeScript / JavaScript / CSS", "stars": 86489, "forks": 13110, "license": "AGPL-3.0", "install": "", "compatibility": "", "evidence": "抖音原视频页面、GitHub仓库主页、官方README及许可证信息已核验。", "metadataVerifiedAt": "2026-09-16T02:17:48.123Z", "source": "抖音【AI腻味】《盘点本周github AI热榜》", "video": "https://www.douyin.com/video/7667879568950609139", "icon": "auto", "color": "", "createdAt": "2026-08-01T09:45:00Z", "updatedAt": "2026-09-16T02:06:50Z" }, { "id": "github-com-ayghri-i-have-adhd", "titleZh": "让编程Agent先给结论的表达技能", "name": "i-have-adhd", "repo": "https://github.com/ayghri/i-have-adhd", "homepage": "", "summary": "面向Claude Code、Codex等编程Agent的输出风格技能，强调行动优先、编号步骤和减少无效铺垫。", "details": "该Skill用于改变编程助手的表达结构：先说结论和动作，再给关键步骤，减少冗长铺垫；支持Claude Code、Codex及其他Agent。它改善的是信息呈现而非模型正确性，不会替代测试、来源核验或安全审查，也不等同医疗建议或ADHD诊断。", "why": "配置轻量、效果直观，可降低长篇Agent输出的阅读成本，适合任务执行和故障处理场景。", "tags": [ "Agent Skills", "Claude Code", "Codex", "表达优化", "ADHD友好", "效率", "adhd", "claude-", "claude-code-plugin", "claude-skills", "developer-tools", "productivity" ], "language": "Python / TypeScript / JavaScript", "stars": 46303, "forks": 2690, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音原视频页面、GitHub仓库主页、官方README及许可证信息已核验。", "metadataVerifiedAt": "2026-09-16T02:17:48.164Z", "source": "抖音【AI腻味】《盘点本周github AI热榜》", "video": "https://www.douyin.com/video/7667879568950609139", "icon": "auto", "color": "", "createdAt": "2026-08-01T09:45:00Z", "updatedAt": "2026-09-16T02:13:28Z" }, { "id": "github-com-earendil-works-pi", "titleZh": "统一LLM接口与终端编程Agent工", "name": "pi", "repo": "https://github.com/earendil-works/pi", "homepage": "https://pi.dev", "summary": "整合多模型LLM API、Agent运行时、工具状态管理、终端UI和交互式编程Agent CLI的工具箱。", "details": "Pi monorepo包含统一多Provider LLM API、Agent Core、终端TUI和Coding Agent CLI，可用于构建或直接运行终端Agent，并支持工具调用和状态管理。项目迭代快、版本多，集成前应锁定版本、核验Provider费用、权限边界、插件与工具执行安全。", "why": "把构建Coding Agent常见的模型接口、Agent Loop、终端界面和CLI集中在同一套模块中，适合二次开发和架构研究。", "tags": [ "Coding Agent", "LLM API", "Agent Runtime", "TUI", "CLI", "TypeScript", "MIT" ], "language": "TypeScript / JavaScript / C", "stars": 105767, "forks": 13293, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音原视频页面、GitHub仓库主页、官方README及许可证信息已核验。", "metadataVerifiedAt": "2026-09-16T02:17:48.456Z", "source": "抖音【AI腻味】《盘点本周github AI热榜》", "video": "https://www.douyin.com/video/7667879568950609139", "icon": "auto", "color": "", "createdAt": "2026-08-01T09:45:00Z", "updatedAt": "2026-09-16T02:16:47Z" }, { "id": "github-com-citrolabs-ego-lite", "titleZh": "可复用登录状态的AI Agent专用", "name": "ego-lite", "repo": "https://github.com/citrolabs/ego-lite", "homepage": "https://lite.ego.app", "summary": "让人和AI Agent并行使用浏览器，并在独立空间中复用本机登录状态执行多项网页任务。", "details": "ego-lite主打本地浏览器状态复用和Agent并行Spaces，Agent可在不打扰用户当前标签页的情况下执行网页自动化。复用登录态意味着Agent可能接触高权限会话；使用前必须限制任务范围、隔离敏感账号、检查数据上传与扩展权限，并避免在未确认时执行发布、支付、删除等副作用操作。当前README下载入口重点提供macOS版本，其他平台支持应以官方文档为准。", "why": "解决浏览器Agent反复登录和抢占用户标签页的问题，对多任务网页自动化和Coding Agent浏览器协作有实用价值。", "tags": [ "Browser Agent", "浏览器自动化", "登录状态", "并行任务", "Codex", "Claude Code", "JavaScript", "agent-skills", "ai-agent", "automation", "browser", "browser-automation" ], "language": "JavaScript / TypeScript / Shell", "stars": 15972, "forks": 834, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音原视频页面、GitHub仓库主页、官方README及许可证信息已核验。", "metadataVerifiedAt": "2026-09-16T02:17:48.344Z", "source": "抖音【AI腻味】《盘点本周github AI热榜》", "video": "https://www.douyin.com/video/7667879568950609139", "icon": "auto", "color": "", "createdAt": "2026-08-01T09:45:00Z", "updatedAt": "2026-09-16T02:17:33Z" }, { "id": "github-com-jane-xiaoer-claude-skill-web-clone", "titleZh": "真实源码优先的网站复刻技能", "name": "claude-skill-web-clone", "repo": "https://github.com/Jane-xiaoer/claude-skill-web-clone", "homepage": "", "summary": "面向Claude Code、Codex和Cursor的网站复刻方法论与可执行探针，覆盖静态站、React/Vue/Next、多页面及WebGL重前端。", "details": "项目提供从真源码检索、浏览器侦察、资源抓取、路由爬取、交互状态记录、源码映射追踪，到重建和多视口视觉验收的完整工作流。其核心原则是“真源码优先”，对AI生成的二手分析和可执行代码默认不信任，必须逐行核验。适合授权的网站迁移、离线保存、学习和自有品牌重构；不得用于绕过登录、复制私有内容、盗用品牌素材或侵犯版权。", "why": "它不是只靠截图猜CSS的一次性提示词，而是将源码、资源、交互、WebGL效果和浏览器验收组织成可重复Skill，可显著提高复杂UI复刻的真实性和可验证性。", "tags": [ "网站复刻", "UI复刻", "Agent Skills", "Claude Code", "Codex", "Cursor", "WebGL", "浏览器自动化", "真源码优先", "视觉验收", "合规使用", "MIT" ], "language": "JavaScript", "stars": 1005, "forks": 122, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音视频画面、GitHub官方API、README和SKILL.md已核验；授权、版权和私域边界已写入卡片。", "metadataVerifiedAt": "2026-09-16T02:17:48.416Z", "source": "douyin", "video": "https://www.douyin.com/video/7650344851245894950", "icon": "auto", "color": "#43b9b1", "createdAt": "2026-07-28T23:50:39.225250+00:00", "updatedAt": "2026-09-15T16:32:40Z" }, { "id": "github-com-tibixdev-winboat", "titleZh": "Linux无缝运行Windows应用", "name": "WinBoat", "repo": "https://github.com/winboat-org/winboat", "homepage": "https://www.winboat.app/", "summary": "通过容器化Windows环境与桌面集成，让Windows应用在Linux桌面中以独立窗口运行。", "details": "WinBoat面向Linux用户提供Windows应用的无缝桌面集成，底层仍需要Windows环境、容器与虚拟化支持，并非Wine兼容层。适合运行Office、Adobe及其他Windows专用软件；部署前需确认KVM、Docker/Podman、RDP、磁盘和Windows授权要求。", "why": "它把Windows虚拟环境封装为接近原生Linux应用的窗口体验，降低双系统和整桌面虚拟机切换成本；适合需要少量Windows专用软件的Linux工作站。", "tags": [ "Linux", "Windows应用", "虚拟化", "容器", "RDP", "桌面集成", "TypeScript", "开源", "MIT", "docker", "docker-compose", "linux" ], "language": "TypeScript / Vue / PowerShell", "stars": 22849, "forks": 739, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音画面明确显示WinBoat；GitHub官方仓库、API和官网已核验。", "metadataVerifiedAt": "2026-09-16T02:17:48.588Z", "source": "douyin", "video": "https://www.douyin.com/video/7634379626910960923", "icon": "auto", "color": "#5b9df6", "createdAt": "2026-07-28T07:11:51.637476+00:00", "updatedAt": "2026-09-16T02:02:24Z" }, { "id": "github-com-lorryjovens-hub-laap-agi", "titleZh": "数字生命体零LLM认知架构", "name": "laap-AGI", "repo": "https://github.com/lorryjovens-hub/laap-AGI", "homepage": "https://laap-agi.netlify.app", "summary": "面向数字生命体的实验性认知架构，将PSI核心、情景记忆、规则引擎和量子启发式推理组织在LLM之外。", "details": "LAAP由Lorry发起并有社区贡献，强调持续认知状态、PSI核心、EpisodicMemory、RulesEngine、QRE量子启发式推理及Hermes等Agent集成。README中的“觉醒”“数字生命”属于项目定位与叙事，不代表机器意识已获科学验证；2000Hz和全本地推理等性能与能力声明需在实际环境复测。", "why": "适合研究把LLM降为认知组件、在模型外建立持续状态、动机、记忆和规则系统的Agent Harness路线，也可为小七AI公司的长期AI员工状态模型提供架构参考。", "tags": [ "数字生命体", "认知架构", "Agent Harness", "PSI", "情景记忆", "规则引擎", "Rust", "Python", "Hermes集成", "实验性", "agent-framework", "agi" ], "language": "Python / Shell / PowerShell", "stars": 12, "forks": 5, "license": "AGPL-3.0", "install": "", "compatibility": "", "evidence": "抖音作者主页简介、GitHub仓库API、README及提交贡献者已核验。", "metadataVerifiedAt": "2026-09-16T02:17:48.638Z", "source": "douyin", "video": "https://www.douyin.com/video/7661192130346725477", "icon": "auto", "color": "#8b7cff", "createdAt": "2026-07-28T07:11:51.637476+00:00", "updatedAt": "2026-09-15T16:47:14Z" }, { "id": "github-com-lorryjovens-hub-claude-code-rust", "titleZh": "Rust原生多模型AI桌面编码工作台", "name": "claude-code-rust", "repo": "https://github.com/lorryjovens-hub/claude-code-rust", "homepage": "", "summary": "基于Tauri 2、Rust与React构建的社区AI桌面客户端，整合多模型聊天、MCP、Skills、多智能体、终端、代码差异和本地记忆。", "details": "该仓库当前README将项目定位为非官方、仍在开发中的Claude Desktop社区实现，而不是Anthropic官方Claude Code的等价重写。它提供Anthropic、OpenAI、DeepSeek等多模型流式聊天，本地SQLite历史与记忆，MCP、Skills、多智能体编排、终端、文件浏览、代码执行与Diff、知识库、研究、GitHub和IM集成。仓库简介宣称启动快2.5倍、体积小97%，但当前clean-main分支只有1个提交，README也明确无正式发布包、测试覆盖不均且部分能力仍属实验；性能数字应视为作者声明，使用前需自行构建、审计权限与沙箱边界并验证。", "why": "它把Rust/Tauri桌面壳、本地持久化、MCP、Skills、多智能体和编码工具集中在一个开源项目中，适合观察本地AI编码工作台的工程实现；同时应明确其非官方和早期开发属性。", "tags": [ "AI桌面客户端", "Rust", "Tauri", "React", "多模型", "MCP", "Skills", "多智能体", "本地记忆", "代码助手", "社区项目", "早期开发" ], "language": "TypeScript / Rust / Shell", "stars": 1660, "forks": 639, "license": "", "install": "", "compatibility": "", "evidence": "GitHub仓库页面、官方GitHub API与clean-main分支README已核验；非官方身份、开发状态、功能边界和作者性能声明已在卡片中区分。", "metadataVerifiedAt": "2026-09-16T02:17:48.669Z", "source": "用户提供GitHub仓库", "video": "", "icon": "auto", "color": "", "createdAt": "2026-07-28T06:38:39.054450Z", "updatedAt": "2026-09-16T02:12:02Z" }, { "id": "github-com-yuan1z0825-nature-skills", "titleZh": "Nature风格科研写作与学术工作流", "name": "nature-skills", "repo": "https://github.com/Yuan1z0825/nature-skills", "homepage": "https://yuan1z0825.github.io/nature-skills/", "summary": "面向AI科研工作流的可复用Skills合集，覆盖论文精读、Nature风格写作润色、科研绘图、文献检索、审稿回复、PPT和专利草稿。", "details": "nature-skills以SKILL.md组织可安装科研工作流，当前覆盖投稿级科研图、学术润色与写作、审稿模拟、引用与参考文献核验、全文中英对照阅读、论文精读卡片、返修回复、论文转PPT、论文转中国发明专利、多源学术检索、合法全文获取、每日文献管线和实验日志等。支持Codex、Claude Code及其他Agent，但部分技能仍标记Draft或Beta，脚本、MCP、Python/R、浏览器和第三方数据库凭据需按需单独配置。它可辅助形成科研产物，不应替代作者对事实、统计、引用、版权、期刊规则与研究伦理的最终核验。", "why": "技能覆盖科研从检索、阅读、分析、写作、绘图到投稿返修的完整链路，并强调来源锚点、可验证流程和可直接使用的产物，适合作为AI辅助科研工作流参考。", "tags": [ "科研", "Agent Skills", "Nature风格", "学术写作", "科研绘图", "论文阅读", "文献检索", "审稿回复", "PPT", "专利", "Codex", "Claude Code" ], "language": "Python / JavaScript / HTML", "stars": 42052, "forks": 2253, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "GitHub官方API、仓库README、技能索引、安装说明及Apache-2.0许可证已核验。", "metadataVerifiedAt": "2026-09-16T02:17:48.799Z", "source": "用户提供GitHub仓库", "video": "", "icon": "auto", "color": "", "createdAt": "2026-07-28T06:35:53.790909Z", "updatedAt": "2026-09-16T02:17:35Z" }, { "id": "github-com-oso95-scroll-world", "titleZh": "滚动穿越式3D世界网页生成Skill", "name": "scroll-world", "repo": "https://github.com/oso95/scroll-world", "homepage": "", "summary": "scroll-world 是面向 Claude Code、Codex 及兼容 SKILL.md 智能体的沉浸式滚动网页生成 Skill。", "details": "项目可依据品牌与行业生成连续镜头场景和连接片段，通过滚动位置驱动视频时间线，制作无切镜的“飞入世界”落地页，并支持桌面横屏及独立移动端 9:16 素材。技术栈为 JavaScript、Python 与 HTML，已核验 5,518 Stars、633 Forks。仓库标注 MIT 许可证，但使用素材及生成内容时仍须自行核查版权与第三方许可风险。", "why": "适合快速探索品牌沉浸式叙事页面，滚动驱动时间线且兼顾桌面端与移动端素材，社区关注度较高。", "tags": [ "JavaScript", "Python", "HTML", "MIT", "Claude Code", "Codex", "SKILL.md", "滚动交互" ], "language": "JavaScript / Python / HTML", "stars": 9291, "forks": 1035, "license": "MIT", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-09-16T02:17:48.893Z", "source": "抖音「AI前端仔」第159集推荐；GitHub官方API与README核验", "video": "https://www.douyin.com/video/7663046947632155938", "icon": "auto", "color": "", "createdAt": "2026-07-28T02:17:51.414Z", "updatedAt": "2026-09-16T02:09:52Z" }, { "id": "github-com-comfy-org-comfyui", "titleZh": "节点式AI图像与视频生产工作流", "name": "ComfyUI", "repo": "https://github.com/Comfy-Org/ComfyUI", "homepage": "https://www.comfy.org/", "summary": "通过可视化节点连接模型、提示词、ControlNet、图像、视频和音频处理步骤，搭建可复用的生成式AI工作流。", "details": "ComfyUI是模块化的扩散模型GUI、API和后端，支持节点式工作流、工作流JSON复现、队列执行、局部重绘、ControlNet、LoRA以及庞大的社区自定义节点生态。适合生图、分镜、视频、配音等AI内容生产，但截图中“脚本到配音全自动”通常依赖额外模型和第三方节点，并非核心仓库开箱即用；第三方节点可能执行任意代码，安装前需审查来源和权限，模型及生成素材也需遵守许可证与版权要求。", "why": "截图正文和相关搜索均明确指向ComfyUI，官方当前仓库为Comfy-Org/ComfyUI（旧地址comfyanonymous/ComfyUI已重定向）；它是节点式生成工作流领域的核心开源项目。", "tags": [ "AI", "ComfyUI", "节点工作流", "Stable Diffusion", "生图", "视频生成", "分镜", "配音", "自托管", "截图推荐", "第三方节点需审查", "版权需注意" ], "language": "Python / GLSL / Batchfile", "stars": 133404, "forks": 15782, "license": "GPL-3.0", "install": "", "compatibility": "", "evidence": "截图字幕、作者文案和“comfyui是免费的吗”相关搜索共同确认项目；GitHub官方API确认当前仓库、Stars、Forks和许可证。", "metadataVerifiedAt": "2026-09-16T02:17:48.947Z", "source": "抖音截图【小乖ai剧】ComfyUI保姆级教程", "video": "", "icon": "auto", "color": "", "createdAt": "2026-07-26T02:42:41Z", "updatedAt": "2026-09-16T02:15:53Z" }, { "id": "github-com-666ghj-mirofish", "titleZh": "多智能体社会仿真与预测引擎", "name": "MiroFish", "repo": "https://github.com/666ghj/MiroFish", "homepage": "https://mirofish.ai", "summary": "利用群体智能和多智能体社会仿真，根据新闻、政策、报告等种子材料构建平行数字世界并生成情景预测报告。", "details": "MiroFish会从用户上传的种子资料构建知识图谱，生成具有记忆、性格和立场的AI角色，让它们在模拟社会中交流、结盟、争论和演化，再汇总为预测报告。适合舆情、政策、市场和故事情景推演研究，但它生成的是模型驱动的情景模拟，不是真实世界的确定预测；结果高度依赖输入资料、模型偏差和参数设定，不能直接代替投资、医疗、法律或公共决策。项目采用AGPL-3.0，自托管或网络服务二次开发应遵守开源义务。", "why": "视频章节摘要明确点名MiroFish，并描述群体智能、平行数字世界、百万Agent并发和上传资料生成预测报告，与666ghj/MiroFish官方项目完全一致。", "tags": [ "抖音推荐", "多智能体", "群体智能", "社会仿真", "情景预测", "知识图谱", "AI Agent", "自托管", "agent-memory", "financial-forecasting", "future-prediction", "knowledge-graph" ], "language": "Python / Vue / JavaScript", "stars": 73651, "forks": 11320, "license": "AGPL-3.0", "install": "", "compatibility": "", "evidence": "抖音标题、章节摘要、项目名称MiroFish、GitHub官方API和许可证已交叉核验", "metadataVerifiedAt": "2026-09-16T02:17:49.017Z", "source": "抖音【开源项目盘点】《多智能体仿真引擎，预测万物推演未来》", "video": "https://www.douyin.com/video/7619703359829724451", "icon": "auto", "color": "", "createdAt": "2026-07-19T17:41:44.169412Z", "updatedAt": "2026-09-16T01:42:49Z" }, { "id": "github-com-ihmily-douyinliverecorder", "titleZh": "多平台直播自动录制工具", "name": "DouyinLiveRecorder", "repo": "https://github.com/ihmily/DouyinLiveRecorder", "homepage": "", "summary": "支持抖音、TikTok、YouTube、快手、B站、Twitch等40多个平台的循环值守、多人并发直播录制工具。", "details": "DouyinLiveRecorder可按直播间地址循环监测开播状态，并自动调用FFmpeg录制，支持多个主播并发、分段保存、录制格式选择、转码、Cookie或代理配置以及开播和录制通知。适合个人直播存档和已获授权的内容归档；使用时需遵守平台条款、主播版权和所在地法律，不应绕过访问控制、付费限制或用于未经授权的批量采集。", "why": "覆盖平台广、支持无人值守与多人并发录制，适合需要长期归档公开直播或自有直播内容的用户。", "tags": [ "直播录制", "抖音", "TikTok", "YouTube", "FFmpeg", "自动化", "多平台", "Python", "acfun-live", "douyin", "douyin-api", "douyin-live" ], "language": "Python / JavaScript / HTML", "stars": 10880, "forks": 1449, "license": "MIT", "install": "", "compatibility": "", "evidence": "GitHub官方仓库API、README定位和许可证已核验", "metadataVerifiedAt": "2026-09-16T02:17:49.147Z", "source": "用户直接提供GitHub仓库", "video": "", "icon": "auto", "color": "", "createdAt": "2026-07-25T01:32:22Z", "updatedAt": "2026-09-15T18:11:43Z" }, { "id": "github-com-graphify-labs-graphify", "titleZh": "一条命令生成代码知识图谱", "name": "graphify", "repo": "https://github.com/Graphify-Labs/graphify", "homepage": "https://www.graphify.com", "summary": "把代码、文档、SQL结构、配置和PDF转换成可查询知识图谱，让Claude Code、Codex、Cursor等Agent更准确理解大型项目。", "details": "Graphify通过tree-sitter进行本地确定性AST解析，把代码实体及调用、继承、依赖等关系组织成可查询图谱，并区分直接提取的EXTRACTED事实与模型推断的INFERRED关系。它支持增量更新和持久化，可作为Claude Code、Codex、Cursor、Gemini CLI等工具的Skill使用。代码建图默认无需向量数据库，但文档和媒体的语义处理可能调用模型；图谱不是源码本身，推断边和高风险修改仍应回到代码与测试验证。", "why": "视频标题和章节摘要明确介绍Graphify将项目转换为知识图谱、三级置信度、持久化与增量更新；与Graphify-Labs/graphify官方定位一致。", "tags": [ "抖音推荐", "代码知识图谱", "AI编程", "AST", "tree-sitter", "Claude Code", "Codex", "Agent Skill", "ai-agents", "antigravity", "ast", "claude-code" ], "language": "Python / Dockerfile", "stars": 118065, "forks": 11411, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音页面章节摘要、项目名与搜索词、GitHub官方API和README已交叉核验", "metadataVerifiedAt": "2026-09-16T02:17:49.234Z", "source": "抖音【ـﮩﮩ٨ـ♡ﮩ٨ـﮩﮩ】《GitHub星探：Graphify》", "video": "https://www.douyin.com/video/7662265652635274547", "icon": "auto", "color": "#8b5cf6", "createdAt": "2026-07-23T12:56:13+00:00", "updatedAt": "2026-09-16T02:17:17Z" }, { "id": "github-com-hoainho-img2threejs", "titleZh": "单图生成可动画Three.js 3D", "name": "img2threejs", "repo": "https://github.com/img2threejs/img2threejs", "homepage": "https://hoainho.github.io/img2threejs-showcase/", "summary": "根据单张参考图，让AI编码Agent重建代码驱动、程序化、可质检并可动画的Three.js模型，适合网页展示、商品可视化和游戏资产原型。", "details": "img2threejs不是传统扩散式网格生成器，而是一套面向Claude Code、Codex等编码Agent的图像到程序化Three.js建模工作流。它先分析参考图的形状、材质和结构，再用代码搭建几何体、材质、灯光与动画，并通过渲染对比和质量门禁迭代。生成结果便于继续修改和嵌入网页，但单图存在被遮挡面、真实尺寸与材质信息缺失，复杂产品仍需人工校模，并须确认参考图及生成资产的版权。", "why": "抖音页面明确出现搜索词“img2threejs免费吗”，视频主题为单图生成网页3D、电商模型和游戏资产；与hoainho/img2threejs的名称、定位和官方描述一致。", "tags": [ "抖音推荐", "图片转3D", "Three.js", "AI编程", "程序化建模", "网页3D", "电商展示", "游戏资产", "3d", "ai-agents", "claude-code", "computer-graphics" ], "language": "Python / JavaScript / Shell", "stars": 16143, "forks": 1339, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音标题、页面搜索词img2threejs、GitHub官方API和仓库描述已交叉核验", "metadataVerifiedAt": "2026-09-16T02:17:49.233Z", "source": "抖音【阿甘探AI】《一张图生成3D模型》", "video": "https://www.douyin.com/video/7665435663575076139", "icon": "auto", "color": "", "createdAt": "2026-07-24T07:24:36Z", "updatedAt": "2026-09-16T01:58:32Z" }, { "id": "github-com-greensock-gsap-skills", "titleZh": "AI网页动画技能包", "name": "gsap-skills", "repo": "https://github.com/greensock/gsap-skills", "homepage": "https://gsap.com", "summary": "GSAP官方面向AI编码Agent的网页动画技能包，让Codex、Claude Code、Cursor等工具正确生成时间轴、滚动、路径和交互动效。", "details": "仓库提供GSAP官方AI Skills，覆盖核心动画API、ScrollTrigger、MotionPath、Flip及React、Vue、Svelte和原生JavaScript集成方式，并包含常见模式、性能建议、插件使用和排错指导。它是指导Agent写动画的技能资料，不等于零代码动画生成器；最终页面仍需在真实浏览器验证性能、响应式布局、可访问性和授权边界。", "why": "视频画面清晰显示GreenSock/gsap-skills仓库及“Official AI skills for GSAP”，与视频“AI写前端补齐网页动画”主题完全一致。", "tags": [ "抖音推荐", "AI编程", "网页动画", "GSAP", "Agent Skill", "前端开发", "ScrollTrigger", "交互动效", "MIT" ], "language": "Markdown / JSON / JavaScript / TypeScript / JSX", "stars": 15069, "forks": 895, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音视频逐帧画面、GitHub官方API、仓库描述和许可证已核验", "metadataVerifiedAt": "2026-09-09T02:17:15.100Z", "source": "抖音【程序员虾仁】《AI写前端终于不只会页面了》", "video": "https://www.douyin.com/video/7644510312363798907", "icon": "spark", "color": "", "createdAt": "2026-07-19T20:03:56.712Z", "updatedAt": "2026-09-09T01:22:34Z" }, { "id": "github-com-1c7-chinese-independent-developer", "titleZh": "中国独立开发者项目大全", "name": "chinese-independent-developer", "repo": "https://github.com/1c7/chinese-independent-developer", "homepage": "", "summary": "收集中国独立开发者及其产品，方便发现个人开发者正在制作的应用、网站、工具与创业项目。", "details": "这是一个社区维护的中国独立开发者项目列表，按开发者和产品汇总大家正在做什么，适合寻找产品灵感、同类项目、独立开发案例和合作对象。它是项目索引而非质量认证；收录不代表代码开源、持续维护或商业可信，使用其中产品前仍需单独核验隐私、价格、许可证与服务状态。", "why": "视频画面明确显示“GitHub上一个近58.5k star的中国独立开发者项目列表”，与1c7/chinese-independent-developer的仓库名称、描述和星标规模一致。", "tags": [ "抖音推荐", "独立开发者", "中国开发者", "项目合集", "产品灵感", "创业", "开源清单", "社区资源", "china", "indie", "indie-developer", "Python" ], "language": "Markdown / JSON / Python", "stars": 61410, "forks": 5382, "license": "", "install": "", "compatibility": "", "evidence": "抖音视频画面与评论搜索词、GitHub官方API和仓库描述已交叉核验", "metadataVerifiedAt": "2026-09-13T02:17:15.399Z", "source": "抖音【花开富贵文定吉祥】《行业冷知识科普》", "video": "https://www.douyin.com/video/7663882960270038897", "icon": "auto", "color": "", "createdAt": "2026-07-24T06:41:33Z", "updatedAt": "2026-09-13T00:32:43Z" }, { "id": "github-com-memvid-memvid", "titleZh": "Agent本地永久记忆引擎", "name": "memvid", "repo": "https://github.com/memvid/memvid", "homepage": "https://www.memvid.com", "summary": "面向AI Agent的本地记忆引擎，把文本、文档和对话写入单文件知识库，并通过语义检索快速召回长期上下文。", "details": "Memvid提供Python SDK和本地优先的持久化记忆能力，可创建brain文件、持续添加上下文并按自然语言搜索；适合Agent长期记忆、RAG与离线知识库。接入生产前仍需测试检索质量、并发写入、备份恢复、敏感数据保护及SDK版本兼容性。", "why": "视频明确演示memvid-sdk、Memory、add和search三步工作流，与官方memvid仓库定位一致。", "tags": [ "抖音推荐", "Agent记忆", "长期记忆", "RAG", "本地优先", "向量检索", "Python SDK", "知识库", "ai", "context", "embedded", "faiss" ], "language": "Rust", "stars": 16541, "forks": 1420, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音原视频页面、GitHub官方仓库API与项目元数据已核验", "metadataVerifiedAt": "2026-09-12T02:17:12.884Z", "source": "抖音【AI小杨】《Memvid 登顶 GitHub 从零到 Agent 永久记忆》", "video": "https://www.douyin.com/video/7665563483269827878", "icon": "auto", "color": "", "createdAt": "2026-07-24T02:55:15Z", "updatedAt": "2026-09-11T13:36:29Z" }, { "id": "github-com-0x0funky-agent-sprite-forge", "titleZh": "Codex二维游戏精灵与帧动画工坊", "name": "agent-sprite-forge", "repo": "https://github.com/0x0funky/agent-sprite-forge", "homepage": "", "summary": "把自然语言和参考图转换成2D精灵表、透明PNG帧、GIF、地图及可导入Godot或Unity的游戏资产。", "details": "项目提供generate2dsprite、generate2dmap和video2dsprite工作流：由Agent规划和生成原始视觉，再用确定性脚本完成去背、切帧、对齐、透明导出、地图分层、碰撞与元数据。Codex可运行精灵和地图流程；video2dsprite生成阶段依赖Grok Build的image_to_video。输出仍需人工检查帧一致性、透明边缘、引擎导入和素材版权，示例动漫角色不可默认商用。", "why": "视频章节摘要直接点名agent-sprite-forge和Generate2D Sprite，功能与官方仓库说明完全一致。", "tags": [ "抖音推荐", "Codex Skill", "游戏开发", "2D精灵", "帧动画", "Sprite Sheet", "像素动画", "Godot", "Unity", "透明PNG", "MIT", "agent-skills" ], "language": "Markdown / Python", "stars": 4082, "forks": 395, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音章节摘要与相关搜索、GitHub官方API、README、Skill目录和LICENSE已核验", "metadataVerifiedAt": "2026-09-13T02:17:15.204Z", "source": "抖音【小螃蟹的游戏开发日志】《用Codex制作游戏帧动画》", "video": "https://www.douyin.com/video/7650613797056498970", "icon": "auto", "color": "#d59b43", "createdAt": "2026-07-23T18:32:32Z", "updatedAt": "2026-09-13T02:14:58Z" }, { "id": "github-com-hkuds-deeptutor", "titleZh": "开源个性化AI家教与学习工作台", "name": "DeepTutor", "repo": "https://github.com/HKUDS/DeepTutor", "homepage": "https://deeptutor.info/", "summary": "港大数据智能实验室开源的个性化AI学习工作台，支持课程资料问答、引导式学习、题库、记忆和知识库。", "details": "DeepTutor提供文档与知识库RAG、Guided Learning、Question Bank、Book、Co-Writer、Deep Research/Solve、可视化、Partner Agents和三层记忆，并支持Docker、多用户隔离、OpenAI/Anthropic及OpenAI兼容或本地模型。源码免费开源不代表云模型、搜索、解析和算力完全免费；AI教学内容可能出错，重要知识、考试答案和专业建议仍需教师或权威资料复核。", "why": "视频明确标注DeepTutor和港大开源AI家教，Stars增长描述与官方README的“111天突破2万星”记录一致。", "tags": [ "抖音推荐", "AI教育", "AI家教", "个性化学习", "RAG", "知识库", "题库", "引导式学习", "本地模型", "Docker", "Apache-2.0", "ai-agents" ], "language": "Python", "stars": 39062, "forks": 4918, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音标题与DeepTutor标签、GitHub官方仓库API、README、版本记录与LICENSE已核验", "metadataVerifiedAt": "2026-09-09T02:17:15.354Z", "source": "抖音【AI替你干活】《港大开源AI家教》", "video": "https://www.douyin.com/video/7664757379447820666", "icon": "auto", "color": "#5b9df6", "createdAt": "2026-07-23T18:21:25Z", "updatedAt": "2026-09-09T02:16:01Z" }, { "id": "github-com-firecrawl-firecrawl", "titleZh": "面向AI的网页搜索与数据抓取平台", "name": "firecrawl", "repo": "https://github.com/firecrawl/firecrawl", "homepage": "https://firecrawl.dev", "summary": "把网页搜索、抓取、交互和站点遍历结果转换为适合AI使用的Markdown、JSON、截图和结构化数据。", "details": "Firecrawl支持JS重页面、搜索、单页抓取、整站Crawl、URL Map、批量抓取、页面交互、媒体解析、Agent、Skill和MCP。视频所称“免API调用”主要对应CLI/Skill的浏览器模式或自部署开源版；官方托管API仍要求API Key，规模化代理、反爬和运行资源也不是零成本。使用时应遵守robots、目标站条款、版权、隐私和访问频率限制。", "why": "视频标题、章节摘要与官方Firecrawl的Markdown/JSON/截图、动态页面和Agent数据获取能力逐项吻合。", "tags": [ "抖音推荐", "网页抓取", "爬虫", "AI数据", "Markdown", "结构化数据", "Agent Skill", "MCP", "自托管", "AGPL-3.0", "合规抓取", "ai" ], "language": "TypeScript", "stars": 179238, "forks": 9751, "license": "AGPL-3.0", "install": "", "compatibility": "", "evidence": "抖音章节摘要、GitHub官方仓库API、README、CLI/Skill说明与LICENSE已核验", "metadataVerifiedAt": "2026-09-12T02:17:13.062Z", "source": "抖音【二次螈Ai工具箱】《firecrawl免API调用》", "video": "https://www.douyin.com/video/7658567098902280369", "icon": "globe", "color": "#ff6759", "createdAt": "2026-07-23T18:12:58Z", "updatedAt": "2026-09-12T02:06:53Z" }, { "id": "github-com-openhands-openhands", "titleZh": "自托管AI编码Agent控制中心", "name": "OpenHands", "repo": "https://github.com/OpenHands/OpenHands", "homepage": "https://openhands.dev", "summary": "把OpenHands、Claude Code、Codex、Gemini等编码Agent统一放进自托管控制台运行和自动化。", "details": "OpenHands Agent Canvas支持本地、Docker、远程和云端Agent后端、ACP兼容Agent及自动化集成。官方代码正拆分到agent-canvas、software-agent-sdk和automation；无沙箱模式会直接访问主机文件，应优先使用Docker或VM隔离。", "why": "视频项目线索指向OpenHands，官方“自托管开发者控制中心”定位与内容吻合。", "tags": [ "抖音推荐", "AI编程", "Agent", "自托管", "Claude Code", "Codex", "Gemini", "ACP", "Docker沙箱", "agent", "artificial-intelligence", "chatgpt" ], "language": "TypeScript / JavaScript / Python", "stars": 82724, "forks": 10637, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音项目线索、GitHub官方API、README与LICENSE已核验", "metadataVerifiedAt": "2026-08-01T02:17:43.757Z", "source": "抖音【兔叽准点溜】《全能项目又来了》", "video": "https://www.douyin.com/video/7658219212044864809", "icon": "code", "color": "#8b5cf6", "createdAt": "2026-07-23T17:44:23Z", "updatedAt": "2026-08-01T01:37:10Z" }, { "id": "github-com-dontbesilent2025-dbskill", "titleZh": "多维文案与商业诊断技能箱", "name": "dbskill", "repo": "https://github.com/dontbesilent2025/dbskill", "homepage": "https://skills.sh/dontbesilent2025/dbskill", "summary": "面向创业者和内容创作者的28个AI Skills，覆盖商业诊断、选题、标题、开头、逐字稿和传播优化。", "details": "dbskill通过/dbs路由到合适Skill，可做内容方向、短视频开头、小红书标题、共鸣、逻辑和传播性检查，也包含商业模式、定价、行动与知识库治理。采用CC BY-NC 4.0，商业用途需要单独授权。", "why": "视频标题是多维文案优化，页面相关搜索明确出现dbskill，仓库能力完全吻合。", "tags": [ "抖音推荐", "Agent Skill", "文案优化", "内容创作", "商业诊断", "短视频", "Claude Code", "Codex", "非商业许可", "JavaScript", "NOASSERTION" ], "language": "JavaScript / Python / Shell", "stars": 9061, "forks": 1055, "license": "NOASSERTION", "install": "", "compatibility": "", "evidence": "抖音“dbskill”线索、GitHub官方API、README与LICENSE已核验", "metadataVerifiedAt": "2026-08-01T02:17:43.979Z", "source": "抖音【Max的AI干货】《多维优化文案的skill》", "video": "https://www.douyin.com/video/7665231032383180095", "icon": "auto", "color": "#ec5ba7", "createdAt": "2026-07-23T17:44:23Z", "updatedAt": "2026-08-01T02:12:07Z" }, { "id": "github-com-juneyaooo-nihaisha-nishi-tcm", "titleZh": "倪海厦中医课程学习助手", "name": "nihaisha-nishi-tcm", "repo": "https://github.com/JuneYaooo/nihaisha-nishi-tcm", "homepage": "", "summary": "把倪海厦课程整理成可检索、可追溯的Agent Skill，用于课程复习、方证穴位比较和证据定位。", "details": "项目覆盖伤寒、金匮、内经、本草和针灸等课程，提供轻量检索、截图/PDF证据及可选RAG知识图谱；完整RAG约3.68GB。仓库无LICENSE，README声明仅个人学习和非商业交流；明确不用于诊断、处方、剂量或自我用药。", "why": "视频和相关搜索均指向倪海厦skill；该仓库是当前同主题中热度最高且描述最吻合的学习型Skill。", "tags": [ "抖音推荐", "Agent Skill", "中医学习", "课程检索", "RAG", "知识图谱", "证据索引", "非医疗建议", "许可未明确", "acupuncture", "agent-skills", "claude-code-skills" ], "language": "Python", "stars": 2005, "forks": 433, "license": "", "install": "", "compatibility": "", "evidence": "抖音“倪海厦skill”线索、GitHub热度、README和目录结构已核验", "metadataVerifiedAt": "2026-08-24T02:17:57.691Z", "source": "抖音【欧耶学AI】《倪海厦AI智能体》", "video": "https://www.douyin.com/video/7661597019413010085", "icon": "auto", "color": "#48ae78", "createdAt": "2026-07-23T17:44:23Z", "updatedAt": "2026-08-24T01:12:17Z" }, { "id": "github-com-opencut-app-opencut", "titleZh": "", "name": "OpenCut", "repo": "https://github.com/OpenCut-app/OpenCut", "homepage": "https://opencut.app", "summary": "免费开源的 CapCut 替代视频编辑器，目标覆盖浏览器、桌面和移动端。", "details": "OpenCut 以无订阅、无水印和本地处理为方向，正在重写为插件优先架构，并规划 Rust 核心、Editor API、MCP、无头批处理以及跨 Web/桌面/移动端的统一代码库。当前主分支仍处于从头重写阶段，官方明确建议需要立即使用时选择经典版本；因此它适合关注和参与演进，不应把路线图中的 MCP、移动端或自动化能力当作已经稳定交付。", "why": "视频第 1 个项目明确展示 OpenCut 仓库、CapCut 替代定位与插件/MCP 路线图，和官方仓库逐项吻合。", "tags": [ "抖音推荐", "视频剪辑", "CapCut 替代", "开源", "本地优先", "跨平台", "MCP", "插件", "重写中", "MIT", "editor", "oss" ], "language": "TypeScript / Rust / CSS", "stars": 80187, "forks": 7962, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音视频逐帧、GitHub 官方仓库 API、语言统计与 README 已核验", "metadataVerifiedAt": "2026-08-01T02:17:44.431Z", "source": "抖音《GitHub 本周热榜来了！开发者集体在卷 AI Agent》", "video": "https://www.douyin.com/video/7664073102514523392", "icon": "auto", "color": "#ff6759", "createdAt": "2026-07-23T12:56:13+00:00", "updatedAt": "2026-08-01T02:07:53Z" }, { "id": "github-com-iofficeai-officecli", "titleZh": "", "name": "OfficeCLI", "repo": "https://github.com/iOfficeAI/OfficeCLI", "homepage": "https://officecli.ai", "summary": "让 AI Agent 通过单文件 CLI、MCP 和 Skill 创建、读取及修改 Word、Excel 与 PowerPoint。", "details": "OfficeCLI 是无需安装 Microsoft Office 的跨平台单文件工具，内置 DOCX、XLSX、PPTX 的读写、渲染、结构化查询、模板合并、Excel 公式与数据透视表能力，并可通过 MCP 或自动安装的 Skill 接入 Codex、Claude Code、Cursor 和 Copilot。它适合无头服务器与文档流水线，但生成文档仍应在交付前进行版式、公式、数据和素材授权检查，不能只依据命令成功判断成品正确。", "why": "视频第 8 个项目展示 iOfficeAI/OfficeCLI、无需 Office、内置 HTML 渲染和 MCP Server，与官方说明完全一致。", "tags": [ "抖音推荐", "Office", "Word", "Excel", "PowerPoint", "CLI", "MCP", "Agent Skill", "文档自动化", "无头运行", "Apache-2.0", "agent" ], "language": "C# / Shell / JavaScript", "stars": 23852, "forks": 1603, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音视频逐帧、GitHub 官方仓库 API、语言统计与 README 已核验", "metadataVerifiedAt": "2026-08-01T02:17:44.714Z", "source": "抖音《GitHub 本周热榜来了！开发者集体在卷 AI Agent》", "video": "https://www.douyin.com/video/7664073102514523392", "icon": "auto", "color": "#20b8a6", "createdAt": "2026-07-23T12:56:13+00:00", "updatedAt": "2026-08-01T02:14:02Z" }, { "id": "github-com-openinterpreter-openinterpreter", "titleZh": "", "name": "openinterpreter", "repo": "https://github.com/openinterpreter/openinterpreter", "homepage": "http://openinterpreter.com/", "summary": "面向 Kimi、DeepSeek 等低成本或开放模型的 Codex 兼容编码智能体。", "details": "Open Interpreter 现已基于 OpenAI Codex 代码重构，重点复现不同模型最适合的 Agent Harness，可在 native、Claude Code、Kimi Code、Qwen Code、DeepSeek TUI 等多种执行框架之间切换，并兼容 ACP 与 Codex exec 协议。它支持 MCP、Skills、Hooks、AGENTS.md、浏览器与桌面 QA，以及本地会话配置；由于会执行命令并控制计算机，应保留权限确认、在隔离环境运行，并审查模型生成的每一次高风险操作。", "why": "视频第 9 个项目展示 openinterpreter/openinterpreter、Kimi K3、跨平台 TUI 与 Codex 兼容性，和迁移后的官方仓库一致。", "tags": [ "抖音推荐", "AI 编程", "Coding Agent", "低成本模型", "Kimi", "DeepSeek", "Codex 兼容", "ACP", "Harness", "MCP", "高权限", "acp" ], "language": "Rust / Python / Starlark", "stars": 67476, "forks": 5800, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音视频逐帧、GitHub 官方仓库 API、语言统计与 README 已核验", "metadataVerifiedAt": "2026-08-01T02:17:44.938Z", "source": "抖音《GitHub 本周热榜来了！开发者集体在卷 AI Agent》", "video": "https://www.douyin.com/video/7664073102514523392", "icon": "terminal", "color": "#f59e0b", "createdAt": "2026-07-23T12:56:13+00:00", "updatedAt": "2026-08-01T00:44:06Z" }, { "id": "github-com-openai-codex", "titleZh": "", "name": "codex", "repo": "https://github.com/openai/codex", "homepage": "", "summary": "OpenAI 官方开源的本地终端编码智能体，可通过 ChatGPT 登录或 API 使用。", "details": "Codex CLI 在本地终端读取和修改代码、运行命令、执行测试并协助完成工程任务，核心实现以 Rust 为主，同时提供 SDK、IDE 集成和桌面入口。它支持使用 ChatGPT 计划登录，也可配置 API；仓库对应开源 CLI 和相关基础组件，不等同于所有 Codex 云端或桌面产品代码。由于它能够操作本地文件和命令，应按项目设置权限、审查差异并保留回滚点。", "why": "视频第 10 个项目直接展示 openai/codex，并将 Codex 协议描述为开放 Agent 事实标准；官方仓库证实其为本地终端编码智能体。", "tags": [ "抖音推荐", "OpenAI", "Codex", "AI 编程", "Coding Agent", "Rust", "终端", "SDK", "高权限", "Apache-2.0" ], "language": "Rust / Python / Starlark", "stars": 101446, "forks": 15208, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音视频逐帧、GitHub 官方仓库 API、语言统计与 README 已核验", "metadataVerifiedAt": "2026-07-26T02:17:55.827Z", "source": "抖音《GitHub 本周热榜来了！开发者集体在卷 AI Agent》", "video": "https://www.douyin.com/video/7664073102514523392", "icon": "code", "color": "#10a37f", "createdAt": "2026-07-23T12:56:13+00:00", "updatedAt": "2026-07-26T02:15:32Z" }, { "id": "github-com-headroomlabs-ai-headroom", "titleZh": "", "name": "headroom", "repo": "https://github.com/headroomlabs-ai/headroom", "homepage": "https://headroom-docs.vercel.app/docs", "summary": "面向 AI Agent 的上下文压缩层，在工具输出、日志、RAG、文件和对话进入模型前去除冗余内容。", "details": "Headroom 可作为 CLI 包装器、OpenAI 兼容代理、Python/TypeScript SDK、MCP 服务或 Docker 服务接入 Claude Code、Codex、Cursor、Aider、Copilot CLI 等 Agent。官方基准中的“最高节省 92% Token”对应代码搜索和 SRE 故障排查等特定样例，并不是所有任务都固定节省 92%；实际收益取决于输入结构、压缩策略、模型缓存和任务容错要求。压缩会在成本、延迟与信息保真度之间取舍，正式工作流应先对自己的数据做准确率回归，并保留原始上下文以便审计和恢复。", "why": "视频画面同时出现 headroom-ai 的 pip、pipx 与 npm 安装命令，标题中的 92% 也与官方基准一致，因此可以唯一确认是 Headroom，而不是同类的 RTK 终端输出压缩工具。", "tags": [ "抖音推荐", "Token 优化", "上下文压缩", "AI Agent", "Codex", "Claude Code", "Cursor", "MCP", "代理", "RAG", "本地优先", "基准非保证" ], "language": "Python / Rust / TypeScript", "stars": 62421, "forks": 4717, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音视频画面、GitHub 官方仓库 API、语言统计与 README 已核验", "metadataVerifiedAt": "2026-07-26T02:17:56.064Z", "source": "douyin", "video": "https://www.douyin.com/video/7665622044674903336", "icon": "spark", "color": "#7c3aed", "createdAt": "2026-07-23T12:29:47+00:00", "updatedAt": "2026-07-26T02:17:18Z" }, { "id": "github-com-penpot-penpot", "titleZh": "", "name": "penpot", "repo": "https://github.com/penpot/penpot", "homepage": "https://penpot.app", "summary": "面向设计与开发协作的开源设计平台，支持实时协作、原型、设计系统与自托管。", "details": "Penpot 以开放 Web 标准和 SVG 为核心，提供界面设计、交互原型、组件/设计系统、团队实时协作以及面向开发者的检查和代码信息，可用官方云服务或自行部署。它能改善设计交付，但生成的样式和布局仍需要工程人员审查，并不等同于把完整产品一键转换成生产代码。", "why": "视频章节虽然被识别为“PMPOT”，但开源设计、实时协作和打通设计到开发交付的描述明确对应 Penpot。", "tags": [ "抖音推荐", "开源设计", "UI UX", "实时协作", "设计系统", "SVG", "自托管", "设计交付", "clojure", "clojurescript", "design", "prototyping" ], "language": "Clojure / JavaScript / Rust", "stars": 57351, "forks": 3798, "license": "MPL-2.0", "install": "", "compatibility": "", "evidence": "抖音视频章节、GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-26T02:17:56.325Z", "source": "douyin", "video": "https://www.douyin.com/video/7656321732915039529", "icon": "code", "color": "#5b9df6", "createdAt": "2026-07-23T06:13:17+00:00", "updatedAt": "2026-07-25T23:13:32Z" }, { "id": "github-com-jamiepine-voicebox", "titleZh": "", "name": "voicebox", "repo": "https://github.com/jamiepine/voicebox", "homepage": "https://voicebox.sh", "summary": "本地优先的开源 AI 语音工作室，覆盖声音克隆、多引擎 TTS、听写与 MCP 接入。", "details": "Voicebox 把多种开源语音模型封装进桌面工作流，支持零样本声音克隆、多语言文本转语音、语音转写、模型管理和 MCP 调用，并强调本地处理与隐私。不同模型对 GPU、显存和下载空间要求不同；克隆或生成他人声音前必须取得授权，并为合成内容提供清晰标识。", "why": "视频明确展示 Voicebox，并给出开源语音工作室、多语言、零样本声音克隆等可核对特征。", "tags": [ "抖音推荐", "AI 语音", "声音克隆", "TTS", "听写", "MCP", "本地优先", "隐私", "声音授权", "ai", "cuda", "mlx" ], "language": "TypeScript / Python / Rust", "stars": 46747, "forks": 5727, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音视频章节、GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-26T02:17:56.571Z", "source": "douyin", "video": "https://www.douyin.com/video/7656321732915039529", "icon": "spark", "color": "#ec5ba7", "createdAt": "2026-07-23T06:13:17+00:00", "updatedAt": "2026-07-26T02:17:27Z" }, { "id": "github-com-topoteretes-cognee", "titleZh": "", "name": "cognee", "repo": "https://github.com/topoteretes/cognee", "homepage": "https://www.cognee.ai", "summary": "用知识图谱与向量检索为 AI Agent 建立可持久化、可关联查询的长期记忆。", "details": "Cognee 将文档、对话和业务数据经过摄取、分块、实体关系抽取后写入图结构与向量存储，再通过语义检索和图遍历帮助智能体跨会话回忆事实及其关系。它适合知识助手、研究代理和多 Agent 系统，但效果取决于嵌入/大模型、图数据库配置与数据治理，部署前应评估推理费用、隐私和删除策略。", "why": "视频章节中的“cogny”是语音识别误差；长期记忆、图谱构建和语义检索三个特征唯一对应 Cognee。", "tags": [ "抖音推荐", "AI Agent", "长期记忆", "知识图谱", "向量检索", "RAG", "语义检索", "数据治理", "agent-memory", "agent-skills", "ai", "ai-agents" ], "language": "Python / TypeScript / JavaScript", "stars": 29331, "forks": 2809, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "抖音视频章节、GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-26T02:17:56.828Z", "source": "douyin", "video": "https://www.douyin.com/video/7656321732915039529", "icon": "database", "color": "#8b7cff", "createdAt": "2026-07-23T06:13:17+00:00", "updatedAt": "2026-07-26T02:01:10Z" }, { "id": "github-com-panniantong-agent-reach", "titleZh": "", "name": "Agent-Reach", "repo": "https://github.com/Panniantong/Agent-Reach", "homepage": "", "summary": "用一个 CLI 让 AI Agent 读取和检索 X、Reddit、YouTube、GitHub、B站与小红书等平台。", "details": "Agent Reach 把多个现成命令行工具、网页读取能力和登录态整合成统一入口，并在直接读取失败时尝试备用方案，适合为编码或研究智能体补齐跨平台信息获取能力。所谓“零 API 费用”是指不依赖各平台付费 API，并不等于没有账号、Cookie、频率限制和平台条款；启用登录态前应检查凭据保存方式并遵守各平台规则。", "why": "视频章节将名称识别为“agent rich”，但画面功能描述与 Agent Reach 的多平台读取、搜索和无付费 API 定位逐项吻合。", "tags": [ "抖音推荐", "AI Agent", "联网检索", "多平台", "CLI", "GitHub", "B站", "小红书", "平台条款", "agent-infrastructure", "ai-agent", "ai-search" ], "language": "Python / Shell / HTML", "stars": 60839, "forks": 4923, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音视频章节、GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-26T02:17:57.052Z", "source": "douyin", "video": "https://www.douyin.com/video/7656321732915039529", "icon": "globe", "color": "#20b8a6", "createdAt": "2026-07-23T06:13:17+00:00", "updatedAt": "2026-07-26T02:00:38Z" }, { "id": "github-com-oraios-serena", "titleZh": "", "name": "Serena", "repo": "https://github.com/oraios/serena", "homepage": "https://oraios.github.io/serena", "summary": "给 Codex、Claude Code 等编码智能体提供符号级代码检索、编辑与重构能力的开源 MCP 工具。", "details": "Serena 通过语言服务器或 JetBrains 后端理解符号、引用、声明和实现，使智能体只读取完成任务所需的代码结构，而不是反复吞入整份文件；同时提供安全重命名、符号体替换、引用查找、项目记忆和可配置工具集。它能显著降低大型代码库的上下文浪费，但并不保证固定比例的 Token 节省，实际效果取决于项目规模、语言服务器和智能体是否正确调用语义工具。", "why": "抖音视频画面、搜索词和作者评论均明确指向 Serena；它解决的是输入侧代码上下文浪费，和已收录的 RTK 终端输出压缩属于互补方案。", "tags": [ "抖音推荐", "AI 编程", "MCP", "Codex", "Claude Code", "语义检索", "符号级编辑", "LSP", "Token 优化", "MIT", "agent", "ai" ], "language": "Python / JavaScript / CSS", "stars": 26809, "forks": 1775, "license": "MIT", "install": "", "compatibility": "", "evidence": "抖音视频页面、作者评论、GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-24T02:18:09.112Z", "source": "douyin", "video": "https://www.douyin.com/video/7665140840184007988", "icon": "code", "color": "#5b9df6", "createdAt": "2026-07-23T06:02:51+00:00", "updatedAt": "2026-07-24T02:13:37Z" }, { "id": "github-com-a16z-infra-ai-town", "titleZh": "", "name": "AI Town", "repo": "https://github.com/a16z-infra/ai-town", "homepage": "https://convex.dev/ai-town", "summary": "可部署、可扩展的 AI 虚拟小镇，让角色在持续运行的像素世界中生活、移动、聊天和建立关系。", "details": "项目受 Generative Agents 论文启发，提供共享世界状态、事务、模拟引擎、向量检索和多人访问，可连接 Ollama 或 OpenAI 兼容模型。它更擅长承载地点、时间、相遇和社会事件，角色的深层心理与长期记忆仍建议接入独立认知层。", "why": "最适合作为“苏晚”数字生命的生活世界：角色不只停留在聊天框里，而是拥有位置、日程、相遇和会改变计划的环境事件。", "tags": [ "数字生命", "AI Town", "生活模拟", "多智能体", "像素世界", "Ollama", "TypeScript", "MIT" ], "language": "TypeScript / JavaScript / HTML", "stars": 10202, "forks": 1133, "license": "MIT", "install": "", "compatibility": "", "evidence": "GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-24T02:18:09.665Z", "source": "web-research", "video": "", "icon": "globe", "color": "#43b9b1", "createdAt": "2026-07-22T22:59:47+00:00", "updatedAt": "2026-07-23T15:39:44Z" }, { "id": "github-com-joonspk-research-generative-agents", "titleZh": "", "name": "Generative Agents", "repo": "https://github.com/joonspk-research/generative_agents", "homepage": "", "summary": "斯坦福生成式智能体研究原型，通过记忆流、反思、计划和行动模拟可信的人类日常行为。", "details": "角色会记录完整经历，按重要性、相关性和时间检索记忆，定期把经历归纳为高层反思，并生成日程、响应新事件。仓库偏研究复现，运行环境和模型接口较旧，适合作为认知架构参考而不是直接当生产聊天产品。", "why": "它最接近“苏晚为什么这样思考”的底层大脑，可提供三层时间结构所需要的即时反应、日程计划和跨事件反思。", "tags": [ "数字生命", "生成式智能体", "长期记忆", "反思", "日程规划", "社会模拟", "研究原型", "Apache-2.0" ], "language": "Markdown / JSON / HTML / Python", "stars": 21791, "forks": 3057, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-24T02:18:11.230Z", "source": "web-research", "video": "", "icon": "bot", "color": "#8b7cff", "createdAt": "2026-07-22T22:59:47+00:00", "updatedAt": "2026-07-24T01:25:18Z" }, { "id": "github-com-letta-ai-characterai-memory", "titleZh": "", "name": "CharacterPlus / Letta", "repo": "https://github.com/letta-ai/characterai-memory", "homepage": "https://characterplus.vercel.app/", "summary": "基于 Letta 的有状态角色聊天应用，让不同角色保持身份、长期记忆和跨会话关系。", "details": "项目展示了角色身份块、持久记忆、共享用户资料、工具调用及 REST API 接入，适合快速搭建一个不会每次重置的陪伴角色。它本身没有持续生活世界，用户不发消息时通常不会自动产生完整人生轨迹。", "why": "适合作为“苏晚”的长期关系与对话记忆层，先低成本做出可交流原型，再连接自主日程和虚拟世界。", "tags": [ "数字生命", "角色陪伴", "Letta", "持久记忆", "有状态Agent", "TypeScript", "长期关系", "MIT" ], "language": "TypeScript / CSS / JavaScript", "stars": 19, "forks": 5, "license": "MIT", "install": "", "compatibility": "", "evidence": "GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-24T02:18:11.859Z", "source": "web-research", "video": "", "icon": "database", "color": "#5b9df6", "createdAt": "2026-07-22T22:59:47+00:00", "updatedAt": "2026-06-30T16:51:16Z" }, { "id": "github-com-lynpoint-cyberverse", "titleZh": "", "name": "CyberVerse", "repo": "https://github.com/Lynpoint/CyberVerse", "homepage": "https://www.cyberverse.cc", "summary": "自托管实时数字人平台，通过 WebRTC、语音、人格记忆、工具和 RAG 构建可面对面交流的 AI 角色。", "details": "CyberVerse 重点解决数字生命的表现层：实时音视频、语音优先交互、人物设定、记忆和工具连接。原检索地址 dsd2077/CyberVerse 已不可访问，GitHub 官方搜索确认当前公开仓库为 Lynpoint/CyberVerse。", "why": "适合给“苏晚”增加可见形象、语音和视频通话能力，但自主生活与深层认知仍需要组合其他项目。", "tags": [ "数字生命", "数字人", "WebRTC", "实时语音", "视频通话", "RAG", "自托管", "虚拟形象", "ai-agents", "ai-companion", "digital-human", "digital-life" ], "language": "Python / Go / Vue", "stars": 1464, "forks": 199, "license": "GPL-3.0", "install": "", "compatibility": "", "evidence": "GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-24T02:18:12.449Z", "source": "web-research", "video": "", "icon": "spark", "color": "#ec5ba7", "createdAt": "2026-07-22T22:59:47+00:00", "updatedAt": "2026-07-23T17:45:53Z" }, { "id": "github-com-elophanto-elophanto", "titleZh": "", "name": "EloPhanto", "repo": "https://github.com/elophanto/EloPhanto", "homepage": "https://elophanto.com", "summary": "具有身份、人格、情绪、记忆和自主日程的实验性数字生命，可持续运行并自行使用工具完成目标。", "details": "项目允许智能体建立身份与价值观、进行反思、积累知识、安排自主任务并按需创建专业分身，也能接触邮箱、账号或钱包等外部系统。其自治权限高且仓库目前未声明 SPDX 许可证，应在沙箱内测试，禁止直接授予生产凭据。", "why": "它最接近“用户不找她，她也在继续生活和做事”的目标，可参考自主唤醒、人格成长和长期运行机制。", "tags": [ "数字生命", "自主Agent", "人格成长", "情绪", "长期记忆", "自主日程", "高权限", "许可证待核验", "agent", "agentic-ai", "ai", "ai-assistant" ], "language": "Python / TypeScript / Shell", "stars": 115, "forks": 9, "license": "NOASSERTION", "install": "", "compatibility": "", "evidence": "GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-24T02:18:13.013Z", "source": "web-research", "video": "", "icon": "rocket", "color": "#ff6759", "createdAt": "2026-07-22T22:59:47+00:00", "updatedAt": "2026-07-23T07:46:51Z" }, { "id": "github-com-camel-ai-oasis", "titleZh": "", "name": "OASIS", "repo": "https://github.com/camel-ai/oasis", "homepage": "https://docs.oasis.camel-ai.org/", "summary": "可扩展到大规模智能体的开源社会模拟平台，复现发帖、评论、关注、推荐和群体行为。", "details": "OASIS 面向 Twitter、Reddit 式社会环境，支持多种社交动作、动态网络、推荐算法和最多百万级角色研究。它不适合作为单个陪伴角色的第一层，但能让数字生命进入复杂社会并观察关系和信息传播。", "why": "适合“苏晚”后期进入由大量 AI 人物组成的社会，研究群体关系、舆论、从众和长期社会演化。", "tags": [ "数字生命", "社会模拟", "百万智能体", "多智能体", "社交网络", "群体行为", "Python", "Apache-2.0", "agent-based-framework", "agent-based-simulation", "ai-societies", "deep-learning" ], "language": "Python / Dockerfile", "stars": 4959, "forks": 610, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "GitHub 官方仓库 API 与 README 已核验", "metadataVerifiedAt": "2026-07-24T02:18:13.580Z", "source": "web-research", "video": "", "icon": "globe", "color": "#d59b43", "createdAt": "2026-07-22T22:59:47+00:00", "updatedAt": "2026-07-23T20:23:29Z" }, { "id": "github-com-rtk-ai-rtk", "titleZh": "", "name": "RTK（Rust Token Killer）", "repo": "https://github.com/rtk-ai/rtk", "homepage": "https://www.rtk-ai.app", "summary": "在终端输出进入 AI 上下文前进行命令感知压缩，减少常见开发命令 60%–90% Token 消耗的 Rust CLI。", "details": "RTK 为 Git、测试、构建、搜索等命令生成更紧凑的模型输入，并保留失败摘要和统计。过滤输出可能改变依赖空输出或机器可读格式的脚本语义，因此自动 Hook 前要审查绕过规则，精确验收仍应读取原始输出。", "why": "视频逐项展示 rtk git status、rtk git diff 和 rtk test，与 rtk-ai/rtk 官方命令和节省比例完全一致。", "tags": [ "抖音推荐", "AI 编程", "Token 优化", "终端输出压缩", "Rust", "CLI", "Claude Code", "Codex", "语义风险", "Apache-2.0", "agentic-coding", "ai-coding" ], "language": "Rust / Shell / TypeScript", "stars": 72866, "forks": 4546, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-24T02:18:14.141Z", "source": "douyin", "video": "https://v.douyin.com/kuBI9TtL_RA/", "icon": "terminal", "color": "#f97316", "createdAt": "2026-07-22T22:29:54+00:00", "updatedAt": "2026-07-24T02:12:37Z" }, { "id": "github-com-tashfeenahmed-freellmapi", "name": "FreeLLMAPI", "repo": "https://github.com/tashfeenahmed/freellmapi", "homepage": "https://freellmapi.co", "summary": "把多家 LLM 免费额度整合为单一 OpenAI 兼容接口，并提供路由、故障切换和加密密钥管理。", "details": "当前官方仓库已扩展到约 28 家提供商和约 40 亿 Token/月的理论免费额度。额度、模型和条款随供应商变化，项目明确面向个人实验；不能把理论总量理解为稳定生产 SLA，也不应转售或共享上游账号。", "why": "视频所说的“16 家、每月 17 亿 Token”与 FreeLLMAPI 的早期版本描述完全一致；GitHub 官方元数据已显示项目继续扩展。", "tags": [ "抖音推荐", "OpenAI 兼容", "LLM 路由", "免费额度", "自动故障切换", "TypeScript", "个人实验", "服务条款需核验", "MIT" ], "language": "TypeScript / HTML / JavaScript", "stars": 16690, "forks": 2420, "license": "MIT", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:05.914Z", "source": "douyin", "video": "https://v.douyin.com/lCoMdn9qlxw/", "icon": "spark", "color": "#8b5cf6", "createdAt": "2026-07-22T22:29:54+00:00", "updatedAt": "2026-07-23T02:16:36Z" }, { "id": "github-com-soxoj-maigret", "name": "Maigret", "repo": "https://github.com/soxoj/maigret", "homepage": "https://maigret.app", "summary": "根据用户名在 3000 多个网站检索公开账号并生成 OSINT 调查报告的 Python 工具。", "details": "Maigret 支持命令行、网页界面和结构化报告，可用于经授权的安全调查、账号自查和公开情报研究。用户名匹配可能产生同名误报，也涉及隐私与平台条款，必须限定合法用途并人工复核。", "why": "“2 万+星标”视频页面的关联搜索直接显示 maigret项目地址，视频内容也展示了按用户名跨平台搜索的功能。", "tags": [ "抖音推荐", "OSINT", "用户名搜索", "公开情报", "Python", "隐私敏感", "仅限合法调查", "MIT", "cli", "cybersecurity", "identification", "information-gathering" ], "language": "Python / HTML / Go Template", "stars": 35681, "forks": 2728, "license": "MIT", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:06.135Z", "source": "douyin", "video": "https://v.douyin.com/nIa9olfwswc/", "icon": "shield", "color": "#ef4444", "createdAt": "2026-07-22T22:29:54+00:00", "updatedAt": "2026-07-23T02:03:28Z" }, { "id": "github-com-codecrafters-io-build-your-own-x", "name": "Build Your Own X", "repo": "https://github.com/codecrafters-io/build-your-own-x", "homepage": "https://codecrafters.io", "summary": "通过从零复刻数据库、操作系统、编译器、游戏和网络工具，系统掌握软件工程原理的超大型教程索引。", "details": "仓库按技术类别整理大量公开教程，核心内容是 Markdown 链接清单，不是可直接安装的 Agent Skill。适合用项目制学习理解底层实现；每篇外部教程的许可证和质量需要分别核验。", "why": "“AI 时代想造专属软件”和“教会你开发游戏”两段视频画面都展示了 CodeCrafters 的 Build Your Own X 页面及其约 52 万 Stars 数据。", "tags": [ "抖音推荐", "编程学习", "从零实现", "教程合集", "游戏开发", "底层原理", "Markdown", "许可证需逐项核验", "awesome-list", "free", "programming", "tutorial-code" ], "language": "Markdown", "stars": 530226, "forks": 50157, "license": "", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:06.338Z", "source": "douyin", "video": "https://v.douyin.com/mW30ItxMZ_0/", "icon": "code", "color": "#2563eb", "createdAt": "2026-07-22T22:29:54+00:00", "updatedAt": "2026-07-23T02:08:49Z" }, { "id": "github-com-nvlabs-eagle", "name": "NVIDIA Eagle / LocateAnything-3B", "repo": "https://github.com/NVlabs/Eagle", "homepage": "https://huggingface.co/nvidia/LocateAnything-3B", "summary": "NVIDIA LocateAnything-3B 所在的官方 Eagle 视觉语言模型仓库，支持文本指向定位、密集检测、GUI 元素定位与 OCR。", "details": "LocateAnything 位于仓库 Embodied 目录，采用并行框解码一次预测完整坐标，在视觉定位任务上兼顾速度与几何一致性，并公开推理、评测和微调代码。模型权重采用 NVIDIA 许可且仅限研究与非商业使用，不能把代码的 Apache-2.0 误当成模型权重许可。", "why": "“英伟达又开源了重磅项目”视频搜索词明确为 LocateAnything-3B；官方 Hugging Face 模型卡直接指向 NVlabs/Eagle/Embodied 代码。", "tags": [ "抖音推荐", "NVIDIA", "LocateAnything-3B", "视觉语言模型", "目标检测", "GUI Grounding", "Python", "模型非商业许可", "Apache-2.0", "demo", "eagle", "gpt4" ], "language": "Python / C / Shell", "stars": 3235, "forks": 314, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:06.542Z", "source": "douyin", "video": "https://v.douyin.com/-84FEIxvYM0/", "icon": "spark", "color": "#48ae78", "createdAt": "2026-07-22T22:17:13+00:00", "updatedAt": "2026-07-22T23:40:33Z" }, { "id": "github-com-heygen-com-hyperframes", "name": "HyperFrames", "repo": "https://github.com/heygen-com/hyperframes", "homepage": "https://hyperframes.heygen.com", "summary": "HeyGen 开源、为智能体设计的 HTML 原生确定性视频制作与渲染框架。", "details": "用 HTML、CSS、GSAP、媒体文件和可寻址时间轴生成 MP4，提供 CLI、预览、检查、渲染、素材处理和配套 Agent Skills，适合产品宣传、字幕、动态图表与自动化视频流水线。", "why": "“三个 Codex 神级 Skill”视频第二项明确展示 HyperFrames；官方仓库同时是完整视频框架和 Skills 载体。", "tags": [ "抖音推荐", "视频", "HTML", "GSAP", "FFmpeg", "Agent Skills", "TypeScript", "Apache-2.0", "ai", "animation", "ffmpeg", "framework" ], "language": "TypeScript / JavaScript / Shell", "stars": 36936, "forks": 3481, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:06.783Z", "source": "manual", "video": "https://v.douyin.com/_fgGyfGBJ3c/", "icon": "rocket", "color": "#ff6759", "createdAt": "2026-07-22T22:11:54.511Z", "updatedAt": "2026-07-23T02:15:16Z" }, { "id": "github-com-lingfengqaq-webnovel-writer", "name": "Webnovel Writer", "repo": "https://github.com/lingfengQAQ/webnovel-writer", "homepage": "", "summary": "基于 Claude Code、面向 200 万字级长篇连载的网文辅助创作系统。", "details": "用 RAG、角色和世界观状态、规划/写作/审查 Agent 以及只读可视化面板降低长篇小说中的遗忘、设定冲突和幻觉，支持多题材模板与持续复盘。GPL-3.0 对分发修改版本有传染性要求。", "why": "视频明确描述 200 万字长篇网文创作，与 Webnovel Writer 的长期记忆和章节管理目标完全一致。", "tags": [ "抖音推荐", "网文", "长篇小说", "RAG", "Claude Code", "Agent Skills", "Python", "GPL-3.0" ], "language": "Python / JavaScript / CSS", "stars": 6074, "forks": 1052, "license": "GPL-3.0", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:07.003Z", "source": "manual", "video": "https://v.douyin.com/q03MZ3tZn50/", "icon": "database", "color": "#5b9df6", "createdAt": "2026-07-22T22:11:54.511Z", "updatedAt": "2026-07-23T01:07:33Z" }, { "id": "github-com-htdt-godogen", "name": "Godogen", "repo": "https://github.com/htdt/godogen", "homepage": "https://x.com/alex_erm", "summary": "让 Claude Code 或 Codex 自主规划、编写、运行、观察并修复 Godot、Bevy 与 Babylon.js 游戏。", "details": "项目发布一组游戏生成 Skills 到新仓库：Agent 会生成代码与素材、启动引擎、检查截图，再依据可见缺陷自我修复；支持 Godot C#、Rust/Bevy、TypeScript/Babylon.js 和 Android 调试导出。部分素材生成需要第三方付费 API。", "why": "“让你学会开发游戏”视频帧中的仓库简介、三种引擎与当时 4.2k Stars 均精确匹配 Godogen。", "tags": [ "抖音推荐", "游戏开发", "Godot", "Bevy", "Babylon.js", "Codex Skill", "Claude Code Skill", "MIT", "claude", "claude-code", "claude-code-skill", "codex" ], "language": "Python / Shell", "stars": 4900, "forks": 438, "license": "MIT", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:07.211Z", "source": "manual", "video": "https://v.douyin.com/tkUa-6MqCV4/", "icon": "bot", "color": "#43b9b1", "createdAt": "2026-07-22T22:11:54.511Z", "updatedAt": "2026-07-23T02:07:42Z" }, { "id": "github-com-f-prompts-chat", "name": "Prompts.chat", "repo": "https://github.com/f/prompts.chat", "homepage": "https://prompts.chat", "summary": "原 Awesome ChatGPT Prompts 演进而来的大型开源提示词分享、发现与自托管平台。", "details": "收录并组织社区提示词，支持搜索、收藏与组织内自托管，兼容 ChatGPT、Claude、Gemini 等模型。仓库未声明标准 SPDX 许可证，复制提示词和部署前要核验具体授权。", "why": "“全球最大最具影响力的提示词宝藏库”对应原 f/awesome-chatgpt-prompts；该仓库现已重定向并更名为 f/prompts.chat。", "tags": [ "抖音推荐", "提示词", "Prompt Engineering", "开源清单", "自托管", "许可证待核验", "ai", "artificial-intelligence", "awesome-list", "chatgpt", "chatgpt-prompts", "claude" ], "language": "HTML / MDX / TypeScript", "stars": 166218, "forks": 21482, "license": "NOASSERTION", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:07.476Z", "source": "manual", "video": "https://v.douyin.com/4W4N7ey2Ulw/", "icon": "spark", "color": "#ff6759", "createdAt": "2026-07-22T22:11:54.511Z", "updatedAt": "2026-07-23T01:46:42Z" }, { "id": "github-com-project-n-e-k-o-n-e-k-o", "name": "N.E.K.O（苏晚相似替代）", "repo": "https://github.com/Project-N-E-K-O/N.E.K.O", "homepage": "https://store.steampowered.com/app/4099310", "summary": "会主动联系用户、拥有情绪与记忆并能使用工具的实时 AI 陪伴平台；是“苏晚虚拟生命引擎”的公开相似替代，不是其原仓。", "details": "项目以具身情绪引擎、主动行为、长期记忆、数字人形态和工具执行构建持续存在的 AI 伙伴，适合作为训练类人陪伴智能体的技术参考。涉及情感依赖、隐私数据和主动行为时必须设置透明边界与安全开关。", "why": "“苏晚”视频展示的虚拟生命引擎尚未找到公开仓库；N.E.K.O 是功能方向最接近且公开可核验的替代方案，因此显著标注为相似替代。", "tags": [ "相似替代", "苏晚", "AI 陪伴", "数字生命", "主动 Agent", "长期记忆", "Python", "Apache-2.0", "ai-agents", "companion", "digital-human", "jarvis-assistant" ], "language": "Python / JavaScript / TypeScript", "stars": 2201, "forks": 248, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:07.941Z", "source": "manual", "video": "https://v.douyin.com/2ZF_GA6IBo8/", "icon": "spark", "color": "#d59b43", "createdAt": "2026-07-22T22:11:54.511Z", "updatedAt": "2026-07-23T02:11:17Z" }, { "id": "github-com-public-apis-public-apis", "name": "Public APIs", "repo": "https://github.com/public-apis/public-apis", "homepage": "https://publicapis.dev", "summary": "社区维护的免费公共 API 分类清单，覆盖天气、资讯、图片、金融、开发工具等大量类别。", "details": "以表格形式记录接口用途、认证方式、HTTPS 与 CORS 情况，适合项目原型和接口调研。列表收录不等于长期可用或安全背书，接入前仍需检查服务条款、配额和隐私政策。", "why": "视频画面地址栏直接显示 public-apis/public-apis，是原仓。", "tags": [ "抖音推荐", "API", "开发资源", "免费接口", "开源清单", "MIT", "api", "apis", "dataset", "development", "free", "list" ], "language": "Python / Shell", "stars": 452025, "forks": 49760, "license": "MIT", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:08.172Z", "source": "manual", "video": "https://v.douyin.com/ySDYKJCvUL0/", "icon": "globe", "color": "#5b9df6", "createdAt": "2026-07-22T22:11:54.511Z", "updatedAt": "2026-07-23T02:17:31Z" }, { "id": "github-com-nousresearch-hermes-agent", "name": "Hermes Agent", "repo": "https://github.com/NousResearch/hermes-agent", "homepage": "https://hermes-agent.nousresearch.com", "summary": "Nous Research 开源、能持续学习用户偏好并通过 Skills 扩展能力的个人 AI Agent。", "details": "支持终端、浏览器、消息网关、定时任务、记忆与技能系统，可接入多种模型和 Telegram 等渠道。部署时要隔离高权限工具、保护 API 密钥，并审查第三方 Skills。", "why": "“让你拥有真人 AI Agent”视频搜索词直接显示 Hermes 安装，仓库与我们当前使用的 Hermes 完全一致。", "tags": [ "抖音推荐", "Hermes", "AI Agent", "长期记忆", "Skills", "Telegram", "Python", "MIT", "ai", "ai-agent", "ai-agents", "anthropic" ], "language": "Python / TypeScript / TeX", "stars": 218997, "forks": 41502, "license": "MIT", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:08.451Z", "source": "manual", "video": "https://v.douyin.com/xlvR6Jb2ink/", "icon": "bot", "color": "#ec5ba7", "createdAt": "2026-07-22T22:11:54.511Z", "updatedAt": "2026-07-23T02:15:53Z" }, { "id": "github-com-pixel-agents-hq-pixel-agents", "name": "Pixel Agents", "repo": "https://github.com/pixel-agents-hq/pixel-agents", "homepage": "", "summary": "把 Claude Code 等编码智能体显示成像素办公室员工的 VS Code 扩展。", "details": "在像素办公室中把 Agent 会话、任务与状态可视化，让多个编码智能体像《模拟人生》里的员工一样工作；核心价值是工作状态观察与趣味化协作界面，不应误认为真正的任务编排后端。", "why": "视频明确描述“用模拟人生管理 AI、把 Claude Code 变成像素办公室”，与官方仓库一致。", "tags": [ "抖音推荐", "像素办公室", "VS Code", "Claude Code", "AI Agent", "TypeScript", "MIT" ], "language": "TypeScript / HTML / JavaScript", "stars": 8693, "forks": 1396, "license": "MIT", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:08.650Z", "source": "manual", "video": "https://v.douyin.com/oLYU6ZCz0TI/", "icon": "bot", "color": "#8b7cff", "createdAt": "2026-07-22T22:11:54.511Z", "updatedAt": "2026-07-23T01:36:48Z" }, { "id": "github-com-robbyant-lingbot-map", "name": "LingBot-Map", "repo": "https://github.com/Robbyant/lingbot-map", "homepage": "https://technology.robbyant.com/lingbot-map", "summary": "蚂蚁灵波开源的实时流式三维重建基础模型，仅用普通 RGB 摄像头即可持续估计位姿、深度和三维点云。", "details": "项目以几何上下文 Transformer 和纯自回归建模逐帧处理视频，在不依赖未来帧的情况下完成相机定位与场景重建，面向机器人、自动驾驶、AR/VR 与空间感知研究。实际部署需要 GPU，并应核验模型权重许可。", "why": "视频标题、发布时间和“普通相机流式 3D 重建”描述均与蚂蚁灵波官方仓库一致。", "tags": [ "抖音推荐", "3D 重建", "计算机视觉", "机器人", "流式推理", "Python", "Apache-2.0" ], "language": "Python / JavaScript / Cuda", "stars": 14919, "forks": 1545, "license": "Apache-2.0", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:08.870Z", "source": "manual", "video": "https://v.douyin.com/i_bBsdpDn08/", "icon": "spark", "color": "#43b9b1", "createdAt": "2026-07-22T22:11:54.511Z", "updatedAt": "2026-07-23T02:14:08Z" }, { "id": "github-com-santifer-career-ops", "name": "career-ops", "repo": "https://github.com/santifer/career-ops", "homepage": "https://career-ops.org", "summary": "运行于AI编码CLI中的开源求职评估、简历定制与申请跟踪工具。", "details": "解决职位筛选和求职材料管理效率问题，可扫描招聘门户，以A至F结构化量表生成1.0至5.0评分，并辅助定制简历、填写申请表和跟踪进度。适合主动求职者；默认不代替用户自动提交，事实须人工复核并遵守网站条款。主要语言：JavaScript、TypeScript、Go；61,075 Stars；MIT许可证。", "why": "覆盖岗位评估、材料定制与申请管理的主要流程，同时保留人工决策环节，适合希望在本地AI编码CLI中建立可控求职工作流的用户。抖音来源：https://v.douyin.com/7Sb6WAA36ic/", "tags": [ "抖音推荐", "AI求职", "岗位评估", "简历定制", "申请跟踪", "本地运行", "MIT", "ai", "ai-agent", "anthropic", "ats", "automation" ], "language": "JavaScript / TypeScript / Go", "stars": 61108, "forks": 12033, "license": "MIT", "install": "", "compatibility": "", "evidence": "", "metadataVerifiedAt": "2026-07-23T02:18:09.062Z", "source": "ai-workspace", "video": "", "icon": "bot", "color": "", "createdAt": "2026-07-22T21:38:45.326Z", "updatedAt": "2026-07-23T01:57:32Z" }, { "id": "github-com-antirez-ds4", "name": "ds4", "repo": "https://github.com/antirez/ds4", "homepage": "", "summary": "面向DeepSeek 4 Flash与PRO的本地高性能推理引擎。", "details": "解决DeepSeek 4 Flash及PRO模型的本地推理问题，支持Metal、CUDA和ROCm计算后端，适合具备高性能硬件且需要本地部署的开发者与研究人员。其硬件及内存门槛较高，并非通用模型推理框架。主要语言：C、Cuda、Objective-C；19,075 Stars；MIT许可证。", "why": "项目针对特定DeepSeek模型和本地加速后端，适合明确追求本地推理的技术用户；部署前应核对模型、显存、内存及平台兼容性。抖音来源：https://v.douyin.com/pO5ge5xztNo/", "tags": [ "抖音推荐", "DeepSeek", "本地推理", "CUDA", "Metal", "ROCm", "MIT", "C" ], "language": "C / Cuda / Objective-C", "stars": 19075, "forks": 1673, "license": "MIT", "metadataVerifiedAt": "2026-07-22T21:39:05.753Z", "source": "ai-workspace", "video": "", "icon": "server", "color": "", "createdAt": "2026-07-22T21:38:45.326Z", "updatedAt": "2026-07-22T21:34:47Z" }, { "id": "github-com-nextlevelbuilder-ui-ux-pro-max-skill", "name": "ui-ux-pro-max-skill", "repo": "https://github.com/nextlevelbuilder/ui-ux-pro-max-skill", "homepage": "https://www.uupm.cc/", "summary": "面向多平台界面开发的AI设计规则与检索知识库。", "details": "解决AI生成界面时缺乏系统设计规范的问题，提供可检索的UI/UX设计知识与开发辅助能力，适合产品设计、前端开发及AI编码工具用户。它不能替代专业设计师，关键体验仍需人工验证。主要语言：Python、JavaScript、HTML；108,941 Stars；MIT许可证。", "why": "规则与知识检索定位清晰，适合在Claude Code、Codex、Cursor等工作流中为界面设计提供参考，但不应将输出视为专业设计结论。抖音来源：https://v.douyin.com/ok9fN33MJAo/", "tags": [ "抖音推荐", "UI/UX", "AI技能", "设计知识库", "前端开发", "命令行", "MIT", "ai-skills", "antigravity", "claude", "claude-code", "codex" ], "language": "Python / JavaScript / HTML", "stars": 108941, "forks": 11597, "license": "MIT", "metadataVerifiedAt": "2026-07-22T21:39:05.799Z", "source": "ai-workspace", "video": "", "icon": "spark", "color": "", "createdAt": "2026-07-22T21:38:45.326Z", "updatedAt": "2026-07-22T21:37:52Z" }, { "id": "github-com-digitalplatdev-freedomain", "name": "FreeDomain", "repo": "https://github.com/DigitalPlatDev/FreeDomain", "homepage": "https://domain.digitalplat.org", "summary": "面向个人用户提供免费域名申请与使用入口的开放项目。", "details": "解决个人项目、学习实验和演示站点获取域名的成本问题，为用户提供免费域名相关服务，适合学生、个人开发者、原型验证及非关键展示场景。免费域名的持续可用性、规则和服务稳定性可能变化，正式品牌、商业业务及关键基础设施应评估风险并准备迁移方案。主要语言：Markdown；Stars：187692；许可证：AGPL-3.0。", "why": "免费门槛适合个人实验和临时演示，但不应默认具备商业域名同等的长期稳定性；正式品牌需审慎评估所有权、续期和可迁移性。抖音推荐来源：https://v.douyin.com/GU_Ydi6KYKc/", "tags": [ "抖音推荐", "免费域名", "个人项目", "演示部署", "域名服务", "AGPL", "digitalplat", "domain", "domain-platform", "domainname", "domains", "free" ], "language": "Markdown", "stars": 187692, "forks": 3937, "license": "AGPL-3.0", "metadataVerifiedAt": "2026-07-22T21:37:57.321Z", "source": "ai-workspace", "video": "", "icon": "globe", "color": "", "createdAt": "2026-07-22T21:37:22.216Z", "updatedAt": "2026-07-22T21:28:25Z" }, { "id": "github-com-openbmb-voxcpm", "name": "VoxCPM", "repo": "https://github.com/OpenBMB/VoxCPM", "homepage": "https://voxcpm.com", "summary": "OpenBMB开源的无分词器多语言TTS模型，支持创意音色设计、提示词生成声线和高拟真声音克隆。", "details": "VoxCPM2是VoxCPM仓库中的第二代模型，面向多语言语音合成、创意声音设计和高拟真克隆。它可根据文本描述设计声线，也可在获得合法授权的前提下，根据参考音频保持说话人音色与表达风格。截图中的“99%还原”“任何歌声均可一比一复刻”等为传播文案，不作为官方保证。声音克隆须取得本人或权利人明确授权，并显著标注AI生成，不得用于冒充、欺诈或绕过身份验证。", "why": "兼具多语言合成和声音克隆能力，适合研究与合规内容生产；推荐时特别保留授权、隐私及声音滥用风险提示。抖音推荐来源：https://v.douyin.com/40DZ_lfBrUk/", "tags": [ "抖音推荐", "文本转语音", "声音克隆", "多语言", "PyTorch", "音频生成", "audio", "deeplearning", "minicpm", "multilingual", "python", "pytorch" ], "language": "Python", "stars": 34420, "forks": 3934, "license": "Apache-2.0", "metadataVerifiedAt": "2026-07-29T00:29:39Z", "source": "douyin-screenshot", "video": "", "icon": "spark", "color": "", "createdAt": "2026-07-22T21:37:22.216Z", "updatedAt": "2026-07-29T00:17:59Z", "titleZh": "VoxCPM2多语言语音生成与声音克隆", "evidence": "用户提供的M_Lab抖音截图、GitHub官方API、OpenBMB/VoxCPM仓库已核验；项目已存在，本次更新VoxCPM2信息。" }, { "id": "github-com-donchitos-claude-code-game-studios", "name": "Claude-Code-Game-Studios", "repo": "https://github.com/Donchitos/Claude-Code-Game-Studios", "homepage": "", "summary": "将Claude Code组织成包含49个角色代理和72项工作流技能的协作式游戏开发体系。", "details": "解决独立游戏开发中角色分工与流程协调问题，以真实工作室层级为参照，提供49个AI代理、72项工作流技能及完整协调系统，覆盖游戏设计与开发场景。适合独立开发者和小团队；它是辅助工作流，不会自动生成可直接发布的完整成品，仍需人工决策、实现与测试。主要语言：Shell；Stars：23314；许可证：MIT。", "why": "角色和技能体系较完整，可用于结构化组织Claude Code参与游戏开发，但应将其视为协作框架而非一键成品生成器。抖音推荐来源：https://v.douyin.com/NVybuJ6xSZY/", "tags": [ "抖音推荐", "Claude Code", "游戏开发", "多代理协作", "工作流", "独立游戏", "ai-agents", "ai-assisted-development", "anthropic", "claude", "claude-code", "game-design" ], "language": "Shell", "stars": 23314, "forks": 3341, "license": "MIT", "metadataVerifiedAt": "2026-07-22T21:37:57.162Z", "source": "ai-workspace", "video": "", "icon": "rocket", "color": "", "createdAt": "2026-07-22T21:37:22.216Z", "updatedAt": "2026-07-22T21:23:02Z" }, { "id": "github-com-dietrichgebert-ponytail", "name": "ponytail", "repo": "https://github.com/DietrichGebert/ponytail", "homepage": "https://ponytail.dev", "summary": "一套用YAGNI等原则约束AI代理过度设计和过度编码的规则与技能。", "details": "针对AI编程代理容易增加无关抽象、功能和复杂度的问题，通过规则、插件与代理技能引导其像资深开发者一样优先选择最小可行实现。适合Claude Code、Cursor等用户；它不是代码生成平台，也不能替代架构评审和人工判断。主要语言：JavaScript / Python / PowerShell；Stars：87850；许可证：MIT。", "why": "其价值在于限制AI过度工程化，而非让AI编写更多代码，适合重视可维护性、YAGNI原则和变更范围控制的开发团队。抖音推荐来源：https://v.douyin.com/KVUefbSE3ME/", "tags": [ "抖音推荐", "代理技能", "YAGNI", "Claude Code", "Cursor规则", "提示工程", "agent-skills", "ai-agents", "claude", "claude-code", "claude-code-plugin", "cursor-rules" ], "language": "JavaScript / Python / PowerShell", "stars": 87850, "forks": 4809, "license": "MIT", "metadataVerifiedAt": "2026-07-22T21:37:57.145Z", "source": "ai-workspace", "video": "", "icon": "bot", "color": "", "createdAt": "2026-07-22T21:37:22.216Z", "updatedAt": "2026-07-22T21:33:05Z" }, { "id": "github-com-colbymchenry-codegraph", "name": "codegraph", "repo": "https://github.com/colbymchenry/codegraph", "homepage": "https://colbymchenry.github.io/codegraph/", "summary": "面向多种AI编程工具的本地代码知识图谱，通过预索引与自动同步减少上下文消耗和工具调用。", "details": "解决AI代理反复扫描代码库、消耗大量令牌的问题，提供预索引代码知识图谱并随代码变更自动同步，适合使用Claude Code、Codex、Gemini或Cursor的开发者。数据可完全本地处理，但实际理解效果仍取决于索引质量与项目结构。主要语言：C / TypeScript / Rust；Stars：61694；许可证：MIT。", "why": "项目定位明确，兼容多类主流AI编程代理，适合希望降低令牌成本并提升大型代码库检索效率的团队。抖音推荐来源：https://v.douyin.com/QONUEJBgmtc/", "tags": [ "抖音推荐", "代码知识图谱", "AI编程", "本地运行", "代码索引", "开发者工具", "C", "MIT" ], "language": "C / TypeScript / Rust", "stars": 61694, "forks": 3860, "license": "MIT", "metadataVerifiedAt": "2026-07-22T21:37:57.142Z", "source": "ai-workspace", "video": "", "icon": "code", "color": "", "createdAt": "2026-07-22T21:37:22.216Z", "updatedAt": "2026-07-22T21:11:38Z" }, { "id": "github-com-modelcontextprotocol-servers", "name": "MCP Servers", "repo": "https://github.com/modelcontextprotocol/servers", "homepage": "https://modelcontextprotocol.io", "summary": "MCP 官方参考服务器与开发示例集合，演示如何让模型受控访问工具和数据。", "details": "Model Context Protocol 官方维护的参考实现集合，包含 Everything、Fetch、Filesystem、Git、Memory、Sequential Thinking、Time 等服务器，并链接各语言 SDK 与社区资源。它用于展示 MCP 协议、SDK 和权限设计，不是完整的生产服务器目录；官方明确提示参考实现不等于生产就绪，GitHub、PostgreSQL、Puppeteer、Slack 等一批旧参考服务器已迁入归档。新贡献采用 Apache-2.0，既有代码部分为 MIT。", "why": "适合 Claude Code 等 Agent 用户理解 MCP 配置、权限边界和服务器实现，也能作为开发自有 MCP 的最可信起点。实际部署前必须逐个审计文件系统路径、令牌、数据库权限、网络访问和命令执行能力，并优先从 MCP Registry 查找当前维护版本。", "tags": [ "抖音推荐", "MCP", "Model Context Protocol", "TypeScript", "Python", "参考实现", "权限隔离", "安全审查", "Apache-2.0", "MIT", "NOASSERTION" ], "language": "TypeScript / Python / JavaScript", "stars": 88788, "forks": 11275, "license": "NOASSERTION", "metadataVerifiedAt": "2026-07-22T21:17:35.572Z", "source": "manual", "video": "https://v.douyin.com/iBc8PUVLovU/", "icon": "terminal", "color": "#0ea5e9", "createdAt": "2026-07-22T21:17:35.572Z", "updatedAt": "2026-07-22T20:56:54Z" }, { "id": "github-com-touchhle-touchhle", "name": "touchHLE", "repo": "https://github.com/touchHLE/touchHLE", "homepage": "https://touchhle.org", "summary": "通过高层模拟运行部分早期 iPhone OS 应用和游戏的跨平台兼容层。", "details": "touchHLE 重新实现早期 iPhone OS 的部分系统框架与运行环境，使特定旧版 iPhone/iPod touch 应用和游戏可在现代桌面及移动平台运行。它采用高层模拟而非完整硬件虚拟化，重点是历史软件保存与兼容性；支持范围依赖逐个应用适配，不能运行现代 iOS、App Store 新应用，也不是 Xcode 官方 Simulator 的替代品。许可证与应用资产版权需分别核验。", "why": "对数字保存、老游戏研究和早期移动软件兼容具有独特价值，Rust 核心也适合学习高层模拟器设计。使用者必须自行合法取得应用文件，不能把“完整 iOS 模拟器”当作功能承诺；先查兼容列表再投入迁移。", "tags": [ "抖音推荐", "Rust", "iPhone OS", "模拟器", "兼容层", "游戏保存", "跨平台", "版权需注意", "MPL-2.0" ], "language": "Rust", "stars": 3811, "forks": 276, "license": "MPL-2.0", "metadataVerifiedAt": "2026-07-22T21:17:08.410Z", "source": "manual", "video": "https://v.douyin.com/FPaHHoX6-UU/", "icon": "terminal", "color": "#f59e0b", "createdAt": "2026-07-22T21:17:08.410Z", "updatedAt": "2026-07-22T18:24:00Z" }, { "id": "github-com-cloakhq-cloakbrowser", "name": "CloakBrowser", "repo": "https://github.com/CloakHQ/CloakBrowser", "homepage": "https://cloakhq.com", "summary": "带源码级指纹补丁的定制 Chromium，可作为 Playwright/Puppeteer 的自动化浏览器。", "details": "面向浏览器自动化与授权测试的定制 Chromium 项目，通过浏览器源码级补丁处理常见指纹与自动化检测差异，提供接近 Playwright/Puppeteer 的调用方式、多身份配置和跨平台支持。它不是普通网页爬虫库，而是高权限浏览器运行环境。仓库、预编译二进制和商业使用条款可能存在不同边界，部署前必须分别核验源码、二进制来源与许可证。", "why": "适合研究浏览器指纹、防自动化检测和合规测试环境，也能减少传统脚本在自动化标记上的兼容问题。仅限自有系统或明确授权目标，禁止用于账号滥用、规避访问控制或违反网站条款；应在隔离环境运行并限制凭据、下载和网络权限。", "tags": [ "抖音推荐", "Chromium", "Playwright", "Puppeteer", "浏览器自动化", "仅限授权测试", "安全", "隐私", "许可证需核验", "ai-agents", "anti-detect", "antidetect-browser" ], "language": "Python / TypeScript / C#", "stars": 28929, "forks": 2342, "license": "MIT", "metadataVerifiedAt": "2026-07-22T21:17:05.405Z", "source": "manual", "video": "https://v.douyin.com/mODnHjphlPg/", "icon": "shield", "color": "#1f2937", "createdAt": "2026-07-22T21:17:05.405Z", "updatedAt": "2026-07-22T21:14:53Z" }, { "id": "github-com-harry0703-moneyprinterturbo", "name": "MoneyPrinterTurbo", "repo": "https://github.com/harry0703/MoneyPrinterTurbo", "homepage": "", "summary": "输入主题或关键词，自动完成文案、素材、字幕、配音、配乐与高清短视频合成。", "details": "一套端到端 AI 短视频生产流水线，支持自动生成或自定义文案，抓取/使用本地素材，合成语音、字幕和背景音乐，输出横屏或竖屏高清视频，并提供 Web 界面、API、批量生成与 Docker 部署。可接 OpenAI、DeepSeek、Gemini、Ollama、通义、Moonshot 等多种模型，字幕可选择轻量方案或 Whisper。MIT 许可证。", "why": "适合把选题快速变成可发布的视频初稿，尤其适合自媒体选题验证、批量栏目和内部宣传。生成质量取决于模型、素材源与配音服务；上线前必须人工复核事实、字幕、肖像与素材版权，并预算模型、语音和素材接口成本。", "tags": [ "抖音推荐", "AI 视频", "内容创作", "视频剪辑", "字幕", "配音", "FFmpeg", "Python", "Docker", "MIT", "素材版权", "ai-video-generator" ], "language": "Python / CSS / Dockerfile", "stars": 98652, "forks": 14590, "license": "MIT", "metadataVerifiedAt": "2026-07-22T21:17:02.482Z", "source": "manual", "video": "https://v.douyin.com/yd11kCoFFNQ/", "icon": "spark", "color": "#22c55e", "createdAt": "2026-07-22T21:17:02.482Z", "updatedAt": "2026-07-22T21:01:09Z" }, { "id": "github-com-ziwang-physics-agentchat", "name": "AgentChat", "repo": "https://github.com/ziwang-Physics/AgentChat", "homepage": "", "summary": "让 Claude Code 通过本地 Chrome/CDP 调用多个免费网页 AI 的 WebSubAgent 工作流。", "details": "Claude Code 原生 Skill 集合，通过 Playwright/CDP 接管本地 Chrome，把 Gemini、ChatGPT、Claude、Qwen、Kimi、MiniMax、MiMo、DeepSeek 等网页端 AI 作为子代理。包含单模型自动降级链、独立任务并行编排和“规划→搜索→推理→合成→审查”的串行管道，支持代码与多模态任务。MIT 许可证。使用时必须已有对应网页账号登录会话，网页结构变化会导致适配失效。", "why": "能把多个网页 AI 的免费额度与能力组织成可复用的 Claude Code 工作流，适合大量独立任务和交叉审查。它依赖浏览器会话与第三方服务条款，可能触发验证码、限流或账号风控；不得绕过平台限制，敏感数据也不应发送给不受控网页模型。", "tags": [ "抖音推荐", "Claude Code", "Agent Skills", "CDP", "Playwright", "多模型", "浏览器自动化", "MIT", "账号风险", "JavaScript" ], "language": "JavaScript / Python / Shell", "stars": 409, "forks": 60, "license": "MIT", "metadataVerifiedAt": "2026-07-22T21:16:59.346Z", "source": "manual", "video": "https://v.douyin.com/ljd4_IXdl6Q/", "icon": "bot", "color": "#7c3aed", "createdAt": "2026-07-22T21:16:59.346Z", "updatedAt": "2026-07-22T18:05:19Z" }, { "id": "github-com-davidhdev-react-bits", "name": "react-bits", "repo": "https://github.com/DavidHDev/react-bits", "homepage": "https://reactbits.dev", "summary": "140+ 个可复制、可定制的 React 动画与交互组件，覆盖文字、背景和界面效果。", "details": "面向现代 React 网站的动画组件与创意工具库，提供文字动画、UI 元素、动态背景等 140+ 组件，并为每个组件提供 JS+CSS、JS+Tailwind、TS+CSS、TS+Tailwind 变体。组件可通过 shadcn/jsrepo 安装，也可直接复制源码修改；另带 Background Studio、Shape Magic、Texture Lab 等视觉工具。它主要服务 React 项目，Vue 与 Svelte 需使用官方移植版本。许可证为 MIT + Commons Clause，商用前应阅读具体限制。", "why": "视频关联搜索直接显示 react bits，并展示覆盖常见场景的 React 动效组件库；原项目已存在，本次补充视频证据。", "tags": [ "抖音推荐", "React", "TypeScript", "JavaScript", "前端动画", "UI 设计", "动效", "MIT+Commons Clause", "3d", "animations", "component-library", "components" ], "language": "JavaScript / TypeScript / CSS", "stars": 43950, "forks": 2069, "license": "NOASSERTION", "metadataVerifiedAt": "2026-07-22T21:16:56.626Z", "source": "manual", "video": "https://v.douyin.com/5fBQugeHY_k/", "icon": "globe", "color": "#6c63ff", "createdAt": "2026-07-22T21:16:56.626Z", "updatedAt": "2026-07-22T20:48:25Z" }, { "id": "github-com-dexhunter-seedance2-skill", "name": "seedance2-skill", "repo": "https://github.com/dexhunter/seedance2-skill", "homepage": "https://seedance2.help", "summary": "Seedance 2.0 专业视频提示词 Skill，帮助 Agent 编写可执行的分镜与多模态引用提示词。", "details": "面向 Seedance 2.0 视频生成的提示词工程指南与 Agent Skill。它系统整理 @图片/@视频/@音频引用语法、镜头运动、分段时间轴、角色一致性、视频续写与编辑、音乐卡点、电商广告、短剧和科普内容等模板。适合内容创作者、导演型 Agent 与提示词工程师使用；项目本身只负责生成和优化提示词，不负责运行视频模型，也不能替代平台内容审核。许可证以仓库当前文件为准。", "why": "结构化程度高，能把模糊创意转成含主体、场景、动作、镜头、时序、转场和声音的可执行提示词，特别适合作为 Codex、Claude Code 等 Agent 的视频导演技能。使用前仍需核对 Seedance 当前素材数量、时长、人脸与内容合规限制。", "tags": [ "抖音推荐", "Seedance 2.0", "AI 视频", "Agent Skills", "提示词", "分镜", "内容创作", "prompt", "seedance", "seedance-2", "video-prompt", "MIT" ], "language": "Markdown", "stars": 2937, "forks": 289, "license": "MIT", "metadataVerifiedAt": "2026-07-28T07:11:51.637476+00:00", "source": "manual", "video": "https://www.douyin.com/video/7659295544153492774", "icon": "auto", "color": "#ec5ba7", "createdAt": "2026-07-22T21:15:52.062Z", "updatedAt": "2026-07-22T15:29:27Z" }, { "id": "github-com-wshobson-agents", "name": "agents", "repo": "https://github.com/wshobson/agents", "homepage": "https://sethhobson.com", "summary": "包含大量专业角色、Skills 与编排模式的 Claude Code 智能体集合，可组成完整的 AI 员工团队。", "details": "该项目按工程、架构、测试、安全、数据、产品和运营等领域提供专业智能体与协作流程。它是视频“80 个专业 AI 员工”对应仓库，也是完整 AI 公司套件的员工库；使用高权限 Agent 时必须限定目录、命令、外部系统和审批范围。", "why": "最高优先收藏：它是这组“完整 AI 公司”的员工主体，配合 DeerFlow 调研、Prompt Optimizer 翻译、Ruflo 编排与 Letta 记忆形成完整闭环。", "tags": [ "最高优先收藏", "完整AI公司", "AI员工库", "Claude Code", "多智能体", "Agent Skills", "工作流", "权限隔离", "agent-skills", "agentic-ai", "agents", "ai-agents" ], "language": "Python / C# / Shell", "stars": 38128, "forks": 4089, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:30:17.765Z", "source": "抖音《这5件让Claude替你打工》", "video": "https://www.douyin.com/video/7637155097838914843", "icon": "bot", "color": "#f59e0b", "createdAt": "2026-07-22T07:30:17.765Z", "updatedAt": "2026-07-22T07:18:36Z" }, { "id": "github-com-ruvnet-ruflo", "name": "ruflo", "repo": "https://github.com/ruvnet/ruflo", "homepage": "https://Cognitum.One", "summary": "面向企业级多智能体编排的 Claude Flow 后继项目，用 Swarm、记忆与工作流协调复杂任务。", "details": "Ruflo（原 Claude-Flow）提供多智能体拓扑、任务编排、持久记忆、MCP 与开发工作流能力。在 AI 公司套件中承担“总经理”角色；高自治编排必须配合最小权限、隔离工作区、额度限制和人工审批。", "why": "完整 AI 公司核心套件的总经理与项目编排层，用于把复杂目标拆给不同智能体执行。", "tags": [ "最高优先收藏", "完整AI公司", "多智能体", "Swarm", "任务编排", "MCP", "Claude Code", "权限隔离", "agentic-ai", "agentic-framework", "agentic-workflow", "agents" ], "language": "TypeScript / JavaScript / Shell", "stars": 65476, "forks": 7771, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:30:16.168Z", "source": "抖音《这5件让Claude替你打工》", "video": "https://www.douyin.com/video/7637155097838914843", "icon": "bot", "color": "#0ea5e9", "createdAt": "2026-07-22T07:30:16.168Z", "updatedAt": "2026-07-22T07:22:27Z" }, { "id": "github-com-simplex-chat-simplex-chat", "name": "simplex-chat", "repo": "https://github.com/simplex-chat/simplex-chat", "homepage": "https://simplex.chat", "summary": "不依赖用户标识符的开源私密通信平台，支持消息、群组、文件与音视频通话。", "details": "SimpleX Chat 通过去中心化队列和可轮换连接标识降低元数据暴露，提供移动端、桌面端与命令行客户端。隐私设计不等于绝对匿名，终端安全、备份、联系人验证和网络环境仍会影响整体风险。", "why": "热榜视频第 6 个项目，是研究去中心化通信和元数据隐私的成熟开源实现。", "tags": [ "隐私通信", "端到端加密", "去中心化", "聊天", "元数据保护", "Haskell", "移动端", "安全", "chat", "double-ratchet", "e2ee", "encryption" ], "language": "Haskell / Kotlin / Swift", "stars": 18873, "forks": 1191, "license": "AGPL-3.0", "metadataVerifiedAt": "2026-07-22T07:30:12.910Z", "source": "抖音 GitHub 本周热榜", "video": "https://www.douyin.com/video/7658853026634976521", "icon": "shield", "color": "#2563eb", "createdAt": "2026-07-22T07:30:12.910Z", "updatedAt": "2026-07-22T07:26:21Z" }, { "id": "github-com-thedotmack-claude-mem", "name": "claude-mem", "repo": "https://github.com/thedotmack/claude-mem", "homepage": "https://claude-mem.ai", "summary": "为 Claude Code 等编码智能体建立跨会话持久记忆，自动保存、压缩并检索项目上下文。", "details": "Claude-Mem 通过 Hooks 和本地服务记录会话观察，再把相关记忆注入后续任务，减少智能体遗忘偏好、命名规范和历史决策。记忆库可能包含源码、路径或敏感上���文，部署时应限制访问、审查保存范围并妥善保护本地数据。", "why": "视频中作为五件套的“记忆层”，解决长项目反复重新解释约定和历史决策的问题。", "tags": [ "Claude Code", "Agent Memory", "持久记忆", "上下文", "Hooks", "本地数据", "隐私", "重点收藏", "ai", "ai-agents", "ai-memory", "anthropic" ], "language": "JavaScript / TypeScript / Shell", "stars": 88177, "forks": 7654, "license": "Apache-2.0", "metadataVerifiedAt": "2026-07-22T07:30:11.395Z", "source": "抖音《这5个Skill让你的Claude Code提高十倍生产力》", "video": "https://www.douyin.com/video/7636723951208500532", "icon": "database", "color": "#ec4899", "createdAt": "2026-07-22T07:30:11.395Z", "updatedAt": "2026-07-22T07:24:16Z" }, { "id": "github-com-yt-dlp-yt-dlp", "name": "yt-dlp", "repo": "https://github.com/yt-dlp/yt-dlp", "homepage": "https://discord.gg/H5MNcFW63r", "summary": "支持数千站点的功能丰富型命令行音视频下载、信息提取与媒体处理工具。", "details": "yt-dlp 支持大量视频网站与音频站点，可选择清晰度和编码格式、下载字幕/封面/元数据/播放列表、断点续传、并发分片，并通过 FFmpeg 合并音视频或执行后处理。它常被 AI Agent 和媒体工作流作为底层下载引擎调用。工具开源不代表可以绕过付费、DRM、账号限制或版权授权；使用时必须遵守平台条款和当地法律。源码仓库采用 Unlicense，但部分官方打包产物会捆绑其他许可证组件，分发前需查看对应发布说明。", "why": "本次抖音视频开头直接展示 yt-dlp 官方 README、项目标志和文件名；此前收藏视频也已独立确认同一仓库，因此合并为多视频来源而不创建重复项目。", "tags": [ "视频下载", "音频下载", "命令行", "Python", "FFmpeg", "字幕", "元数据", "版权合规", "cli", "downloader", "python", "sponsorblock", "断点续传", "播放列表", "格式选择", "多站点", "媒体处理", "平台条款" ], "language": "Python / Makefile / Shell / Dockerfile", "stars": 179639, "forks": 15293, "license": "Unlicense", "metadataVerifiedAt": "2026-07-23T12:19:10+00:00", "source": "抖音收藏（多条视频确认）", "video": "https://www.douyin.com/video/7652671296039341375", "icon": "terminal", "color": "#dc2626", "createdAt": "2026-07-22T07:29:55.072Z", "updatedAt": "2026-07-23T11:55:26Z", "relatedVideos": [ "https://www.douyin.com/video/7654896385111903529", "https://www.douyin.com/video/7652671296039341375" ], "evidence": "两条抖音视频画面、GitHub 官方仓库 API 与 README 已核验" }, { "id": "github-com-letta-ai-letta", "name": "letta", "repo": "https://github.com/letta-ai/letta", "homepage": "https://docs.letta.com/", "summary": "用于构建有状态、可长期学习和持续记忆的 AI Agent 平台。", "details": "Letta 提供 Agent 状态、核心记忆、归档记忆、工具、服务端和开发者接口，使智能体能跨会话保留偏好与工作上下文。在 AI 公司套件中承担长期大脑角色，涉及个人偏好与业务资料时应配置数据保留、访问控制和删除策略。", "why": "完整 AI 公司核心套件的长期记忆层，让员工持续积累上下文而不是每次从零开始。", "tags": [ "最高优先收藏", "完整AI公司", "Agent Memory", "长期记忆", "有状态Agent", "Python", "数据治理", "隐私", "ai", "ai-agents", "llm", "llm-agent" ], "language": "Python / Go / Shell", "stars": 23910, "forks": 2544, "license": "Apache-2.0", "metadataVerifiedAt": "2026-07-22T07:29:53.574Z", "source": "抖音《这5件让Claude替你打工》", "video": "https://www.douyin.com/video/7637155097838914843", "icon": "database", "color": "#ec4899", "createdAt": "2026-07-22T07:29:53.574Z", "updatedAt": "2026-07-22T02:46:14Z" }, { "id": "github-com-linshenkx-prompt-optimizer", "name": "prompt-optimizer", "repo": "https://github.com/linshenkx/prompt-optimizer", "homepage": "https://prompt.always200.com", "summary": "可视化、多模型、跨平台的提示词优化工具，把自然语言需求迭代成更清晰可执行的 AI 指令。", "details": "Prompt Optimizer 支持多模型对比、提示词迭代、变量、历史记录、测试和多种部署方式。在 AI 公司套件中承担“翻译师”角色；优化后的提示仍需人工确认边界，避免把危险或含敏感信息的请求自动放大。", "why": "视频页面搜索词为“多功能 prompt 优化器”，与 Prompt Optimizer 的多模型提示词重写和对比功能一致。", "tags": [ "最高优先收藏", "完整AI公司", "提示词优化", "多模型", "Prompt", "AI工具", "自托管", "安全审查", "ai-prompts", "ai-tools", "llm", "prompt" ], "language": "TypeScript / Vue / JavaScript", "stars": 32407, "forks": 3790, "license": "NOASSERTION", "metadataVerifiedAt": "2026-07-22T07:29:51.266Z", "source": "抖音《这5件让Claude替你打工》", "video": "https://v.douyin.com/K0y_zMcQJY0/", "icon": "spark", "color": "#6366f1", "createdAt": "2026-07-22T07:29:51.266Z", "updatedAt": "2026-07-22T07:02:31Z" }, { "id": "github-com-luanti-org-luanti", "name": "luanti", "repo": "https://github.com/luanti-org/luanti", "homepage": "https://www.luanti.org/", "summary": "支持 Lua 模组和游戏定义的开源体素游戏引擎，可自由制作方块、生物、地图与玩法。", "details": "Luanti 原名 Minetest，支持 Windows、macOS、Linux、BSD 与 Android，围绕 Lua API、模组、游戏包和社区内容构建开放创作生态。引擎开源不代表社区纹理、音乐和模组都可任意商用，发布作品前应逐项核对许可。", "why": "对应视频展示的“卢安提”，适合新手从 Lua 脚本入门独立游戏和模组开发。", "tags": [ "游戏引擎", "体素游戏", "Lua", "模组", "开源游戏", "跨平台", "新手开发", "素材许可", "c-plus-plus", "cpp17", "game-engine", "lua" ], "language": "C++ / Lua / C", "stars": 13320, "forks": 2389, "license": "NOASSERTION", "metadataVerifiedAt": "2026-07-22T07:29:48.112Z", "source": "抖音《新手也能徒手做独立游戏》", "video": "https://www.douyin.com/video/7659712785407823104", "icon": "code", "color": "#65a30d", "createdAt": "2026-07-22T07:29:48.112Z", "updatedAt": "2026-07-22T00:01:04Z" }, { "id": "github-com-xbtlin-ai-berkshire", "name": "ai-berkshire", "repo": "https://github.com/xbtlin/ai-berkshire", "homepage": "https://github.com/xbtlin/ai-berkshire#readme", "summary": "基于 Claude Code 与 Codex 的多智能体价值投资研究框架。", "details": "AI Berkshire 将巴菲特、芒格、段永平与李录等方法论组织为并行研究和交叉质疑流程，用于公司基本面与风险分析。项目属于研究工具，模型结论可能错误或过时，不构成投资建议，真实交易前必须核对原始财报并独立决策。", "why": "热榜视频第 5 个项目，适合研究多角色协作、证据链和反方审查在金融分析中的应用。", "tags": [ "AI Agent", "价值投资", "金融研究", "Claude Code", "Codex", "多智能体", "风险提示", "非投资建议", "ai", "ai-agent", "anthropic", "berkshire-hathaway" ], "language": "Python / Shell / Batchfile", "stars": 13645, "forks": 1909, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:29:44.496Z", "source": "抖音 GitHub 本周热榜", "video": "https://www.douyin.com/video/7658853026634976521", "icon": "bot", "color": "#16a34a", "createdAt": "2026-07-22T07:29:44.496Z", "updatedAt": "2026-07-22T07:29:40Z" }, { "id": "github-com-calesthio-openmontage", "name": "OpenMontage", "repo": "https://github.com/calesthio/OpenMontage", "homepage": "https://www.openmontage.video/", "summary": "面向 AI 编码智能体的开源 Agent 视频生产系统，覆盖研究、脚本、素材、剪辑与合成。", "details": "OpenMontage 提供多条制作流水线、工具与大量 Agent Skills，可调用免费素材库、本地 TTS、FFmpeg 以及多种付费生成服务。生成视频仍需核对素材、音乐、人物肖像和平台条款，API 服务也可能产生费用。", "why": "热榜视频第 4 个项目，把编码智能体变成可复现的视频制作工作室。", "tags": [ "AI 视频", "Agent Skills", "视频剪辑", "FFmpeg", "Python", "素材版权", "生成式 AI", "Codex", "agent", "agentic-ai", "ai", "claude" ], "language": "Python / HTML / JavaScript", "stars": 40983, "forks": 4843, "license": "AGPL-3.0", "metadataVerifiedAt": "2026-07-22T07:29:42.967Z", "source": "抖音 GitHub 本周热榜", "video": "https://www.douyin.com/video/7658853026634976521", "icon": "spark", "color": "#ef4444", "createdAt": "2026-07-22T07:29:42.967Z", "updatedAt": "2026-07-22T07:25:58Z" }, { "id": "github-com-deusdata-codebase-memory-mcp", "name": "codebase-memory-mcp", "repo": "https://github.com/DeusData/codebase-memory-mcp", "homepage": "https://deusdata.github.io/codebase-memory-mcp/", "summary": "把代码库高速索引成持久知识图谱并通过 MCP 提供结构查询的代码智能后端。", "details": "项目以 Tree-sitter、静态二进制和本地图数据库索引函数、类型、调用链、路由与跨服务关系，面向 Claude Code、Codex 等 MCP 客户端减少逐文件读取。它会读取源码并可写入 Agent 配置，安装前应审查脚本和代码仓库权限。", "why": "热榜视频第 2 个项目，适合大型代码库的调用链、影响分析和低 Token 结构检索。", "tags": [ "MCP", "代码智能", "知识图谱", "Tree-sitter", "调用链", "本地索引", "Claude Code", "Codex", "aider", "ast", "claude-code", "code-analysis" ], "language": "C / C++ / Shell", "stars": 33848, "forks": 2595, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:29:41.271Z", "source": "抖音 GitHub 本周热榜", "video": "https://www.douyin.com/video/7658853026634976521", "icon": "database", "color": "#0ea5e9", "createdAt": "2026-07-22T07:29:41.271Z", "updatedAt": "2026-07-22T07:29:24Z" }, { "id": "github-com-msitarzewski-agency-agents", "name": "agency-agents", "repo": "https://github.com/msitarzewski/agency-agents", "homepage": "", "summary": "可直接调用的专业 AI 智能体角色库，覆盖工程、设计、产品、营销、测试与运营。", "details": "Agency Agents 以角色说明和工作流程组织多种专业智能体，可用于 Claude Code 等 Agent 环境。角色提示能提高分工效率，但不能替代权限隔离、事实核验和人工审批，尤其不应让营销或运维角色无监督操作生产系统。", "why": "视频所说的“144 位各领域专家”对应 Agency Agents 的多岗位 Agent 角色库；仓库已在项目页，本次补充视频证据。", "tags": [ "AI Agent", "角色库", "多智能体", "工程", "设计", "产品", "营销", "Claude Code", "Shell", "MIT" ], "language": "Shell / Python / PowerShell", "stars": 135611, "forks": 22083, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:29:39.700Z", "source": "抖音 GitHub 本周热榜", "video": "https://v.douyin.com/9LcQs06Li_g/", "icon": "bot", "color": "#f97316", "createdAt": "2026-07-22T07:29:39.700Z", "updatedAt": "2026-07-22T07:29:03Z" }, { "id": "github-com-pbakaus-impeccable", "name": "impeccable", "repo": "https://github.com/pbakaus/impeccable", "homepage": "https://impeccable.style", "summary": "用一套设计语言、命令与可检测反模式，让 AI 生成的前端更专业、更一致。", "details": "Impeccable 包含设计 Skill、设计领域参考、审查与润色命令，以及无需模型的反模式检测工具，覆盖字体、颜色、空间、动效、响应式与 UX 文案。抖音作者的自制 VibeCoding UI Skill 没有公开唯一仓库地址，本条收录视频明确提到的另一个上游参考项目。", "why": "它把“不要有廉价 AI 味”转化为可执行、可检查的设计规则，是视频 UI Skill 的另一项明确参考。", "tags": [ "Agent Skills", "UI UX", "设计系统", "反模式", "前端", "审查", "Apache-2.0", "视频参考上游", "JavaScript" ], "language": "JavaScript / CSS / TypeScript", "stars": 48612, "forks": 2837, "license": "Apache-2.0", "metadataVerifiedAt": "2026-07-22T07:29:37.970Z", "source": "抖音《一个Skill让AI做出高级UI》参考项目", "video": "https://www.douyin.com/video/7660003772609072818", "icon": "spark", "color": "#7c3aed", "createdAt": "2026-07-22T07:29:37.970Z", "updatedAt": "2026-07-22T07:23:51Z" }, { "id": "github-com-leonxlnx-taste-skill", "name": "taste-skill", "repo": "https://github.com/Leonxlnx/taste-skill", "homepage": "https://tasteskill.dev", "summary": "面向 AI 编码智能体的反模板化前端设计 Skills，让布局、字体、动效与视觉层级更有设计感。", "details": "Taste Skill 提供通用、Codex、重设计、高端视觉、极简、粗野主义与图像先行等多个可安装 Skill，并以可调的变化度、动效和密度控制结果。抖音作者的自制 VibeCoding UI Skill 没有公开唯一仓库地址，本条收录视频明确提到的上游参考项目。", "why": "可直接提升 AI 生成界面的审美与差异化，是视频自制 UI Skill 的明确参考来源之一。", "tags": [ "Agent Skills", "UI 设计", "前端", "反模板化", "Codex", "Claude Code", "动效", "视频参考上游", "agent", "ai", "claude", "claude-code" ], "language": "JavaScript / Shell", "stars": 66153, "forks": 4568, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:29:36.551Z", "source": "抖音《一个Skill让AI做出高级UI》��考项目", "video": "https://www.douyin.com/video/7660003772609072818", "icon": "spark", "color": "#e11d48", "createdAt": "2026-07-22T07:29:36.551Z", "updatedAt": "2026-07-22T07:29:33Z" }, { "id": "github-com-iofficeai-aionui", "name": "AionUi", "repo": "https://github.com/iOfficeAI/AionUi", "homepage": "https://www.aionui.com", "summary": "统一管理 Claude Code、Codex、Hermes 等终端智能体的开源 Cowork 图形界面。", "details": "AionUi 提供多智能体会话、工作区、文件操作、定时任务、远程访问、语音与多模型接入，可把分散的命令行 Agent 放进统一控制台。远程模式和任意 API Key 能力需要严格保护鉴权、文件权限与网络暴露面。", "why": "视频中作为五件套的“入口层”，也直接支持 Hermes 与 Codex，适合统一查看和控制多任务。", "tags": [ "Cowork", "多智能体", "Claude Code", "Codex", "Hermes", "GUI", "远程访问", "重点收藏", "acp", "agent-team", "ai", "ai-agent" ], "language": "TypeScript / CSS / JavaScript", "stars": 30624, "forks": 3067, "license": "Apache-2.0", "metadataVerifiedAt": "2026-07-22T07:29:34.996Z", "source": "抖音《这5个Skill让你的Claude Code提高十倍生产力》", "video": "https://www.douyin.com/video/7636723951208500532", "icon": "bot", "color": "#06b6d4", "createdAt": "2026-07-22T07:29:34.996Z", "updatedAt": "2026-07-22T07:19:47Z" }, { "id": "github-com-hkuds-lightrag", "name": "LightRAG", "repo": "https://github.com/HKUDS/LightRAG", "homepage": "https://arxiv.org/abs/2410.05779", "summary": "港大数据智能实验室开源的轻量级图增强检索框架，用知识图谱与向量检索理解复杂资料和代码关系。", "details": "LightRAG 将实体、关系与文本块结合，支持多种图存储、向量库、模型和 API 服务，可用于代码库、文档与知识库问答。索引私有代码或业务资料时应确认模型与存储边界，避免敏感内容被发送到未授权的外部服务。", "why": "视频中作为五件套的“上下文层”，让智能体从逐文件猜测升级为理解跨文件关系和整体结构。", "tags": [ "RAG", "GraphRAG", "知识图谱", "代码理解", "HKU", "Python", "隐私", "重点收藏", "genai", "gpt", "gpt-4", "graphrag" ], "language": "Python / TypeScript / Shell", "stars": 37985, "forks": 5349, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:29:33.300Z", "source": "抖音《这5个Skill让你的Claude Code提高十倍生产力》", "video": "https://www.douyin.com/video/7636723951208500532", "icon": "database", "color": "#8b5cf6", "createdAt": "2026-07-22T07:29:33.300Z", "updatedAt": "2026-07-22T07:19:30Z" }, { "id": "github-com-mattpocock-skills", "name": "skills", "repo": "https://github.com/mattpocock/skills", "homepage": "", "summary": "Matt Pocock 维护的 AI 编程 Skills 集合，覆盖需求追问、规格、PRD、任务拆分、TDD 与代码库改进。", "details": "该仓库把资深 TypeScript 与产品工程经验整理成可安装的 Agent Skills，例如 grill-me、grill-with-docs、to-prd、to-issues 与 tdd。Skills 属于高权限工作指令，使用前应阅读内容并确认写文件、执行命令的边界。", "why": "视频中作为五件套的“代码质量层”，尤其适合先把模糊需求问清楚，再进入实现。", "tags": [ "Agent Skills", "Claude Code", "Codex", "TypeScript", "TDD", "PRD", "需求澄清", "重点收藏", "Shell", "MIT" ], "language": "Shell / JavaScript", "stars": 183735, "forks": 15726, "license": "MIT", "metadataVerifiedAt": "2026-07-23T12:56:13+00:00", "source": "抖音《这5个Skill让你的Claude Code提高十倍生产力》", "video": "https://www.douyin.com/video/7636723951208500532", "icon": "spark", "color": "#3178c6", "createdAt": "2026-07-22T07:29:30.960Z", "updatedAt": "2026-07-23T12:54:02Z" }, { "id": "github-com-kangarooking-cangjie-skill", "name": "cangjie-skill", "repo": "https://github.com/kangarooking/cangjie-skill", "homepage": "", "summary": "把书籍、长视频、播客、课程与长文中的方法论蒸馏成可调用、可组合、可测试的 Agent Skills。", "details": "仓颉 Skill 使用 RIA-TV++ 流水线完成整体理解、并行提取、三重验证、可执行结构化、知识链接、压力测试和交付，可生成 BOOK_OVERVIEW、INDEX、DIGEST、术语表、多个 SKILL.md 与测试用例。视频通常先由配套工具提取字幕、转写和素材，再做方法论蒸馏。蒸馏能力不等于获得内容权利：处理受版权保护的视频、课程、书籍、人物声音或表达风格前，应取得授权并遵守平台条款；产出不得冒充原作者、传播大段受保护原文或用于未经同意的人格仿制。", "why": "相比普通摘要，它重点保留触发条件、执行步骤、适用边界、Skill 关联和压力测试，让收藏的视频与长内容能在 Agent 的真实任务中被反复调用。", "tags": [ "Agent Skills", "视频蒸馏", "知识蒸馏", "长视频", "播客", "书籍", "方法论", "Claude Code", "OpenClaw", "Python", "版权需授权", "MIT" ], "language": "Python", "stars": 4319, "forks": 565, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:08:50.419Z", "source": "douyin-favorite-300", "video": "", "icon": "spark", "color": "#f5b942", "createdAt": "2026-07-22T07:08:50.442Z", "updatedAt": "2026-07-22T07:08:34Z" }, { "id": "github-com-withmarbleapp-os-taxonomy", "name": "os-taxonomy", "repo": "https://github.com/withmarbleapp/os-taxonomy", "homepage": "", "summary": "面向小学阶段的开放教育知识地图，把八个学科拆成细粒度知识点并连接成可计算的先修关系图。", "details": "Marble Skill Taxonomy v1 收录 1,590 个微知识点、3,221 条先修依赖、183 个领域摘要和八个学科，并与 Common Core、NGSS、英国国家课程等标准对齐。项目以 UTF-8 JSON、JSON Schema、校验脚本和校验和发布，可用于学习路径、AI 辅导、自适应教育和知识图谱产品。数据库采用 ODbL 1.0、Marble 原创文本采用 CC BY-SA 4.0；课程标准仍遵循各上游许可，复用时必须署名、遵守相同方式共享要求并核对 PROVENANCE。", "why": "它不是单一教学应用，而是一套可直接被教育产品和 AI 系统读取的底层知识结构，能回答某个知识点依赖什么、掌握标准是什么以及下一步能解锁什么。", "tags": [ "教育知识地图", "知识图谱", "小学教育", "先修关系", "课程标准", "AI 辅导", "自适应学习", "JSON", "开放数据", "ODbL-1.0", "CC BY-SA 4.0", "JavaScript" ], "language": "JavaScript", "stars": 3552, "forks": 608, "license": "ODbL-1.0", "metadataVerifiedAt": "2026-07-22T07:08:50.442Z", "source": "douyin-favorite-300", "video": "", "icon": "database", "color": "#0ea5e9", "createdAt": "2026-07-22T07:08:50.442Z", "updatedAt": "2026-07-22T06:39:19Z" }, { "id": "github-com-usestrix-strix", "name": "strix", "repo": "https://github.com/usestrix/strix", "homepage": "https://strix.ai", "summary": "面向开发与安全团队的开源自主 AI 渗透测试框架，可发现、验证并辅助修复应用漏洞。", "details": "STRIX 在 Docker 沙箱中编排多个 AI 安全智能体，提供侦察���HTTP 拦截、浏览器测试、Shell、动态与静态分析、漏洞验证、PoC、修复建议、报告及 CI/CD 集成。它具备真实利用验证和攻击性工具能力，只能用于自有系统或已获得书面授权的测试范围；严禁扫描、利用或访问未授权目标，运行时还应隔离网络和凭据、限定目标、保存审计记录并人工复核结果。", "why": "相比仅报告疑似问题的传统扫描器，STRIX 能动态运行目标并验证漏洞，同时输出复现步骤与修复建议；适合在明确授权和隔离环境中研究 AI 辅助应用安全与持续渗透测试。", "tags": [ "AI 安全", "自动化渗透测试", "多智能体", "漏洞验证", "PoC", "DevSecOps", "CI/CD", "Python", "Docker 沙箱", "仅限授权测试", "攻击性工具", "Apache-2.0" ], "language": "Python / TypeScript / Jinja", "stars": 43261, "forks": 4468, "license": "Apache-2.0", "metadataVerifiedAt": "2026-07-22T07:03:59.368Z", "source": "douyin-favorite-300", "video": "", "icon": "shield", "color": "#dc2626", "createdAt": "2026-07-22T07:03:59.368Z", "updatedAt": "2026-07-22T07:00:21Z" }, { "id": "github-com-paperclipai-paperclip", "name": "paperclip", "repo": "https://github.com/paperclipai/paperclip", "homepage": "https://paperclip.ing", "summary": "用于组建和管理 AI 员工团队的开源控制台，把目标、组织架构、任务、预算、审批与审计集中在一个“AI 公司”里。", "details": "Paperclip 由 Node.js 服务端和 React 界面组成，可接入 OpenClaw、Claude Code、Codex、Cursor、命令行及 HTTP 智能体，通过心跳调度员工工作。它提供多公司隔离、组织层级、任务委派、预算限制、审批关卡、持久会话、密钥边界、成本追踪和不可变审计记录。自主运行不代表无需监督；生产环境仍应采用最小权限、沙箱、额度上限和人工审批，避免智能体误操作或失控消耗。", "why": "它把多个零散的 AI Agent 终端提升为可管理的组织系统，适合构建 AI 公司、持续运行的智能体团队，以及统一观察 Hermes、Codex 等员工的任务与成本。", "tags": [ "AI 公司", "多智能体", "Agent 编排", "组织架构", "任务管理", "预算控制", "审批治理", "Codex", "Claude Code", "OpenClaw", "TypeScript", "MIT" ], "language": "TypeScript / JavaScript / Shell", "stars": 74414, "forks": 13852, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:03:59.348Z", "source": "douyin-favorite-300", "video": "", "icon": "bot", "color": "#f59e0b", "createdAt": "2026-07-22T07:03:59.368Z", "updatedAt": "2026-07-22T06:32:16Z" }, { "id": "github-com-datawhalechina-easy-vibe", "name": "easy-vibe", "repo": "https://github.com/datawhalechina/easy-vibe", "homepage": "https://datawhalechina.github.io/easy-vibe/", "summary": "Datawhale 面向零基础到进阶开发者的 Vibe Coding 多语言教程，带你从自然语言想法逐步做出可上线应用。", "details": "Easy Vibe 提供从 AI 编程初体验、产品原型、全栈开发到 Claude Code、MCP、Skills、Agent Teams 和跨平台交付的分阶段学习路径，并包含可视化、交互式示例及十种语言文档。GitHub API 当前未识别到许可证，仓库根目录也没有可访问的 LICENSE 文件；README 徽章宣称 CC BY-NC-SA 4.0 且许可证链接失效，因此在作者补齐正式许可前，不应默认用于商业复制或再分发。", "why": "这是同名仓库中活跃度和关注度最高的 DatawhaleChina 官方项目，适合初学者系统学习 Vibe Coding，也能作为开发者查阅 AI 原生工程流程的路线图。", "tags": [ "Vibe Coding", "AI 编程", "Datawhale", "多语言教程", "全栈开发", "Claude Code", "MCP", "Agent Teams", "JavaScript", "许可证待确认", "agent", "ai" ], "language": "JavaScript / Dockerfile / Shell", "stars": 18415, "forks": 1754, "license": "", "metadataVerifiedAt": "2026-07-22T06:02:36.531Z", "source": "douyin-favorite-300", "video": "", "icon": "spark", "color": "#7c3aed", "createdAt": "2026-07-22T06:02:36.531Z", "updatedAt": "2026-07-22T04:10:13Z" }, { "id": "github-com-ruvnet-ruview", "name": "RuView", "repo": "https://github.com/ruvnet/RuView", "homepage": "https://Cognitum.One/RuView", "summary": "利用 WiFi CSI 和低成本边缘传感器进行穿墙存在、动作、姿态与生命体征感知的平台。", "details": "RuView 通过 ESP32 等 CSI 采集节点分析无线信号扰动，可用于房间占用、动作、呼吸、心率、跌倒风险和穿墙检测，并支持本地边缘运行及智能家居集成。高精度能力依赖兼容硬件、多个采集节点、环境校准和模型数据，并非任意普通路由器安装软件后即可获得宣传中的透视效果；人体位置和生命体征仍属于敏感个人数据，部署必须获得授权并遵守隐私与无线电法规。", "why": "用户点名的“RuView WiFi 变透视”对应该官方仓库；项目展示了无摄像头空间感知的完整工程路径，同时值得关注其硬件门槛、精度边界和隐私风险。", "tags": [ "WiFi CSI", "穿墙感知", "ESP32", "边缘计算", "姿态估计", "生命体征", "Rust", "隐私敏感", "硬件要求", "授权使用", "awesome", "claude" ], "language": "Rust / Python / JavaScript", "stars": 82839, "forks": 11161, "license": "MIT", "metadataVerifiedAt": "2026-07-22T05:54:58.231Z", "source": "douyin-favorite-300", "video": "", "icon": "shield", "color": "#00a6a6", "createdAt": "2026-07-22T05:54:58.249Z", "updatedAt": "2026-07-22T05:54:15Z" }, { "id": "github-com-manalkaff-opendesign", "name": "opendesign", "repo": "https://github.com/manalkaff/opendesign", "homepage": "", "summary": "把 Claude Design 的结构化设计方法封装成可在 Codex、Claude Code 等本地智能体中安装的开源 Skills。", "details": "OpenDesign 提供设计系统提取、前端设计、线框探索、交互原型、演示文稿、可调参数设计和开发交接等技能，并能先读取现有代码库与品牌上下文再输出 HTML 设计成果。它是社区实现，并非 OpenAI 或 Anthropic 官方项目；使用前应审查 Skills 内容及其对本地项目文件的访问范围。", "why": "用户点名的 OpenAI Design 项目与收藏中的 OpenDesign 对应；它可安装到 Codex、Claude Code、Cursor、Gemini CLI 和 OpenCode，适合把可复用的设计工作流带入本地代码项目。", "tags": [ "OpenDesign", "Agent Skills", "UI 设计", "设计系统", "Codex", "Claude Code", "交互原型", "HTML", "社区项目", "MIT" ], "language": "HTML / JavaScript / Shell", "stars": 219, "forks": 28, "license": "MIT", "metadataVerifiedAt": "2026-07-22T05:54:58.249Z", "source": "douyin-favorite-300", "video": "", "icon": "spark", "color": "#8b5cf6", "createdAt": "2026-07-22T05:54:58.249Z", "updatedAt": "2026-07-22T02:21:56Z" }, { "id": "github-com-alibaba-page-agent", "name": "page-agent", "repo": "https://github.com/alibaba/page-agent", "homepage": "https://alibaba.github.io/page-agent/", "summary": "阿里巴巴开源的网页内 JavaScript GUI Agent，可用自然语言直接操作 Web 界面。", "details": "Page Agent 直接嵌入网页运行，通过文本化 DOM 理解和操作界面，无需 Python、无头浏览器或独立扩展；支持自带模型、表单自动化、产品内 AI Copilot，并可选接入扩展与 MCP。生产使用时不应把长期 API 密钥直接暴露在前端，建议通过受控后端代理并配置允许操作范围。", "why": "收藏中点名的 Page Agent.js 对应阿里巴巴官方仓库；它与 Browser Use 的外部浏览器自动化路线不同，适合给现有 SaaS 或管理后台快速嵌入自然语言操作能力。", "tags": [ "AI Agent", "网页自动化", "JavaScript", "TypeScript", "DOM", "MCP", "SaaS Copilot", "阿里巴巴", "前端密钥需保护", "MIT", "agent", "ai" ], "language": "TypeScript / JavaScript / CSS", "stars": 27438, "forks": 2406, "license": "MIT", "metadataVerifiedAt": "2026-07-22T05:51:24.609Z", "source": "douyin-favorite-300", "video": "", "icon": "globe", "color": "#ff6a00", "createdAt": "2026-07-22T05:51:24.609Z", "updatedAt": "2026-07-22T05:36:20Z" }, { "id": "github-com-github-spec-kit", "name": "spec-kit", "repo": "https://github.com/github/spec-kit", "homepage": "https://github.github.com/spec-kit/", "summary": "GitHub 官方的规格驱动开发工具包，用结构化流程约束 AI 编码。", "details": "通过 Specify CLI 和模板把需求拆为规格、计划、任务与实现阶段，支持 Codex、Copilot、Claude 等多种编码智能体，并可用扩展和工作流定制研发流程。", "why": "收藏视频将它与 OpenSpec、BMAD 对比；它由 GitHub 官方维护，适合为 AI 辅助开发建立可审查的需求与交付链路。", "tags": [ "SDD", "规格驱动开发", "AI 编程", "GitHub", "Python", "Codex", "研发流程", "MIT", "ai", "copilot", "development", "engineering" ], "language": "Python / Shell / PowerShell", "stars": 123331, "forks": 10997, "license": "MIT", "metadataVerifiedAt": "2026-07-23T06:13:17+00:00", "source": "douyin-favorite-300", "video": "https://www.douyin.com/video/7585176612473556287", "icon": "code", "color": "#24292f", "createdAt": "2026-07-22T05:05:11.797Z", "updatedAt": "2026-07-23T06:10:20Z" }, { "id": "github-com-xxhuberrr-mineradio", "name": "Mineradio", "repo": "https://github.com/XxHuberrr/Mineradio", "homepage": "", "summary": "以电影镜头、粒子视觉和歌词舞台为核心的沉浸式开源音乐播放器。", "details": "聚焦本地音乐播放、动态歌词与 3D 可视化体验，适合作为音频可视化和桌面交互设计参考。音乐源、歌词和封面等内容的使用仍需遵守相应版权规则。", "why": "收藏视频直接展示并感谢开源的 Mineradio；GitHub 官方搜索可明确区分主仓库与二次修改版本。", "tags": [ "音乐播放器", "音频可视化", "歌词", "3D", "桌面应用", "开源", "版权需注意", "HTML", "NOASSERTION" ], "language": "HTML / JavaScript / NSIS", "stars": 9042, "forks": 700, "license": "NOASSERTION", "metadataVerifiedAt": "2026-07-28T07:11:51.637476+00:00", "source": "douyin-favorite-300", "video": "https://www.douyin.com/video/7667219322967805210", "icon": "spark", "color": "#d14b8f", "createdAt": "2026-07-22T05:04:55.973Z", "updatedAt": "2026-07-22T04:45:07Z" }, { "id": "github-com-huangchihhungleo-claude-real-video", "name": "claude-real-video", "repo": "https://github.com/HUANGCHIHHUNGLeo/claude-real-video", "homepage": "", "summary": "让 Claude 或其他大模型在本地真正理解视频画面与语音内容的工具。", "details": "从网址或本地文件提取场景感知、去重后的关键帧与转写文本，再交给大模型分析，减少整段视频直接输入的成本；处理私人视频时需注意本地文件和转写内容的隐私。", "why": "收藏列表直接出现 claude-real-video 名称，GitHub 官方搜索确认该主仓库与功能描述一致。", "tags": [ "视频理解", "Claude", "LLM", "关键帧", "语音转写", "本地运行", "隐私", "MIT", "claude", "claude-code", "cli", "ffmpeg" ], "language": "Python / HTML / CSS", "stars": 1814, "forks": 148, "license": "MIT", "metadataVerifiedAt": "2026-07-22T05:04:45.009Z", "source": "douyin-favorite-300", "video": "", "icon": "spark", "color": "#d85f45", "createdAt": "2026-07-22T05:04:45.009Z", "updatedAt": "2026-07-22T05:02:34Z" }, { "id": "github-com-bmad-code-org-bmad-method", "name": "BMAD-METHOD", "repo": "https://github.com/bmad-code-org/BMAD-METHOD", "homepage": "", "summary": "以角色化智能体和敏捷流程组织需求、架构、开发与测试的 AI 方法框架。", "details": "提供分析师、产品经理、架构师、开发者和 QA 等角色及配套工作流，用上下文工程推动复杂项目。GitHub API 未给出标准 SPDX 许可证，采用或商用前需核对仓库最新授权文本。", "why": "收藏视频明确将 BMAD 与 Spec Kit、OpenSpec 对比，适合研究重规划、多角色的 AI 软件交付方法。", "tags": [ "AI Agent", "敏捷开发", "多角色", "工作流", "JavaScript", "上下文工程", "许可证需核验", "NOASSERTION" ], "language": "JavaScript / HTML / Python", "stars": 50932, "forks": 5849, "license": "NOASSERTION", "metadataVerifiedAt": "2026-07-22T05:04:35.281Z", "source": "douyin-favorite-300", "video": "", "icon": "bot", "color": "#f2a900", "createdAt": "2026-07-22T05:04:35.281Z", "updatedAt": "2026-07-22T04:38:58Z" }, { "id": "github-com-xai-org-grok-build", "name": "grok-build", "repo": "https://github.com/xai-org/grok-build", "homepage": "", "summary": "xAI 组织开源的终端 AI 编程智能体 Harness 与交互式 TUI。", "details": "提供全屏终端界面、鼠标交互和可扩展的编码智能体运行层，适合在终端内组织��码理解、修改与评审任务；接入仓库和命令执行时应限制权限与敏感凭据。", "why": "收藏内容点名 Grok Build，GitHub 官方搜索确认主仓库位于 xai-org 组织，避免误收同名第三方项目。", "tags": [ "AI 编程", "Grok", "终端", "TUI", "Agent Harness", "开发工具", "权限隔离", "Rust", "Apache-2.0" ], "language": "Rust / Python / JavaScript", "stars": 21505, "forks": 4007, "license": "Apache-2.0", "metadataVerifiedAt": "2026-07-22T05:03:54.666Z", "source": "douyin-favorite-300", "video": "", "icon": "terminal", "color": "#111111", "createdAt": "2026-07-22T05:03:54.666Z", "updatedAt": "2026-07-22T05:02:44Z" }, { "id": "github-com-facebook-astryx", "name": "astryx", "repo": "https://github.com/facebook/astryx", "homepage": "http://astryx.atmeta.com", "summary": "Meta 开源、可全面定制并面向 AI Agent 使用的设计系统。", "details": "为产品界面提供设计令牌、组件和智能体可读取的设计上下文，目标是在人工设计与 AI 生成代码之间保持一致的视觉语言；落地前应核对组件成熟度和品牌素材授权。", "why": "收藏中的一周 AI 资讯明确点名 Meta 开源 Astryx，GitHub 官方搜索确认仓库归属 facebook 组织。", "tags": [ "设计系统", "Meta", "Agent Ready", "UI", "前端", "设计令牌", "开源", "TypeScript", "MIT" ], "language": "TypeScript / JavaScript / CSS", "stars": 10117, "forks": 748, "license": "MIT", "metadataVerifiedAt": "2026-07-22T05:03:44.737Z", "source": "douyin-favorite-300", "video": "", "icon": "spark", "color": "#1769e0", "createdAt": "2026-07-22T05:03:44.737Z", "updatedAt": "2026-07-22T05:03:02Z" }, { "id": "github-com-ahujasid-blender-mcp", "name": "blender-mcp", "repo": "https://github.com/ahujasid/blender-mcp", "homepage": "https://blendermcp.org/", "summary": "通过 MCP 让 Claude、Codex 等大模型直接读取和控制 Blender 的开源桥接器。", "details": "由 Blender 插件和 Python MCP 服务组成，可操作对象、材质、场景与外部素材。它允许模型在 Blender 内执行 Python 代码，存在高权限与提示注入风险，必须只连接可信客户端并隔离运行环境。", "why": "收藏视频明确展示 Claude 连接 Blender；该仓库是对应的主流开源实现，适合研究 AI 辅助 3D 建模。", "tags": [ "Blender", "MCP", "AI", "3D 建模", "Python", "Claude", "Codex", "高权限", "安全隔离", "MIT" ], "language": "Python", "stars": 24591, "forks": 2371, "license": "MIT", "metadataVerifiedAt": "2026-07-22T05:03:38.531Z", "source": "douyin-favorite-300", "video": "", "icon": "spark", "color": "#e87d0d", "createdAt": "2026-07-22T05:03:38.531Z", "updatedAt": "2026-07-22T04:44:45Z" }, { "id": "github-com-zhulinsen-daily-stock-analysis", "name": "daily_stock_analysis", "repo": "https://github.com/ZhuLinsen/daily_stock_analysis", "homepage": "https://dsa.zhulinsen.tech", "summary": "聚合多市场行情与新闻、由大模型生成分析报告并自动推送的股票研究系统。", "details": "支持 A 股、港股、美股和 ETF，提供 Web 工作台、回测、持仓、Agent 策略问股及 GitHub Actions 零服务器定时任务。分析结果仅供信息整理与研究，不构成投资建议。", "why": "收藏视频描述的五万星 AI 股票分析、Web 工作台和 Actions 定时运行特征与该官方仓库完全一致。", "tags": [ "AI", "股票分析", "量化研究", "GitHub Actions", "Python", "Web 工作台", "非投资建议", "MIT", "a-stock", "ai-agent", "aigc", "llm" ], "language": "Python / TypeScript / JavaScript", "stars": 58195, "forks": 50015, "license": "MIT", "metadataVerifiedAt": "2026-07-22T05:03:17.860Z", "source": "douyin-favorite-300", "video": "https://www.douyin.com/video/7656321732915039529", "icon": "database", "color": "#2f80ed", "createdAt": "2026-07-22T05:03:17.860Z", "updatedAt": "2026-07-22T04:59:48Z" }, { "id": "github-com-earthtojake-text-to-cad", "name": "text-to-cad", "repo": "https://github.com/earthtojake/text-to-cad", "homepage": "https://www.cadskills.xyz", "summary": "供 Codex、Claude Code 等智能体生成和检查 CAD、机器人与硬件设计的技能库。", "details": "覆盖 STEP、STL、3MF、DXF、GLB、URDF、SRDF 和 SDF 的生成、渲染、审查与可复现工作流，并提供本地 CAD 查看器和基准资产。", "why": "收藏视频点名 CAD Skills；它把机械设计规则封装成可安装技能，适合探索 AI 编程智能体参与硬件与机器人设计。", "tags": [ "CAD", "Agent Skills", "机器人", "硬件设计", "JavaScript", "Codex", "3D", "MIT", "3mf", "agents", "ai-agents", "build123d" ], "language": "JavaScript / Python / Shell", "stars": 9266, "forks": 1033, "license": "MIT", "metadataVerifiedAt": "2026-07-22T05:03:12.739Z", "source": "douyin-favorite-300", "video": "", "icon": "code", "color": "#ef6c35", "createdAt": "2026-07-22T05:03:12.739Z", "updatedAt": "2026-07-22T05:02:07Z" }, { "id": "github-com-fission-ai-openspec", "name": "OpenSpec", "repo": "https://github.com/Fission-AI/OpenSpec", "homepage": "https://openspec.dev/", "summary": "为 AI 编码助手提供轻量、可审查的规格驱动开发工作流。", "details": "把当前规格与变更提案分开管理，通过提案、评审、实施和归档阶段保持需求可追踪；无需 API 密钥，可接入多种现有编码助手。", "why": "收藏视频明确比较 OpenSpec、Spec Kit 与 BMAD；它尤其适合在现有项目上用小步变更控制 AI 实现范围。", "tags": [ "SDD", "OpenSpec", "AI 编程", "需求管理", "TypeScript", "变更追踪", "工程流程", "MIT", "ai", "context-engineering", "engineering", "planning" ], "language": "TypeScript / JavaScript / Shell", "stars": 61976, "forks": 4289, "license": "MIT", "metadataVerifiedAt": "2026-07-22T05:02:58.722Z", "source": "douyin-favorite-300", "video": "", "icon": "code", "color": "#00a88f", "createdAt": "2026-07-22T05:02:58.722Z", "updatedAt": "2026-07-22T05:01:53Z" }, { "id": "github-com-esengine-deepseek-reasonix", "name": "DeepSeek-Reasonix", "repo": "https://github.com/esengine/DeepSeek-Reasonix", "homepage": "http://reasonix.io/", "summary": "围绕 DeepSeek 前缀缓存稳定性设计的终端 AI 编程智能体。", "details": "面向长时间编程会话优化提示词前缀和缓存命中，提供终端交互、工具调用、项目上下文与多语言文档；项目专门适配 DeepSeek，选型前需确认模型供应商和费用策略。", "why": "收藏视频明确点名 deepseek-reasonix，适合研究低成本长会话编码 Agent 与缓存友好的运行循环。", "tags": [ "AI", "编程智能体", "DeepSeek", "终端", "提示词缓存", "Go", "TypeScript", "MIT", "agent", "agent-framework", "ai-agent", "ai-coding" ], "language": "Go / TypeScript / CSS", "stars": 27512, "forks": 1760, "license": "MIT", "metadataVerifiedAt": "2026-07-22T05:02:38.265Z", "source": "douyin-favorite-300", "video": "", "icon": "terminal", "color": "#7b61ff", "createdAt": "2026-07-22T05:02:38.265Z", "updatedAt": "2026-07-22T04:49:49Z" }, { "id": "github-com-browser-use-browser-harness", "name": "browser-harness", "repo": "https://github.com/browser-use/browser-harness", "homepage": "https://browser-harness.com", "summary": "通过轻量 CDP 连接让大模型直接控制真实浏览器，并在执行中自我补全工具的浏览器 Harness。", "details": "项目以可编辑、自修复的浏览器执行层连接 Chrome，强调持久会话和任务技能沉淀。它拥有较高浏览器控制权限，部署时必须隔离配置文件、Cookie 和敏感站点。", "why": "收藏视频明确介绍 Browser Harness 的轻量实现与自修复范式，适合研究可持续演进的浏览器 Agent。", "tags": [ "AI", "Agent", "Browser Harness", "CDP", "浏览器自动化", "Python", "安全隔离", "ai-agent", "browser-agent", "browser-automation", "browser-use", "browser-use-box" ], "language": "Python / HTML / Shell", "stars": 16160, "forks": 1523, "license": "MIT", "metadataVerifiedAt": "2026-07-22T04:43:05.630Z", "source": "douyin-favorite", "video": "", "icon": "terminal", "color": "#6f63e8", "createdAt": "2026-07-22T04:43:05.834Z", "updatedAt": "2026-07-22T03:28:18Z" }, { "id": "github-com-awesome-selfhosted-awesome-selfhosted", "name": "awesome-selfhosted", "repo": "https://github.com/awesome-selfhosted/awesome-selfhosted", "homepage": "https://awesome-selfhosted.net/", "summary": "系统整理可在自有服务器部署的自由软件网络服务与 Web 应用。", "details": "按类别收录大量自托管服务及其源码、演示和许可��信息。该仓库本身是目录，列表内各项目许可证不同，采用前需逐项核对维护状态与授权条款。", "why": "收藏视频提到 23.9 万星的自建服务合集；该官方仓库是查找自托管方案的长期维护索引。", "tags": [ "自托管", "开源合集", "服务器", "Web 应用", "隐私", "运维", "许可证需逐项核对", "awesome", "awesome-list", "cloud", "free-software", "hosting" ], "language": "Markdown", "stars": 307265, "forks": 14408, "license": "NOASSERTION", "metadataVerifiedAt": "2026-07-22T04:43:05.834Z", "source": "douyin-favorite", "video": "https://www.douyin.com/video/7532830101835173154", "icon": "server", "color": "#2e9d8f", "createdAt": "2026-07-22T04:32:27.590Z", "updatedAt": "2026-07-22T04:41:12Z" }, { "id": "github-com-shubhamsaboo-awesome-llm-apps", "name": "awesome-llm-apps", "repo": "https://github.com/Shubhamsaboo/awesome-llm-apps", "homepage": "https://www.theunwindai.com", "summary": "汇总100多个可运行的AI Agents、Agent Skills与RAG应用，覆盖常驻代理、多智能体团队、语音、多模态和检索增强生成。", "details": "Awesome LLM Apps是Shubham Saboo维护的开源案例库，按场景组织100多个AI Agent、Agent Skills和RAG应用，包含Always-on Agents、Multi-agent Teams、语音与多模态、金融、法律、研究和自动化等方向。项目适合作为示例索引与学习材料；每个子项目仍需单独核对依赖、第三方API费用、许可证和生产安全边界，不能把合集里的全部示例视为同等成熟。", "why": "它把大量独立AI应用整理为可检索、可运行的工程案例，适合快速寻找原型和参考架构；相比逐个搜索项目，合集能显著降低选型和学习成本。", "tags": [ "AI", "LLM", "Agent", "Skills", "RAG", "Python", "示例合集", "agents", "llms", "python", "rag", "Apache-2.0" ], "language": "Python", "stars": 128708, "forks": 18998, "license": "Apache-2.0", "metadataVerifiedAt": "2026-07-29T17:43:53.002970+00:00", "source": "douyin", "video": "https://www.douyin.com/video/7667502634140320102", "icon": "spark", "color": "#e85aad", "createdAt": "2026-07-22T04:32:27.590Z", "updatedAt": "2026-07-29T17:33:08Z", "titleZh": "100+ AI Agent、Skills与RAG开源应用合集", "relatedVideos": [ "https://www.douyin.com/video/7664073102514523392", "https://www.douyin.com/video/7667502634140320102" ], "evidence": "抖音23秒画面中的Always-on Agents、Multi-agent Teams、EvoAgentX、Release Radar等与官方README交叉核验；GitHub官方API核验仓库、Stars、Forks与许可证。" }, { "id": "github-com-yikart-aitoearn", "name": "AiToEarn", "repo": "https://github.com/yikart/AiToEarn", "homepage": "https://aitoearn.ai/", "summary": "面向一人公司与内容创作者的 AI 内容生产、发布、互动和变现平台。", "details": "提供多平台内容生成与分发、互动运营、MCP 接入和 Docker 自托管能力。连接社交平台前应审查账号权限、平台规则及自动化操作风险。", "why": "收藏视频直接点名 AiToEarn，适合研究一人公司、内容运营 Agent 与跨平台发布工作流。", "tags": [ "AI", "Agent", "内容创作", "一人公司", "自动发布", "MCP", "TypeScript", "自托管", "auto-publish", "douyin", "douyin-api", "electron-app" ], "language": "TypeScript / JavaScript / SCSS", "stars": 24054, "forks": 3708, "license": "MIT", "metadataVerifiedAt": "2026-07-22T04:43:05.638Z", "source": "douyin-favorite", "video": "", "icon": "rocket", "color": "#ff8a3d", "createdAt": "2026-07-22T04:32:27.590Z", "updatedAt": "2026-07-22T03:57:56Z" }, { "id": "github-com-browser-use-browser-use", "name": "browser-use", "repo": "https://github.com/browser-use/browser-use", "homepage": "https://browser-use.com", "summary": "让 AI Agent 能够理解网页并执行浏览器任务的开源自动化框架。", "details": "以 Python 和浏览器自动化为核心，支持 Agent 操作网页、组合模型与工具并运行多步骤在线任务；实际使用应限制允许访问的域名和敏感凭据范围。", "why": "收藏视频明确推荐 BrowserUse，适合浏览器 Agent、自动化流程与网页任务编排。", "tags": [ "AI", "Agent", "浏览器自动化", "Python", "Playwright", "LLM", "安全边界", "ai-agents", "ai-tools", "browser-automation", "browser-use", "llm" ], "language": "Python / Shell / Dockerfile", "stars": 105981, "forks": 11664, "license": "MIT", "metadataVerifiedAt": "2026-07-22T04:43:05.649Z", "source": "douyin-favorite", "video": "https://www.douyin.com/video/7661848299988602148", "icon": "globe", "color": "#4b8bf4", "createdAt": "2026-07-22T04:32:27.590Z", "updatedAt": "2026-07-22T04:37:14Z" }, { "id": "github-com-d4vinci-scrapling", "name": "Scrapling", "repo": "https://github.com/D4Vinci/Scrapling", "homepage": "https://scrapling.readthedocs.io/en/latest/", "summary": "自适应 Python 网页采集框架，从单次请求到大规模爬取都提供统一工具链。", "details": "提供自适应元素定位、浏览器自动化、会话与代理管理、并发抓取及断���恢复能力。使用时仍需遵守目标网站条款、robots 规则与当地法律。", "why": "收藏视频重点介绍其抗页面结构变化与完整采集能力，适合数据采集、自动化和 Agent 工具链研究。", "tags": [ "Python", "爬虫", "数据采集", "自动化", "Playwright", "MCP", "合规使用", "ai", "ai-scraping", "automation", "crawler", "crawling" ], "language": "Python / Dockerfile", "stars": 70609, "forks": 7012, "license": "BSD-3-Clause", "metadataVerifiedAt": "2026-07-22T04:43:05.620Z", "source": "douyin-favorite", "video": "https://www.douyin.com/video/7618005778348063098", "icon": "globe", "color": "#39a96b", "createdAt": "2026-07-22T04:32:27.590Z", "updatedAt": "2026-07-22T04:28:48Z" }, { "id": "github-com-bytedance-deer-flow", "name": "deer-flow", "repo": "https://github.com/bytedance/deer-flow", "homepage": "https://deerflow.tech", "summary": "字节跳动开源的长时任务 SuperAgent 框架，可把研究、编程与内容创作拆解为持续执行的多步骤工作流。", "details": "DeerFlow 围绕沙箱、记忆、工具、Skills、子智能体和消息网关构建长任务运行时，适合深度研究与持续执行。作为 AI 公司套件的调研员，生产使用仍需配置成本上限、沙箱和人工复核。", "why": "完整 AI 公司核心套件的调研员；与提示词优化、员工角色库、总经理编排和长期记忆共同使用。", "tags": [ "最高优先收藏", "完整AI公司", "AI调研员", "SuperAgent", "多智能体", "长任务", "Python", "TypeScript", "agent", "agentic", "agentic-framework", "agentic-workflow" ], "language": "Python / TypeScript / MDX", "stars": 77568, "forks": 10559, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:29:49.690Z", "source": "抖音《这5件让Claude替你打工》", "video": "https://www.douyin.com/video/7637155097838914843", "icon": "bot", "color": "#b565f2", "createdAt": "2026-07-22T04:32:27.590Z", "updatedAt": "2026-07-22T07:24:35Z" }, { "id": "github-com-workbzw-ai-office-react", "name": "AI Office React", "repo": "https://github.com/workbzw/ai-office-react", "homepage": "", "summary": "把 AI 员工状态和任务动作呈现在像素风虚拟办公室中的 React 控制台。", "details": "基于 Vite、React、TypeScript 与 PixiJS 构建的独立 AI 办公室前端，内置 HTTP Action Gateway，可通过 desk_visit、desk_visit_tour、set_state 等动作驱动员工移动、拜访、状态与任务展示。项目适合进一步连接 Hermes 或其他 Agent 系统；生产部署需要由 Nginx 托管静态文件，并把动作网关限制在本机后增加鉴权、限流与参数校验。", "why": "视觉表达清晰，动作协议简单，适合作为多 Agent 状态看板和 Hermes 任务控制台的前端原型；MIT 许可证便于二次开发，但仓库素材另有版权提醒，公开商用前应替换或确认素材授权。", "tags": [ "AI Agent", "虚拟办公室", "React", "TypeScript", "PixiJS", "Vite", "Hermes", "MIT" ], "language": "TypeScript / CSS / JavaScript", "stars": 103, "forks": 40, "license": "MIT", "metadataVerifiedAt": "2026-07-22T02:17:52.898Z", "source": "manual", "video": "", "icon": "auto", "color": "#7C5CFC", "createdAt": "2026-07-22T00:58:57.282Z", "updatedAt": "2026-07-22T00:33:38Z" }, { "id": "github-com-johnson-jia-exercises-dataset", "name": "exercises-dataset", "repo": "https://github.com/Johnson-Jia/exercises-dataset", "homepage": "", "summary": "包含 1324 个健身动作、六语说明和离线 GIF 演示的结构化数据集。", "details": "字段包含身体部位、器械、目标肌群、辅助肌群、六种语言动作说明和本地 GIF，可用附带页面离线浏览。仓库说明指出图片来自上游移除前的社区分支，媒体版权与许可并不明确，仅建议个人学习和离线研究，商业使用前必须自行完成权利核验。", "why": "数据结构完整、离线素材丰富，但许可证和 GIF 素材权利存在风险，因此加入项目库时保留显著警告，不作无条件推荐。", "tags": [ "数据集", "健身动作", "多语言", "离线", "GIF", "HTML", "许可需核验", "仅供学习", "抖音推荐" ], "language": "HTML", "stars": 0, "forks": 0, "license": "", "metadataVerifiedAt": "2026-07-22T02:17:53.480Z", "source": "IT咖啡馆《Github一周热点123期》", "video": "https://www.douyin.com/video/7663547890190634278", "icon": "database", "color": "#3b82f6", "createdAt": "2026-07-19T20:07:54.698Z", "updatedAt": "2026-07-01T04:43:54Z" }, { "id": "github-com-hkuds-vibe-trading", "name": "Vibe-Trading", "repo": "https://github.com/HKUDS/Vibe-Trading", "homepage": "https://vibetrading.wiki/", "summary": "用自然语言进行金融研究、回测和多智能体分析的个人交易 Agent。", "details": "支持跨市场数据研究、策略回测、多智能体团队和 MCP 工具接入，可通过自然语言组织金融分析工作流。项目定位偏研究与模拟，不应视为投资建议，也不应未经风控直接用于真实交易。", "why": "把数据、回测和多智能体研究整合到同一套自然语言工作流中，适合量化研究原型和策略验证。", "tags": [ "AI Agent", "金融研究", "策略回测", "多智能体", "MCP", "Python", "MIT", "非投资建议", "抖音推荐", "ai-agent", "algorithmic-trading", "backtesting" ], "language": "Python / TypeScript / HTML / JavaScript", "stars": 28536, "forks": 4612, "license": "MIT", "metadataVerifiedAt": "2026-07-29T19:34:01.367152+00:00", "source": "IT咖啡馆《Github一周热点123期》", "video": "https://www.douyin.com/video/7663547890190634278", "icon": "bot", "color": "#14b8a6", "createdAt": "2026-07-19T20:07:54.698Z", "updatedAt": "2026-07-29T19:34:01.367152+00:00", "relatedSources": [ "https://www.xiaohongshu.com/explore/6a5b9ac6000000000a0391f1" ] }, { "id": "github-com-voltagent-awesome-design-md", "name": "awesome-design-md", "repo": "https://github.com/VoltAgent/awesome-design-md", "homepage": "https://getdesign.md/", "summary": "为 AI 编码智能体整理的 DESIGN.md 设计系统资源合集。", "details": "收录 73 份来自知名品牌和产品的 DESIGN.md 文件，以纯文本描述设计令牌、组件规范和视觉规则。复制到项目后，可帮助 AI 编码智能体生成风格更统一的界面。", "why": "把设计系统变成智能体可以直接读取的项目上下文，适合快速统一网站和应用的视觉语言。", "tags": [ "DESIGN.md", "设计系统", "AI 编程", "设计令牌", "Figma", "Google Stitch", "MIT", "抖音推荐", "awesome-list", "design-md", "design-system", "design-tokens" ], "language": "Markdown", "stars": 103651, "forks": 11896, "license": "MIT", "metadataVerifiedAt": "2026-07-22T02:17:54.889Z", "source": "IT咖啡馆《Github一周热点123期》", "video": "https://www.douyin.com/video/7663547890190634278", "icon": "code", "color": "#f97316", "createdAt": "2026-07-19T20:07:54.698Z", "updatedAt": "2026-07-22T02:16:42Z" }, { "id": "github-com-stablyai-orca", "name": "orca", "repo": "https://github.com/stablyai/orca", "homepage": "https://onOrca.dev", "summary": "让多个 AI 编程智能体并行工作的 Agent 开发环境。", "details": "可让 Codex、Claude Code、OpenCode 和 Pi 等智能体并排运行，每个任务使用独立 Git worktree。内置终端分屏、设计模式、GitHub 与 Linear 集成、SSH 工作树、差异批注和移动端协作。", "why": "适合同时调度多个编码智能体，又能用 worktree 隔离改动，减少并行任务互相覆盖。", "tags": [ "AI Agent", "并行智能体", "Agent IDE", "Worktree", "Codex", "Claude Code", "TypeScript", "MIT", "抖音推荐", "ade", "agent-ide", "ai-agents" ], "language": "TypeScript / JavaScript / CSS / Swift", "stars": 32632, "forks": 2296, "license": "MIT", "metadataVerifiedAt": "2026-07-29T19:34:01.367152+00:00", "source": "IT咖啡馆《Github一周热点123期》", "video": "https://www.douyin.com/video/7663547890190634278", "icon": "terminal", "color": "#8b5cf6", "createdAt": "2026-07-19T20:07:54.698Z", "updatedAt": "2026-07-29T19:34:01.367152+00:00", "relatedSources": [ "https://www.xiaohongshu.com/explore/6a5b9ac6000000000a0391f1" ] }, { "id": "github-com-openmoss-moss-transcribe-diarize", "name": "MOSS-Transcribe-Diarize", "repo": "https://github.com/OpenMOSS/MOSS-Transcribe-Diarize", "homepage": "https://huggingface.co/OpenMOSS-Team/MOSS-Transcribe-Diarize", "summary": "面向长音频的端到端多说话人转写与说话人分离模型。", "details": "0.9B 开源音频理解模型，一次完成语音转写、说话人分离、时间戳和可选声学事件标注，支持 50 多种语言，适用于会议、播客、访谈、课程和视频。支持 Python、SGLang Omni、vLLM 以及字幕 Web 工作流。", "why": "把传统 ASR 与说话人分离合并为单模型流程，是本期视频首先介绍的项目。", "tags": [ "AI", "ASR", "说话人分离", "音频理解", "长音频", "Python", "多语言", "Apache-2.0", "抖音推荐" ], "language": "Python", "stars": 1090, "forks": 57, "license": "Apache-2.0", "metadataVerifiedAt": "2026-07-22T02:17:56.014Z", "source": "IT咖啡馆《Github一周热点123期》", "video": "https://www.douyin.com/video/7663547890190634278", "icon": "spark", "color": "#ec5ba7", "createdAt": "2026-07-19T20:07:54.698Z", "updatedAt": "2026-07-22T01:56:47Z" }, { "id": "github-com-open-webui-open-webui", "name": "示例：Open WebUI", "repo": "https://github.com/open-webui/open-webui", "homepage": "https://openwebui.com/", "summary": "自托管 AI 聊天与模型管理界面。", "details": "可连接 OpenAI 兼容接口、本地模型和知识库。后续正式项目由你手动上传或我解析视频后补充。", "why": "适合放入热门 AI 工具清单，作为卡片样式验收数据。", "tags": [ "AI", "自托管", "工具", "ai", "llm", "llm-ui", "llm-webui", "llms", "mcp", "ollama", "ollama-webui", "open-webui" ], "language": "Python / Svelte / JavaScript", "stars": 146255, "forks": 21214, "license": "NOASSERTION", "metadataVerifiedAt": "2026-07-22T02:17:57.459Z", "source": "initial-test", "video": "", "icon": "auto", "color": "", "createdAt": "2026-07-09T01:16:00.000Z", "updatedAt": "2026-07-22T02:11:29Z" }, { "id": "example-open-source-curation", "name": "示例：等待上传热门项目", "repo": "https://github.com/", "homepage": "", "summary": "这里会展示你手动上传或我从视频介绍中整理出的 GitHub 热门项目。", "details": "支持项目用途、核心能力、适合人群、部署入口、视频来源、备注与标签。正式数据上传后可删除这张示例卡片。", "why": "先占位验证卡片样式和管理流程。", "tags": [ "待整理", "GitHub", "手动上传" ], "language": "Unknown", "stars": null, "source": "system", "video": "", "createdAt": "2026-07-09", "updatedAt": "2026-07-09T01:18:29.246Z" }, { "id": "github-com-browser-use-video-use", "name": "video-use", "repo": "https://github.com/browser-use/video-use", "homepage": "", "summary": "让 Codex、Claude Code 或 Hermes 通过对话完成转写、剪辑、调色、字幕和动画的视频编辑 Skill。", "details": "Video Use 以逐词转写、EDL、FFmpeg 和按需视觉检查组织视频编辑，可自动去口头禅、剪停顿、调色、生成字幕与覆盖动画，并在输出前自检切点。ElevenLabs、素材和在线下载可能涉及费用、隐私、版权与平台条款。", "why": "热榜视频第 7 个项目，把可审查的文字决策和视频制作工具链结合起来。", "tags": [ "视频剪辑", "Agent Skills", "FFmpeg", "字幕", "Codex", "Claude Code", "Hermes", "版权需注意", "Python", "MIT" ], "language": "Python / HTML / Shell", "stars": 17482, "forks": 2119, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:34:04Z", "source": "抖音 GitHub 本周热榜", "video": "https://www.douyin.com/video/7658853026634976521", "icon": "spark", "color": "#db2777", "createdAt": "2026-07-22T07:34:04Z", "updatedAt": "2026-07-22T07:31:20Z" }, { "id": "github-com-diegosouzapw-omniroute", "name": "OmniRoute", "repo": "https://github.com/diegosouzapw/OmniRoute", "homepage": "https://omniroute.online", "summary": "在本地统一多家模型授权、API Key、路由、重试与 OpenAI 兼容接口的 AI 网关。", "details": "OmniRoute 支持多种 OAuth 与 API Key 提供方、模型格式转换、负载均衡、失败回退、用量统计和 Web 管理界面。它集中持有高价值凭据，部署时必须限制监听地址、开启认证、保护数据库并审查第三方授权条款。", "why": "热榜视频第 8 个项目，适合给 Codex、Claude Code 和其他客户端建立统一的模型入口与故障回退。", "tags": [ "AI Gateway", "模型路由", "OAuth", "OpenAI 兼容", "负载均衡", "失败回退", "凭据安全", "自托管", "a2a", "ai-agents", "ai-gateway", "anthropic" ], "language": "TypeScript / JavaScript / Shell", "stars": 24041, "forks": 3217, "license": "MIT", "metadataVerifiedAt": "2026-07-22T07:34:04Z", "source": "抖音 GitHub 本周热榜", "video": "https://www.douyin.com/video/7658853026634976521", "icon": "server", "color": "#0891b2", "createdAt": "2026-07-22T07:34:04Z", "updatedAt": "2026-07-22T07:32:57Z" }, { "id": "github-com-nutlope-hallmark", "name": "hallmark", "titleZh": "反AI模板化网页设计Skill", "repo": "https://github.com/Nutlope/hallmark", "homepage": "", "summary": "面向Claude Code、Cursor和Codex的反“AI味”设计Skill，通过结构选择、主题系统和质量检查减少千篇一律的生成式网页。", "details": "Hallmark提供网页设计工作流与检查规则，适用于新页面、设计审计和重设计，强调先确定页面结构与视觉方向，再执行一致性和反模板化检查。它能改善设计过程，但不能自动保证品牌原创性、可访问性、版权合规或真实用户体验，生产发布前仍需人工评审与浏览器验收。", "why": "它把“减少AI模板感”从抽象审美要求转成可复用的设计流程和检查项，可直接接入主流AI编程工具。", "tags": [ "AI", "设计", "Skill", "Claude Code", "Cursor", "Codex", "反模板化", "网页设计", "MIT" ], "language": "CSS", "stars": 19654, "forks": 972, "license": "MIT", "metadataVerifiedAt": "2026-07-29T19:34:01.367152+00:00", "source": "xiaohongshu-weekly", "sourceUrl": "https://www.xiaohongshu.com/explore/6a5b9ac6000000000a0391f1", "icon": "palette", "color": "#7c3aed", "createdAt": "2026-07-29T19:34:01.367152+00:00", "updatedAt": "2026-07-29T19:34:01.367152+00:00" }, { "id": "github-com-firecrawl-anydoc", "name": "anydoc", "titleZh": "Firecrawl 全格式文档转 Markdown 引擎", "repo": "https://github.com/firecrawl/anydoc", "homepage": "https://firecrawl.github.io/anydoc/", "summary": "Firecrawl 团队开源的高性能本地文档转换引擎，可将 Word、PowerPoint、Excel、OpenDocument、RTF、EPUB、CSV 和 PDF 统一转换为适合大模型读取的 Markdown。", "details": "Anydoc 以 Rust 实现，并提供 Node.js、Python、WASM 和 Agent Skill。格式自动检测、统一 Markdown 序列化，适合 Hermes 在读取办公文档时先本地转换再按需分段，减少外部工具链、等待时间和上下文浪费；扫描或纯图片 PDF 仍需 OCR。", "why": "它直接补齐 Hermes 对多种办公文档的快速本地读取能力，本机 ARM64 实测 DOCX 转换约 0.13–0.15 秒，无需上传文件或启动 LibreOffice。", "tags": [ "AI", "文档解析", "Markdown", "Rust", "Agent Skill", "Word", "PowerPoint", "Excel", "PDF", "本地优先", "MIT" ], "language": "Rust", "stars": 13908, "forks": 712, "license": "MIT", "install": "本机已安装 @firecrawl/anydoc 0.1.8 与 ARM64 原生绑定；Hermes 可通过 convert-documents-to-markdown Skill 调用。", "compatibility": "支持 Linux ARM64；文本型文档本地转换。扫描/图片型 PDF 不含 OCR，继续使用 OCR 工具。", "evidence": "已核验官方仓库、README、许可证、npm版本，并在当前 Hermes 主机用真实 DOCX 完成冷/热转换测试和输出回读。", "metadataVerifiedAt": "2026-08-11T09:12:36.368326+00:00", "source": "抖音 阿甘探AI", "sourceUrl": "https://www.douyin.com/video/7672507250216373547", "video": "https://v.douyin.com/rkPBOGU_rKE/", "icon": "file-text", "color": "#ef4444", "createdAt": "2026-08-11T09:12:36.368326+00:00", "updatedAt": "2026-08-10T23:33:04Z" }, { "id": "github-com-juliangarnier-anime", "name": "anime", "titleZh": "Anime.js 轻量级前端动画引擎", "repo": "https://github.com/juliangarnier/anime", "homepage": "https://animejs.com/", "summary": "快速、轻量且功能完整的 JavaScript 动画库，可统一驱动 CSS 属性、SVG、DOM 属性和 JavaScript 对象动画。", "details": "Anime.js V4 提供时间线、交错、弹簧、拖拽、滚动观察、SVG 路径与形变等动画能力，采用模块化 ESM API。适合交互演示、产品页面、数据可视化和微动效；不应为普通静态页面滥用动画，仍需尊重 prefers-reduced-motion 并控制主线程开销。", "why": "API 简洁、体积轻、无需大型框架即可制作高质量 Web 动效，能直接提升本地前端项目的交互表现。", "tags": [ "前端", "JavaScript", "动画", "SVG", "CSS", "Web交互", "时间线", "MIT" ], "language": "JavaScript", "stars": 72055, "forks": 4850, "license": "MIT", "install": "npm install animejs", "compatibility": "浏览器端 ES modules；适用于原生网页及主流前端框架。", "evidence": "已核验抖音短链对应视频ID、Anime.js官方GitHub仓库、README、V4用法、许可证和实时元数据。", "metadataVerifiedAt": "2026-08-11T12:18:41.703419+00:00", "source": "抖音 戈攀AI笔记 第117期", "sourceUrl": "https://www.douyin.com/video/7672420491524705551", "video": "https://v.douyin.com/OC8ZnD3BwM8/", "icon": "sparkles", "color": "#ef476f", "createdAt": "2026-08-11T12:18:41.703419+00:00", "updatedAt": "2026-08-09T16:40:09Z" }, { "id": "github-com-prism-shadow-penguin-harness", "name": "PenguinHarness", "titleZh": "PenguinHarness 自进化 Agent 构建平台", "repo": "https://github.com/Prism-Shadow/penguin-harness", "homepage": "https://penguin.ooo/", "summary": "用自然语言一键构建可运行、可管理并能迭代改进的 Agent 应用。", "details": "PenguinHarness 通过文件系统保存任务、状态和产物，以 Harness 约束 Agent 执行流程，并支持 DeepSeek、Kimi、GPT、Claude、Gemini 等模型。适合学习 Agent 工程、RAG、工具调用、状态管理和自进化闭环。", "why": "覆盖当前 Agent 应用工程中的构建、管理、评估与迭代链路，适合作为工程实践和面试项目研究。", "tags": [ "AI", "Agent", "Harness", "RAG", "TypeScript", "自进化", "工作流", "Apache-2.0" ], "language": "TypeScript", "stars": 1174, "forks": 112, "license": "Apache-2.0", "install": "git clone https://github.com/Prism-Shadow/penguin-harness.git", "compatibility": "TypeScript / Node.js；具体运行要求以官方 README 为准。", "evidence": "已核验抖音视频ID 7671151896781999411、官方GitHub仓库、项目主页、许可证及实时GitHub元数据。", "metadataVerifiedAt": "2026-08-11T15:54:00.164003+00:00", "source": "抖音 AI算法工程师Ray", "sourceUrl": "https://www.douyin.com/video/7671151896781999411", "video": "https://v.douyin.com/sdKBIU5OImQ/", "icon": "bot", "color": "#22c55e", "createdAt": "2026-08-11T15:54:00.164003+00:00", "updatedAt": "2026-08-11T15:44:18Z" }, { "id": "github-com-huangruiteng-loopx", "name": "LoopX", "titleZh": "LoopX 长周期 AI Agent 状态控制平面", "repo": "https://github.com/huangruiteng/loopx", "homepage": "https://huangruiteng.github.io/loopx/", "summary": "为长周期、多轮次、多Agent任务保存目标、门禁、待办、证据、配额和交接状态的本地优先控制平面。", "details": "LoopX不替代Hermes、Codex、Claude Code等执行引擎，而是在它们外层维护可恢复的持久状态：目标与范围、可执行待办、人工/安全门禁、证据日志、Agent认领与租约、配额感知自动唤醒和可验证交接。适用于跨天工程、研究、实验、Issue/PR循环及多人/多Agent协作。官方所谓200+小时是项目经过的墙钟周期，并非模型连续运行200小时。", "why": "与小七AI公司“长期任务不丢、可审计、可续做”的需求高度匹配；可吸收其状态内核、证据与交接设计，同时保留Hermes作为唯一执行引擎，避免形成第二套Agent运行时。", "tags": [ "AI Agent", "长周期任务", "控制平面", "状态管理", "多Agent", "证据日志", "任务交接", "Python", "MIT" ], "language": "Python", "stars": 4108, "forks": 334, "license": "MIT", "install": "研究/试用：按官方文档安装后在项目目录执行 loopx connect；小七AI公司暂不直接引入运行依赖。", "compatibility": "Python 3.11+；支持Codex、Claude Code、Cursor、OpenCode、Pi及自定义Runner。", "evidence": "已核验抖音视频ID 7670318446169829114、官方GitHub仓库、README、MIT许可证、实时元数据及200+小时证据边界。", "metadataVerifiedAt": "2026-08-11T15:55:01.130628+00:00", "source": "抖音 沉默聊科技", "sourceUrl": "https://www.douyin.com/video/7670318446169829114", "video": "https://v.douyin.com/HFGzDXsi3sw/", "icon": "repeat", "color": "#6366f1", "createdAt": "2026-08-11T15:55:01.130628+00:00", "updatedAt": "2026-08-11T15:41:22Z" }, { "id": "github-com-opensquilla-opensquilla", "name": "OpenSquilla", "titleZh": "OpenSquilla 低成本高智能密度 AI Agent", "repo": "https://github.com/opensquilla/opensquilla", "homepage": "https://opensquilla.ai", "summary": "通过本地模型路由与任务调度，在接近OpenClaw任务得分的同时显著降低完整运行成本。", "details": "OpenSquilla 是面向桌面与自动化任务的开源 AI Agent。其 PinchBench 1.2.1 公布结果为25项任务平均分0.9251、总成本$0.688；对照OpenClaw为0.9255、$6.233。截图中“Token仅十分之一”不准确：总Token约为对照的57%，接近十分之一的是成本。", "why": "适合研究多模型路由、任务执行和成本优化；先纳入热门项目库，不与现有Hermes或小七AI公司运行时耦合。", "tags": [ "AI Agent", "模型路由", "Token优化", "成本优化", "Python", "Apache-2.0" ], "language": "Python", "stars": 6609, "forks": 523, "license": "Apache-2.0", "install": "以官方 README 为准；热门项目库仅收录，不自动安装。", "compatibility": "以官方 README 与 Release 为准。", "evidence": "已核验截图项目名、官方GitHub仓库、README基准描述、许可证及实时GitHub元数据。", "metadataVerifiedAt": "2026-08-15T19:20:27.274841+00:00", "source": "抖音 阿甘探AI", "sourceUrl": "", "video": "", "icon": "bot", "color": "#ec4899", "createdAt": "2026-08-15T19:20:27.274841+00:00", "updatedAt": "2026-08-15T19:19:58Z" }, { "id": "github-com-csyqlz-vozeb-pro", "name": "VOZEB PRO", "titleZh": "VOZEB PRO 开源 AI 创作工作台", "repo": "https://github.com/csyqlz/VOZEB-PRO", "homepage": "https://www.vozeb.com", "summary": "把统一创作 Agent、Canvas、短剧生产、素材库和商业运营后台整合在同一套 Next.js 全栈应用中。", "details": "支持文本、图片、视频和音频创作，Canvas 节点工作流、短剧剧本到 FFmpeg 成片、素材与提示词管理、模型路由、持久生成任务，以及套餐、积分、订单、支付和审计等商业后台能力。", "why": "功能覆盖完整，适合研究或自建多模态内容生产与短剧工作流；当前版本仍早期，升级前需注意官方数据库兼容说明。", "tags": [ "AI Agent", "多模态创作", "Canvas", "短剧", "Next.js", "PostgreSQL", "AGPL-3.0" ], "language": "TypeScript", "stars": 342, "forks": 122, "license": "AGPL-3.0", "install": "Docker Compose；以官方 README 为准，热门项目库仅收录，不自动安装。", "compatibility": "64位 Linux、Docker Compose v2、PostgreSQL 16；调用外部模型，无需 GPU。", "evidence": "已核验官方 GitHub 仓库、README、许可证、演示站链接及实时元数据。", "metadataVerifiedAt": "2026-08-16T19:49:20.420643+00:00", "source": "抖音 VOZEB", "sourceUrl": "", "video": "", "icon": "palette", "color": "#2563eb", "createdAt": "2026-08-16T19:49:20.420643+00:00", "updatedAt": "2026-08-14T20:14:25Z" }, { "id": "github-com-breezeblue-ai-breeze-tts", "titleZh": "Breeze TTS 2：实时可控语音生成模型", "name": "breeze-tts", "repo": "https://github.com/breezeblue-ai/breeze-tts", "homepage": "https://breezeblue.ai/breeze-tts-2", "summary": "面向实时交互的中英双语语音生成模型，支持声音克隆、自然语言声音设计、声音演绎控制和低延迟流式生成。", "details": "Breeze TTS 2 是 BreezeBlue 发布的 30 亿参数开放权重 TTS 模型。它支持根据参考音频克隆音色、无需参考音频通过自然语言设计声音、在保持说话人特征的同时控制情绪和表达方式，并提供流式 API。官方建议 eager 推理至少使用 12GB 显存；代码采用 Apache-2.0，模型权重及自托管输出仅限研究和非商业用途，商业使用需另行授权。", "why": "它把声音克隆、声音设计、表达控制和实时流式合成集中在一个模型中，适合语音助手、游戏角色、数字人和交互式叙事原型。", "tags": [ "TTS", "语音生成", "声音克隆", "声音设计", "流式推理", "中文", "英文", "PyTorch", "CUDA", "非商业模型" ], "language": "Python / PyTorch", "stars": 164, "forks": 15, "license": "代码 Apache-2.0；模型权重与自托管输出仅限研究及非商业用途", "install": "git clone https://github.com/breezeblue-ai/breeze-tts.git && cd breeze-tts && python -m pip install -r requirements.txt", "compatibility": "需要 NVIDIA CUDA GPU；eager 推理约 7.7 GiB 显存，官方建议至少 12GB；快速路径建议 24GB。支持中文和英文。", "evidence": "已核验官方 GitHub、Hugging Face 模型卡、安装说明、硬件要求及模型许可证。", "metadataVerifiedAt": "2026-08-31T08:13:21.565660Z", "source": "BreezeBlue 官方 GitHub / Hugging Face", "video": "", "icon": "microphone", "color": "", "createdAt": "2026-08-31T08:13:21.565660Z", "updatedAt": "2026-08-31T07:10:56Z" }, { "id": "github-com-sapics-ip-location-db", "titleZh": "IP Location DB：多源离线 IP 地理与 ASN 数据库", "name": "ip-location-db", "repo": "https://github.com/sapics/ip-location-db", "homepage": "https://github.com/sapics/ip-location-db", "summary": "聚合 ASN、GeoFeed、Whois、iptoasn.com、DB-IP Lite 与 GeoLite2，持续生成可下载的离线 IP 国家、城市和 ASN 数据库。", "details": "项目通过 GitHub Actions 定期整合多个公开数据源，提供 CIDR、CSV、MMDB 等格式，适合 IP 归属、国家/城市识别和 ASN 查询。不同上游的许可证、精度和更新频率各异，风控或计费场景应保留来源与置信边界，不把 IP 定位当作精确地址。", "why": "无需在线第三方 API 即可接入 IP 国家、城市与 ASN 查询，适合订阅网关、机场节点审计、日志分析和离线服务。", "tags": [ "IP 地理位置", "GeoIP", "ASN", "离线数据库", "MMDB", "CSV", "GeoFeed", "Whois", "网络运维", "asn", "autonomous-system", "geoip", "ip-country", "ip-geolocation", "ip-location" ], "language": "JavaScript / Shell", "stars": 2165, "forks": 204, "license": "多数据源，按各上游许可证", "install": "直接从仓库 release 分支或对应 raw 路径下载所需数据库文件；以官方 README 为准。", "compatibility": "提供多种文本和 MMDB 格式，可由常见 GeoIP/MMDB 库或标准 CSV 工具读取。", "evidence": "已核验 GitHub 官方仓库、项目定位、数据源、格式、维护状态及实时 GitHub API 元数据。", "metadataVerifiedAt": "2026-09-01T21:36:30.890310Z", "source": "用户推荐", "video": "", "icon": "globe", "color": "", "createdAt": "2026-09-01T21:36:30.890310Z", "updatedAt": "2026-09-01T00:40:05Z" } ] }