[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-HKUDS--OpenSpace":3,"tool-HKUDS--OpenSpace":62},[4,18,28,36,45,54],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":17},4358,"openclaw","openclaw\u002Fopenclaw","OpenClaw 是一款专为个人打造的本地化 AI 助手，旨在让你在自己的设备上拥有完全可控的智能伙伴。它打破了传统 AI 助手局限于特定网页或应用的束缚，能够直接接入你日常使用的各类通讯渠道，包括微信、WhatsApp、Telegram、Discord、iMessage 等数十种平台。无论你在哪个聊天软件中发送消息，OpenClaw 都能即时响应，甚至支持在 macOS、iOS 和 Android 设备上进行语音交互，并提供实时的画布渲染功能供你操控。\n\n这款工具主要解决了用户对数据隐私、响应速度以及“始终在线”体验的需求。通过将 AI 部署在本地，用户无需依赖云端服务即可享受快速、私密的智能辅助，真正实现了“你的数据，你做主”。其独特的技术亮点在于强大的网关架构，将控制平面与核心助手分离，确保跨平台通信的流畅性与扩展性。\n\nOpenClaw 非常适合希望构建个性化工作流的技术爱好者、开发者，以及注重隐私保护且不愿被单一生态绑定的普通用户。只要具备基础的终端操作能力（支持 macOS、Linux 及 Windows WSL2），即可通过简单的命令行引导完成部署。如果你渴望拥有一个懂你",349277,3,"2026-04-06T06:32:30",[13,14,15,16],"Agent","开发框架","图像","数据工具","ready",{"id":19,"name":20,"github_repo":21,"description_zh":22,"stars":23,"difficulty_score":24,"last_commit_at":25,"category_tags":26,"status":17},9989,"n8n","n8n-io\u002Fn8n","n8n 是一款面向技术团队的公平代码（fair-code）工作流自动化平台，旨在让用户在享受低代码快速构建便利的同时，保留编写自定义代码的灵活性。它主要解决了传统自动化工具要么过于封闭难以扩展、要么完全依赖手写代码效率低下的痛点，帮助用户轻松连接 400 多种应用与服务，实现复杂业务流程的自动化。\n\nn8n 特别适合开发者、工程师以及具备一定技术背景的业务人员使用。其核心亮点在于“按需编码”：既可以通过直观的可视化界面拖拽节点搭建流程，也能随时插入 JavaScript 或 Python 代码、调用 npm 包来处理复杂逻辑。此外，n8n 原生集成了基于 LangChain 的 AI 能力，支持用户利用自有数据和模型构建智能体工作流。在部署方面，n8n 提供极高的自由度，支持完全自托管以保障数据隐私和控制权，也提供云端服务选项。凭借活跃的社区生态和数百个现成模板，n8n 让构建强大且可控的自动化系统变得简单高效。",184740,2,"2026-04-19T23:22:26",[16,14,13,15,27],"插件",{"id":29,"name":30,"github_repo":31,"description_zh":32,"stars":33,"difficulty_score":10,"last_commit_at":34,"category_tags":35,"status":17},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,"2026-04-05T11:01:52",[14,15,13],{"id":37,"name":38,"github_repo":39,"description_zh":40,"stars":41,"difficulty_score":24,"last_commit_at":42,"category_tags":43,"status":17},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",161147,"2026-04-19T23:31:47",[14,13,44],"语言模型",{"id":46,"name":47,"github_repo":48,"description_zh":49,"stars":50,"difficulty_score":51,"last_commit_at":52,"category_tags":53,"status":17},8272,"opencode","anomalyco\u002Fopencode","OpenCode 是一款开源的 AI 编程助手（Coding Agent），旨在像一位智能搭档一样融入您的开发流程。它不仅仅是一个代码补全插件，而是一个能够理解项目上下文、自主规划任务并执行复杂编码操作的智能体。无论是生成全新功能、重构现有代码，还是排查难以定位的 Bug，OpenCode 都能通过自然语言交互高效完成，显著减少开发者在重复性劳动和上下文切换上的时间消耗。\n\n这款工具专为软件开发者、工程师及技术研究人员设计，特别适合希望利用大模型能力来提升编码效率、加速原型开发或处理遗留代码维护的专业人群。其核心亮点在于完全开源的架构，这意味着用户可以审查代码逻辑、自定义行为策略，甚至私有化部署以保障数据安全，彻底打破了传统闭源 AI 助手的“黑盒”限制。\n\n在技术体验上，OpenCode 提供了灵活的终端界面（Terminal UI）和正在测试中的桌面应用程序，支持 macOS、Windows 及 Linux 全平台。它兼容多种包管理工具，安装便捷，并能无缝集成到现有的开发环境中。无论您是追求极致控制权的资深极客，还是渴望提升产出的独立开发者，OpenCode 都提供了一个透明、可信",144296,1,"2026-04-16T14:50:03",[13,27],{"id":55,"name":56,"github_repo":57,"description_zh":58,"stars":59,"difficulty_score":24,"last_commit_at":60,"category_tags":61,"status":17},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",109154,"2026-04-18T11:18:24",[14,15,13],{"id":63,"github_repo":64,"name":65,"description_en":66,"description_zh":67,"ai_summary_zh":67,"readme_en":68,"readme_zh":69,"quickstart_zh":70,"use_case_zh":71,"hero_image_url":72,"owner_login":73,"owner_name":74,"owner_avatar_url":75,"owner_bio":76,"owner_company":76,"owner_location":76,"owner_email":76,"owner_twitter":76,"owner_website":77,"owner_url":78,"languages":79,"stars":103,"forks":104,"last_commit_at":105,"license":106,"difficulty_score":24,"env_os":107,"env_gpu":108,"env_ram":108,"env_deps":109,"category_tags":115,"github_topics":76,"view_count":24,"oss_zip_url":76,"oss_zip_packed_at":76,"status":17,"created_at":116,"updated_at":117,"faqs":118,"releases":154},9871,"HKUDS\u002FOpenSpace","OpenSpace","\"OpenSpace: Make Your Agents: Smarter, Low-Cost, Self-Evolving\" -- Community: https:\u002F\u002Fopen-space.cloud\u002F","OpenSpace 是一个旨在让 AI 智能体变得更聪明、更低成本且具备自我进化能力的开源框架。它核心解决了当前 AI 智能体开发中 Token 消耗过高、技能复用困难以及缺乏持续成长机制的痛点。通过“一次命令，全员进化”的理念，OpenSpace 能够统一管理并优化包括 Claude Code、Codex、OpenClaw 等多种主流智能体，显著降低运行成本（官方数据显示可减少 46% 的 Token 用量）。\n\n该平台特别适合 AI 开发者、研究人员以及希望构建自动化工作流的技术团队使用。其独特的技术亮点在于构建了去中心化的“技能商店”与共享机制：智能体在执行任务中积累的高质量技能可被自动提取、评估并持久化存储，供其他智能体学习调用，从而实现群体智慧的自我迭代。此外，OpenSpace 支持多通道通信网关（如 WhatsApp、飞书），允许智能体跨平台接收指令，并提供灵活的 MCP 服务部署方案，轻松突破本地运行限制。无论是想要降低大模型应用成本，还是致力于探索智能体自主进化的开发者，OpenSpace 都提供了一个高效、开放的基础设施。","\u003Cdiv align=\"center\">\n\n\u003Cpicture>\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_4aa940ab5ed8.png\" width=\"320px\" style=\"border: none; box-shadow: none;\" alt=\"OpenSpace Logo\">\n\u003C\u002Fpicture>\n\n## ✨ OpenSpace: Make Your Agents: Smarter, Low-Cost, Self-Evolving ✨\n\n| 🔋 **46% Fewer Tokens** | **💰 $11K earned in 6 Hours** | 🧬 **Self-Evolving Skills** | 🌐 **Agents Experience Sharing** |\n\n[![Agents](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FAgents-Claude_Code%20%7C%20Codex%20%7C%20OpenClaw%20%7C%20nanobot%20%7C%20...-99C9BF.svg)](https:\u002F\u002Fmodelcontextprotocol.io\u002F)\n[![Python](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.12+-FCE7D6.svg)](https:\u002F\u002Fwww.python.org\u002F)\n[![License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-C1E5F5.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT\u002F)\n[![Feishu](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FFeishu-Group-E9DBFC?style=flat&logo=larksuite&logoColor=white)](.\u002FCOMMUNICATION.md)\n[![WeChat](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FWeChat-Group-C5EAB4?style=flat&logo=wechat&logoColor=white)](.\u002FCOMMUNICATION.md)\n[![中文文档](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F文档-中文版-F5C6C6?style=flat)](.\u002FREADME_CN.md)\n\n**One Command to Evolve All Your AI Agents**: OpenClaw, nanobot, Claude Code, Codex, Cursor and etc.\n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_82c25fc4d16b.gif\" width=\"500px\" alt=\"openspace --query your task\">\n\n\u003C\u002Fdiv>\n\n---\n\n## 📢 News\n\n- **2026-04-16** 📊 **Evolution candidate lifecycle tracking** — skill store now records when evolution suggestions are processed (`evolution_processed_at`), cleanly distinguishing pending candidates from already-handled ones.\n- **2026-04-12** 🍎 **macOS platform hardening** — decoupled `atomacos` from core macOS imports so screenshots, window control, and recording work independently without it.\n- **2026-04-10** 🎯 **CAPTURED skills** now persist to the host agent's own skill directory instead of the default registry path. Cloud skill uploads now support **private visibility** correctly.\n- **2026-04-09** 💬 Multi-channel **communication gateway**. OpenSpace can now receive and respond to messages from external platforms. Ships with **WhatsApp** (Baileys bridge + QR auth) and **Feishu** (HTTP webhook) adapters, session management, attachment caching, and allowlist-based access control. See [`openspace\u002Fconfig\u002FREADME.md`](openspace\u002Fconfig\u002FREADME.md) for setup.\n- **2026-04-07** 🌐 OpenSpace MCP now supports standalone **SSE** and **streamable HTTP** startup, making it easier for remote hosts to connect over HTTP instead of stdio and bypass stdio-bound MCP server timeout bottlenecks. See the [host integration guide](openspace\u002Fhost_skills\u002FREADME.md) for setup details.\n- **2026-04-06** 🛠️ Fixed multiple runtime issues across grounding, MCP serving, skill evolution, and persistence, improving execution stability and recovery in long-running workflows.\n- **2026-04-05** 🧭 Cleaned up LLM credential resolution: centralized `.env` loading, improved host config auto-detection, and made provider-native env handling more consistent.\n- **2026-04-03** 🚀 Released **v0.1.0** — Skill quality monitoring: structural patterns extracted from high-quality skills now evaluate every new submission daily. Faster, more relevant cloud search. Production-grade vertical skill clusters emerging organically from the community. Frontend now supports Chinese (zh) i18n.\n- **2026-04-02** ⚡ Cloud search upgraded for higher relevance and lower latency.\n- **2026-03-31** 🛡️ Security hardening: hardened zip extraction and `import_skill` against path traversal. CLI now respects `OPENSPACE_MODEL` and `OPENSPACE_LLM_*` env vars; MiniMax compatibility; workflow ID collision fixes.\n- **2026-03-29** 🔒 Pinned litellm to \u003C1.82.7 to avoid PYSEC-2026-2 supply-chain attack.\n- **2026-03-28** 🔧 Idempotent skill registration — `register_skill_dir` now returns existing `SkillMeta` for already-registered skills. Updated OpenClaw setup docs.\n- **2026-03-27** 🪟 Fixed stdio deadlock on Windows; improved evolver confirmation parsing with stem-style keyword matching.\n- **2026-03-26** 🌱 Dynamic skill directory re-scanning on each call, lightweight local skill search, and streamlined documentation.\n- **2026-03-25** 🎉 OpenSpace is now open source!\n\n---\n\n## The Problem with Today's AI Agents\n\nToday's AI agents — [OpenClaw](https:\u002F\u002Fgithub.com\u002Fopenclaw\u002Fopenclaw), [nanobot](https:\u002F\u002Fgithub.com\u002FHKUDS\u002Fnanobot), [Claude Code](https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Fdocs\u002Fclaude-code), [Codex](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fcodex), [Cursor](https:\u002F\u002Fcursor.com), etc. — are powerful, but they have a critical weakness: they never **Learn**, **Adapt**, and **Evolve** from real-world experience — let alone **Share** with each other.\n- **❌ Massive Token Waste** - How to reuse successful task patterns instead of reasoning from scratch and burning tokens every time?\n- **❌ Repeated Costly Failures** - How to share solutions across agents instead of repeating the same costly exploration and mistakes?\n- **❌ Poor and Unreliable Skills** - How to maintain skill reliability as tools and APIs evolve — while ensuring community-contributed skills meet rigorous quality standards?\n\n## 🎯 What is OpenSpace?\n\n**🚀 🚀 The self-evolving engine where every task makes every agent smarter and more cost-efficient.**\n\nhttps:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fc50f70ab-f6db-47bf-9498-3210c0f0abae\n\nOpenSpace plugs into any agent as skills and evolves it with three superpowers:\n\n### 🧬 Self-Evolution\nSkills that learn and improve themselves automatically\n- ✅ **AUTO-FIX** — When a skill breaks, it fixes itself instantly\n- ✅ **AUTO-IMPROVE** — Successful patterns become better skill versions\n- ✅ **AUTO-LEARN** — Captures winning workflows from actual usage\n- ✅ **Quality monitoring** — Tracks skill performance, error rates, and execution success across all tasks.\n\n**Skills that continuously evolve — turning every failure into improvement, every success into optimization.**\n\n### 🌐 Collective Agent Intelligence\nTurn individual agents into a shared brain\n- ✅ **Shared evolution**: One agent's improvement becomes every agent's upgrade\n- ✅ **Network effects**: More agents → richer data → faster evolution for every agent\n- ✅ **Easy sharing** — Upload and download evolved skills with one simple command\n- ✅ **Access control** — Choose public, private, or team-only access for each skill\n\n**One agent learns, all agents benefit — collective intelligence at scale.**\n\n### 💰 Token Efficiency\nSmarter agents, dramatically lower costs\n- ✅ **Stop repeating work** → Reuse successful solutions instead of starting from zero each time\n- ✅ **Tasks get cheaper** → As skills improve, similar work costs less and less\n- ✅ **Small updates only** → Fix what's broken, don't rebuild everything\n- ✅ **Real savings**: 4.2× better performance with 46% fewer tokens on real-world tasks, delivering measurable economic value. ([GDPVal](#-benchmark-gdpval))\n\nDo more, spend less — agents that actually save you money over time.\n\n---\n\n### The Difference\n\n**❌ Current Agents**\n- Skills degrade silently as tools evolve\n- Failed patterns repeat with no learning mechanism\n- Knowledge remains trapped in individual agents\n\n**✅ OpenSpace-Powered Agents**\n- Multi-layer monitoring catches problems and auto-triggers repairs\n- Successful workflows become reusable, shareable skills\n- When one agent learns something useful, all agents get that knowledge instantly\n\n### 📊 OpenSpace: Turn Your Agent into a Money-Making Coworker\n\n**🎯 Real-World Results That Matter**\nOn 50 professional tasks (**📈 [GDPVal Economic Benchmark](#-benchmark-gdpval)**) across 6 industries, OpenSpace agents earn **4.2× more money** than baseline ([ClawWork](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FClawWork)) agents using the same backbone LLM (Qwen 3.5-Plus). While cutting 46% of costly tokens through skill evolution.\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_18c77ed6baef.png\" width=\"100%\" alt=\"GDPVal Benchmark — Key Results\" \u002F>\n\u003C\u002Fdiv>\n\n**💼 These Aren't Toy Problems**\n- Building payroll calculators from complex union contracts\n- Preparing tax returns from 15 scattered PDF documents\n- Drafting legal memoranda on California privacy regulations\n- Creating compliance forms and engineering specifications\n\n**📈 Consistent Wins Across All Fields**\n- Compliance work: +18.5% higher earnings\n- Engineering projects: +8.7% better performance\n- Professional documents: 56% fewer tokens needed\n- Every category improved — no exceptions\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_73b01c7047c9.png\" width=\"100%\" alt=\"GDPVal Benchmark — Task Showcase by Category\" \u002F>\n\u003C\u002Fdiv>\n\n**OpenSpace doesn't just make agents smarter** — it makes them economically viable. Real work, real money, measurable results.\n\n## Use Case for Autonomous System Development with OpenSpace\n\n**🖥️ [My Daily Monitor](showcase\u002FREADME.md)** — OpenSpace empowers your agent to complete large-scale system development. This personal behavior monitoring system with 20+ live dashboard panels was built entirely by the agent — 60+ skills evolved from scratch through OpenSpace, demonstrating autonomous end-to-end software development capabilities.\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_6c1a26859905.png\" width=\"100%\" alt=\"My Daily Monitor – Dark Mode\" \u002F>\n\u003C\u002Fdiv>\n\n---\n\n## 📋 Table of Contents\n\n- [⚡ Quick Start](#-quick-start)\n  - [🤖 Path A: For Your Agent](#-path-a-for-your-agent)\n  - [👤 Path B: As Your Co-Worker](#-path-b-as-your-co-worker)\n  - [📊 Local Dashboard](#-local-dashboard)\n- [📈 Benchmark: GDPVal](#-benchmark-gdpval)\n- [📊 Showcase: My Daily Monitor](#-showcase-my-daily-monitor)\n- [🏗️ Framework](#️-framework)\n  - [🧬 Self-Evolution Engine](#-self-evolution-engine)\n  - [🌐 Cloud Skill Community](#-cloud-skill-community)\n- [🔧 Advanced Configuration](#-advanced-configuration)\n- [📖 Code Structure](#-code-structure)\n- [🤝 Contribute & Roadmap](#-contribute--roadmap)\n- [🔗 Related Projects](#-related-projects)\n\n---\n\n## ⚡ Quick Start\n\n🌐 **Just want to explore?** Browse community skills, evolution lineage at **[open-space.cloud](https:\u002F\u002Fopen-space.cloud)** — no installation needed.\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace.git && cd OpenSpace\npip install -e .\nopenspace-mcp --help   # verify installation\n```\n\n> [!TIP]\n> **Slow clone?** The `assets\u002F` folder (~50 MB of images) makes the default clone large. Use this lightweight alternative to skip it:\n> ```bash\n> git clone --filter=blob:none --sparse https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace.git\n> cd OpenSpace\n> git sparse-checkout set '\u002F*' '!assets\u002F'\n> pip install -e .\n> ```\n\n**Choose your path:**\n- **[Path A](#-path-a-for-your-agent)** — Plug OpenSpace into your agent\n- **[Path B](#-path-b-as-your-co-worker)** — Use OpenSpace directly as your AI co-worker\n\n### 🤖 Path A: For Your Agent\n\nWorks with any agent that supports skills (`SKILL.md`) — [Claude Code](https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Fdocs\u002Fclaude-code), [Codex](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fcodex), [OpenClaw](https:\u002F\u002Fgithub.com\u002Fopenclaw\u002Fopenclaw), [nanobot](https:\u002F\u002Fgithub.com\u002FHKUDS\u002Fnanobot), etc.\n\n**① Add OpenSpace to your agent's MCP config:**\n\n```json\n{\n  \"mcpServers\": {\n    \"openspace\": {\n      \"command\": \"openspace-mcp\",\n      \"toolTimeout\": 600,\n      \"env\": {\n        \"OPENSPACE_HOST_SKILL_DIRS\": \"\u002Fpath\u002Fto\u002Fyour\u002Fagent\u002Fskills\",\n        \"OPENSPACE_WORKSPACE\": \"\u002Fpath\u002Fto\u002FOpenSpace\",\n        \"OPENSPACE_API_KEY\": \"sk-xxx (optional, for cloud)\"\n      }\n    }\n  }\n}\n```\n\n> [!TIP]\n> Credentials (API key, model) are **auto-detected** from your agent's config; you usually don't need to set them manually.\n\n> [!NOTE]\n> OpenSpace supports 3 launch modes:\n> - **stdio**: keep `command: \"openspace-mcp\"` in the host config.\n> - **SSE**: start `openspace-mcp --transport sse --host 127.0.0.1 --port 8080`.\n> - **streamable HTTP**: start `openspace-mcp --transport streamable-http --host 127.0.0.1 --port 8081`.\n>\n> Common remote endpoints:\n> - SSE endpoint: `http:\u002F\u002F127.0.0.1:8080\u002Fsse`\n> - streamable HTTP endpoint: `http:\u002F\u002F127.0.0.1:8081\u002Fmcp`\n>\n> `stdio` is the simplest option. HTTP modes keep OpenSpace as a standalone server, but **host-specific registration syntax** and **host-side timeouts** still apply.\n\n**② Copy skills** into your agent's skills directory:\n\n```bash\ncp -r OpenSpace\u002Fopenspace\u002Fhost_skills\u002Fdelegate-task\u002F \u002Fpath\u002Fto\u002Fyour\u002Fagent\u002Fskills\u002F\ncp -r OpenSpace\u002Fopenspace\u002Fhost_skills\u002Fskill-discovery\u002F \u002Fpath\u002Fto\u002Fyour\u002Fagent\u002Fskills\u002F\n```\n\nDone. These two skills teach your agent when and how to use OpenSpace — no additional prompting needed. Your agent can now self-evolve skills, execute complex tasks, and access the cloud skill community. You can also add your own custom skills — see [`openspace\u002Fskills\u002FREADME.md`](openspace\u002Fskills\u002FREADME.md).\n\n> [!NOTE]\n> **Cloud community (optional):** Register at **[open-space.cloud](https:\u002F\u002Fopen-space.cloud)** to get a `OPENSPACE_API_KEY`, then add it to the `env` block above. Without it, all local capabilities (task execution, evolution, local skill search) work normally.\n\n📖 Per-agent config (OpenClaw \u002F nanobot), all env vars, advanced settings: [`openspace\u002Fhost_skills\u002FREADME.md`](openspace\u002Fhost_skills\u002FREADME.md)\n\n### 👤 Path B: As Your Co-Worker\n\nUse OpenSpace directly — coding, search, tool use, and more — with self-evolving skills and cloud community built in.\n\n> [!NOTE]\n> Create a `.env` file with your LLM API key and optionally `OPENSPACE_API_KEY` for cloud community access (refer to [`openspace\u002F.env.example`](openspace\u002F.env.example)).\n\n```bash\n# Interactive mode\nopenspace\n\n# Execute task\nopenspace --model \"anthropic\u002Fclaude-sonnet-4-5\" --query \"Create a monitoring dashboard for my Docker containers\"\n```\n\nAdd your own custom skills: [`openspace\u002Fskills\u002FREADME.md`](openspace\u002Fskills\u002FREADME.md).\n\n**Cloud CLI** — manage skills from the command line:\n\n```bash\nopenspace-download-skill \u003Cskill_id>         # download a skill from the cloud\nopenspace-upload-skill \u002Fpath\u002Fto\u002Fskill\u002Fdir   # upload a skill to the cloud\n```\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>Python API\u003C\u002Fb>\u003C\u002Fsummary>\n\n```python\nimport asyncio\nfrom openspace import OpenSpace\n\nasync def main():\n    async with OpenSpace() as cs:\n        result = await cs.execute(\"Analyze GitHub trending repos and create a report\")\n        print(result[\"response\"])\n\n        for skill in result.get(\"evolved_skills\", []):\n            print(f\"  Evolved: {skill['name']} ({skill['origin']})\")\n\nasyncio.run(main())\n```\n\n\u003C\u002Fdetails>\n\n### 📊 Local Dashboard\n\nSee how your skills evolve — browse skills, track lineage, compare diffs.\n\n> Requires **Node.js ≥ 20**.\n\n```bash\n# Terminal 1. Start backend API\nopenspace-dashboard --port 7788\n\n# Terminal 2: Start frontend dev server\ncd frontend\nnpm install        # only needed once\nnpm run dev    \n```\n\n📖 **Frontend setup guide**: [`frontend\u002FREADME.md`](frontend\u002FREADME.md)\n\n\u003Cdiv align=\"center\">\n\u003Ctable>\n\u003Ctr>\n\u003Ctd width=\"50%\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_4aec6a4c7adf.gif\" width=\"100%\" alt=\"Skill Classes\" \u002F>\u003C\u002Ftd>\n\u003Ctd width=\"50%\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_8f6b3d5c77f4.gif\" width=\"100%\" alt=\"Cloud Skill Records\" \u002F>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"center\">\u003Csub>Skill Classes — Browse, Search & Sort\u003C\u002Fsub>\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Csub>Cloud — Browse & Discover Skill Records\u003C\u002Fsub>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd width=\"50%\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_dce23db38fa5.gif\" width=\"100%\" alt=\"Version Lineage\" \u002F>\u003C\u002Ftd>\n\u003Ctd width=\"50%\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_ee8537c24a3a.gif\" width=\"100%\" alt=\"Workflow Sessions\" \u002F>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"center\">\u003Csub>Version Lineage — Skill Evolution Graph\u003C\u002Fsub>\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Csub>Workflow Sessions — Execution History & Metrics\u003C\u002Fsub>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\n---\n\n## 📈 Benchmark: GDPVal\n\nWe evaluate OpenSpace on [GDPVal](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fopenai\u002Fgdpval) — 220 real-world professional tasks spanning 44 occupations — using the [ClawWork](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FClawWork) evaluation protocol with identical productivity tools and LLM-based scoring. Our two-phase design (Cold Start → Warm Rerun) demonstrates how accumulated skills reduce token consumption over time.\n\nFair Benchmark: OpenSpace uses Qwen 3.5-Plus as its backbone LLM — identical to a ClawWork baseline agent — ensuring that performance differences stem purely from skill evolution, not model capabilities.\n\nReal Economic Value: Tasks range from building payroll calculators to preparing tax returns to drafting legal memoranda — the same professional work that generates actual GDP, evaluated on both quality and cost efficiency.\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_45fd6021e81b.png\" width=\"100%\" alt=\"GDPVal Benchmark — Income Comparison\" \u002F>\n\u003C\u002Fdiv>\n\n- **4.2× Higher Income** vs ClawWork with the same backbone LLM (Qwen 3.5-Plus)\n- **72.8% Value Capture** — $11,484 earned out of $15,764 task value, outperforming all agents\n- **70.8% Average Quality** — +30pp above the best ClawWork agent (40.8%)\n− **45.9% Token Usage** in Phase 2 vs Phase 1 — better results with dramatically lower costs\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_d3c2a3500c88.png\" width=\"100%\" alt=\"GDPVal Benchmark — Quality & Token Efficiency\" \u002F>\n\u003C\u002Fdiv>\n\n### What Real-World Tasks Can OpenSpace Handle?\n\nThe 50 GDPVal tasks span 6 real-world work categories. \n- **Phase 1 (Cold Start)** runs all 50 tasks sequentially — skills accumulate in a shared database as each task completes.\n- **Phase 2 (Warm Rerun)** re-executes the same 50 tasks with the full evolved skill database from Phase 1.\n\nIncome Capture = actual payment earned ÷ maximum possible task value\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_73b01c7047c9.png\" width=\"100%\" alt=\"GDPVal Benchmark — Task Showcase by Category\" \u002F>\n\u003C\u002Fdiv>\n\n## 🎯 Where Evolution Delivers Maximum Impact — And Why:\n\n| Category | Income Δ | Token Δ | Why |\n|---|---|---|---|\n| **📝 Documents & Correspondence** (7) | 71→74% (+3.3pp) | −56% | Polished formal output — California privacy law memoranda, surveillance investigation reports, child support case reports. The `document-gen-fallback` skill family evolved through 13 versions, making structure and error recovery near-automatic. |\n| **📋 Compliance & Form** (11) | 51→70% (+18.5pp) | −51% | Structured PDFs — tax returns from 15 source documents, pharmacy compliance checklists, clinical handoff templates. The PDF skill chain (checklist logic → reportlab layout → verification) evolves once, then all form tasks reuse the full pipeline. |\n| **🎬 Media Production** (3) | 53→58% (+5.8pp) | −46% | Audio\u002Fvideo via Python and ffmpeg — bossa-nova instrumental from drum reference, bass stem editing from 5 tracks, CGI show reel from 13 source videos. Evolved skills encode working ffmpeg flags and codec fallbacks, eliminating sandbox trial-and-error. |\n| **🛠️ Engineering** (4) | 70→78% (+8.7pp) | −43% | Multi-deliverable technical projects — Web3 full-stack (Solidity + React + tests), CNC workcell safety system (report + layout + hardware table), aerospace CFD report. Coordination skills transfer universally across these diverse tasks. |\n| **📊 Spreadsheets** (15) | 63→70% (+7.3pp) | −37% | Functional .xlsx tools — payroll calculators from union contracts, sales forecasts from historical data, pricing models with competitor benchmarking. Spreadsheet patterns (formulas, merged cells, validation) are identical across domains. |\n| **📈 Strategy & Analysis** (10) | 88→89% (+1.0pp) | −32% | Strategic recommendations — supplier negotiation strategies, nonprofit program evaluations, energy trading analysis for a $300M desk. Already highest quality (88%); savings from reusing document structure and multi-file orchestration. |\n\n### What Did Evolution Produce? (165 Skills)\n\nAcross 50 Phase 1 tasks, OpenSpace autonomously evolved **165 skills**. The breakthrough insight: these aren't just domain knowledge — they're **resilient execution patterns** and **quality assurance workflows**. The agent learned how to reliably deliver results in an imperfect, real-world environment.\n\n**Key Discovery**: Most skills focus on tool reliability and error recovery, not task-specific knowledge.\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_8bfba90ef327.png\" width=\"100%\" alt=\"GDPVal Benchmark — Evolved Skill Taxonomy\" \u002F>\n\u003C\u002Fdiv>\n\n| Purpose | Count | What It Teaches the Agent |\n|---|---|---|\n| **File Format I\u002FO** | 44 | PDF extraction fallbacks, DOCX parsing, Excel merged-cell handling, PPTX creation. 32\u002F44 *captured* from real failures — each one is a production bug solved. |\n| **Execution Recovery** | 29 | Layered fallback: sandbox fails → shell → file-write-then-run → heredoc. 28\u002F29 *captured* from actual crashes. The foundation that makes everything else reliable. |\n| **Document Generation** | 26 | End-to-end doc pipeline. `document-gen-fallback` evolved from 1 imported skill into **13 derived versions** — the most deeply iterated skill family. |\n| **Quality Assurance** | 23 | Post-write verification: check Excel row counts, validate PDF pages, proof-gate spreadsheet formulas. Why P2 quality improves — the agent *verifies*, not just produces. |\n| **Task Orchestration** | 17 | Multi-file tracking, ZIP packaging, zero-iteration failure detection. Meta-skills that help across all task types with multiple deliverables. |\n| **Domain Workflow** | 13 | SOAP notes, audio production (**4 generations** from 1 template), video pipelines. Small count but deep evolution within each domain. |\n| **Web & Research** | 11 | SSL\u002Fproxy debugging, search fallbacks, JS-heavy page handling. Includes 2 *fixed* skills — web access is inherently unstable. |\n\n**Reproduce experiments, analysis tools, and results**: [`gdpval_bench\u002FREADME.md`](gdpval_bench\u002FREADME.md)\n\n---\n\n## 📊 Showcase: My Daily Monitor\n\n> **Zero human code was written.** 60+ skills evolved from scratch to build a fully working live dashboard.\n\n**My Daily Monitor** is an always-on dashboard streaming processes, servers, news, markets, email, and schedules — with a built-in AI agent.\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_a10f865e4476.png\" width=\"90%\" alt=\"My Daily Monitor – Light Mode\" \u002F>\n\u003C\u002Fdiv>\n\n### How OpenSpace Built It (From Zero)\n\n| Phase | What Happened | Skills |\n|-------|--------------|--------|\n| 🌱 **Seed** | Analyzed open-source [WorldMonitor](https:\u002F\u002Fgithub.com\u002Fkoala73\u002Fworldmonitor), extracted reference patterns | 6 initial skills |\n| 🏗️ **Scaffold** | Generated project structure, Vite config, TypeScript setup | +8 skills |\n| 🎨 **Build** | Created 20+ panels with data services, API routes, grid layout | +25 skills |\n| 🔧 **Fix** | Auto-repaired broken TypeScript, API mismatches, CSS conflicts | +12 FIX evolutions |\n| 🧬 **Evolve** | Derived enhanced patterns, merged complementary skills | +15 DERIVED skills |\n| 📦 **Capture** | Extracted reusable patterns from successful executions | +8 CAPTURED skills |\n\n### 📈 Skill Evolution Graph\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_7431bb292cac.png\" width=\"90%\" alt=\"Skill Evolution Graph\" \u002F>\n\u003C\u002Fdiv>\n\n> Each node is a skill that OpenSpace learned, extracted, or refined. The full evolution history is open-sourced in [`showcase\u002F.openspace\u002Fopenspace.db`](showcase\u002F.openspace\u002Fopenspace.db) — load it in any SQLite browser to explore lineage, diffs, and quality metrics.\n\n**Full details**: [`showcase\u002FREADME.md`](showcase\u002FREADME.md)\n\n---\n\n## 🏗️ OpenSpace's Framework\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_77235adca28f.png\" width=\"90%\" alt=\"OpenSpace Framework\" \u002F>\n\u003C\u002Fdiv>\n\n### 🧬 Self-Evolution Engine\n\nThe core of OpenSpace. Skills aren't static files — they're living entities that automatically select, apply, monitor, analyze, and evolve themselves.\n\n#### 🔄 Autonomous & Continuous Evolution\n\n- **Full Lifecycle Management**: From discovery to application to evolution — all without human intervention. OpenSpace completes tasks regardless of whether matching skills exist.\n\n**Three Evolution Modes**:\n- 🔧 FIX — Repair broken or outdated instructions in-place. Same skill, new version.\n- 🚀 DERIVED — Create enhanced or specialized versions from parent skills. New skill directory, coexists with parents.\n- ✨ CAPTURED — Extract novel reusable patterns from successful executions. Brand new skill, no parent.\n\n**Three Independent Triggers**: Multiple lines of defense against skill degradation — both successful and failed executions drive evolution.\n- **📈 Post-Execution Analysis** — Runs after every task. Analyzes full recordings and suggests FIX\u002FDERIVED\u002FCAPTURED for involved skills.\n- **⚠️ Tool Degradation** — When tool success rates drop, quality monitor finds all dependent skills and batch-evolves them.\n- **📊 Metric Monitor** — Periodically scans skill health metrics (applied rate, completion rate, fallback rate) and evolves underperformers.\n\n#### 📊 Full-Stack Quality Monitoring\nMulti-Layer Tracking: Quality monitoring covers the entire execution stack — from high-level workflows to individual tool calls:\n- **🎯 Skills** — applied rate, completion rate, effective rate, fallback rate\n- **🔨 Tool Calls** — success rate, latency, flagged issues\n- **⚡ Code Execution** — execution status, error patterns\n\n**Cascade Evolution**: When any component degrades — skill workflow or single tool call — evolution automatically triggers for all upstream dependent skills, maintaining system-wide coherence.\n\n#### 🔧 Intelligent & Safe Evolution\n**🤖 Autonomous Evolution**: Each evolution explores the codebase, discovers root causes, and decides fixes autonomously — gathering real evidence before making changes, not generating blindly.\n\n**⚡ Diff-Based & Token-Efficient**: Produces minimal, targeted diffs rather than full rewrites, with automatic retry on failure. Every version stored in a version DAG with full lineage tracking.\n\n**🛡️ Built-in Safeguards**:\n- Confirmation gates reduce false-positive triggers\n- Anti-loop guards prevent runaway evolution cycles\n- Safety checks flag dangerous patterns (prompt injection, credential exfiltration)\n- Evolved skills are validated before replacing predecessors\n\n**🌐 Collaborative Skill Community**\nA collaborative registry where agents share evolved skills. When one agent evolves an improvement, every connected agent can discover, import, and build on it — turning individual progress into collective intelligence.\n\n- **🔐 Flexible Sharing**: Share skills publicly, within groups, or keep them private. Smart search finds what you need and auto-imports it. Every evolution is lineage-tracked with full diffs.\n\n- **☁️ Collaborative Platform**: open-space.cloud — register for an API key, browse community skills, and manage your groups.\n\n---\n\n## 🔧 Advanced Configuration\n\nFor most users, [Quick Start](#-quick-start) is all you need. For advanced options (environment variables, execution modes, security policies, etc.), see [`openspace\u002Fconfig\u002FREADME.md`](openspace\u002Fconfig\u002FREADME.md).\n\n---\n\n\u003Ca id=\"-code-structure\">\u003C\u002Fa>\n\u003Cdetails>\n\u003Csummary>\u003Cb>📖 Code Structure\u003C\u002Fb>\u003C\u002Fsummary>\n\n> **Legend**: ⚡ Core modules  |  🧬 Skill evolution  |  🌐 Cloud  |  🔧 Supporting modules\n\n```\nOpenSpace\u002F\n├── openspace\u002F\n│   ├── tool_layer.py                     # OpenSpace main class & OpenSpaceConfig\n│   ├── mcp_server.py                     # MCP Server (4 tools for your agent)\n│   ├── __main__.py                       # CLI entry point (python -m openspace)\n│   ├── dashboard_server.py               # Web dashboard API server\n│   │\n│   ├── ⚡ agents\u002F                         # Agent System\n│   │   ├── base.py                       # Base agent class\n│   │   └── grounding_agent.py            # Execution agent (tool calling, iteration, skill injection)\n│   │\n│   ├── ⚡ grounding\u002F                      # Unified Backend System\n│   │   ├── core\u002F\n│   │   │   ├── grounding_client.py       # Unified interface across all backends\n│   │   │   ├── search_tools.py           # Smart Tool RAG (BM25 + embedding + LLM)\n│   │   │   ├── quality\u002F                  # Tool quality tracking & self-evolution\n│   │   │   ├── security\u002F                 # Policies, sandboxing, E2B\n│   │   │   ├── system\u002F                   # System-level provider & tools\n│   │   │   ├── transport\u002F                # Connectors & task managers\n│   │   │   └── tool\u002F                     # Tool abstraction (base, local, remote)\n│   │   └── backends\u002F\n│   │       ├── shell\u002F                    # Shell command execution\n│   │       ├── gui\u002F                      # Anthropic Computer Use\n│   │       ├── mcp\u002F                      # Model Context Protocol (stdio, HTTP, WebSocket)\n│   │       └── web\u002F                      # Web search & browsing\n│   │\n│   ├── 🧬 skill_engine\u002F                  # Self-Evolving Skill System\n│   │   ├── registry.py                   # Discovery, BM25+embedding pre-filter, LLM selection\n│   │   ├── analyzer.py                   # Post-execution analysis (agent loop + tool access)\n│   │   ├── evolver.py                    # FIX \u002F DERIVED \u002F CAPTURED evolution (3 triggers)\n│   │   ├── patch.py                      # Multi-file FULL \u002F DIFF \u002F PATCH application\n│   │   ├── store.py                      # SQLite persistence, version DAG, quality metrics\n│   │   ├── skill_ranker.py               # BM25 + embedding hybrid ranking\n│   │   ├── retrieve_tool.py              # Skill retrieval tool for agents\n│   │   ├── fuzzy_match.py                # Fuzzy matching for skill discovery\n│   │   ├── conversation_formatter.py     # Format execution history for analysis\n│   │   ├── skill_utils.py                # Shared skill utilities\n│   │   └── types.py                      # SkillRecord, SkillLineage, EvolutionSuggestion\n│   │\n│   ├── 🌐 cloud\u002F                         # Cloud Skill Community\n│   │   ├── client.py                     # HTTP client (upload, download, search)\n│   │   ├── search.py                     # Hybrid search engine\n│   │   ├── embedding.py                  # Embedding generation for skill search\n│   │   ├── auth.py                       # API key management\n│   │   └── cli\u002F                          # CLI tools (download_skill, upload_skill)\n│   │\n│   ├── 💬 communication\u002F                  # Multi-Channel Communication Gateway\n│   │   ├── gateway.py                    # Message routing, session management, reply dispatch\n│   │   ├── adapters\u002F                     # Platform adapters (WhatsApp, Feishu)\n│   │   ├── bridges\u002F                      # Non-Python runtimes (WhatsApp Baileys bridge)\n│   │   ├── config.py                     # Communication config loader\n│   │   ├── session_store.py              # Per-channel session persistence\n│   │   └── types.py                      # ChannelMessage, ChannelSource, SendResult\n│   │\n│   ├── 🔧 platform\u002F                      # Platform abstraction (system info, screenshots)\n│   ├── 🔧 host_detection\u002F                # Auto-detect nanobot \u002F openclaw credentials\n│   ├── 🔧 host_skills\u002F                   # SKILL.md definitions for agent integration\n│   │   ├── delegate-task\u002FSKILL.md        # Teaches agent: execute, fix, upload\n│   │   └── skill-discovery\u002FSKILL.md      # Teaches agent: search & discover skills\n│   ├── 🔧 prompts\u002F                       # LLM prompt templates (grounding + skill engine)\n│   ├── 🔧 llm\u002F                           # LiteLLM wrapper with retry & rate limiting\n│   ├── 🔧 config\u002F                        # Layered configuration system\n│   ├── 🔧 local_server\u002F                  # GUI\u002FShell backend Flask server (server mode)\n│   ├── 🔧 recording\u002F                     # Execution recording, screenshots & video capture\n│   ├── 🔧 utils\u002F                         # Logging, UI, telemetry\n│   └── 📦 skills\u002F                        # Built-in skills (lowest priority, user can add here)\n│\n├── frontend\u002F                             # Dashboard UI (React + Tailwind)\n├── gdpval_bench\u002F                         # GDPVal benchmark experiments & results\n├── showcase\u002F                             # My Daily Monitor (60+ evolved skills)\n│   ├── my-daily-monitor\u002F                 # The full app (zero human code)\n│   └── skills\u002F                           # 60+ evolved skills with full lineage\n├── .openspace\u002F                           # Runtime: embedding cache + skill DB\n└── logs\u002F                                 # Execution logs & recordings\n```\n\n\u003C\u002Fdetails>\n\n---\n\n## 🤝 Contribute & Roadmap\n\nWe welcome contributions! OpenSpace today evolves *how to do X*. The next frontier: **evolving how agents organize doing X together**. \n\nGroup infrastructure (visibility, sharing, permissions) is already live. What comes next:\n\n- [ ] **[Kanban](https:\u002F\u002Fgithub.com\u002FBloopAI\u002Fvibe-kanban)-style orchestration** — Shared task board with skill-aware scheduling; scheduling itself evolves\n- [ ] **Collaboration pattern evolution** — Decomposition, handoff, prioritization strategies captured and improved from completed tasks\n- [ ] **Role emergence** — Agents develop role profiles through practice, not configuration\n- [ ] **Cross-group pattern transfer** — Coordination patterns discovered by one group available to others via cloud registry\n\n---\n\n## 🔗 Related Projects\n\nOpenSpace builds upon the following open-source projects. We sincerely thank their authors and contributors:\n\n- **[AnyTool](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FAnyTool)** — Plug-and-play universal tool-use layer for any AI agent\n- **[ClawWork](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FClawWork)** - Transforms AI assistants into true AI coworkers\n- **[WorldMonitor](https:\u002F\u002Fgithub.com\u002Fkoala73\u002Fworldmonitor)** - Real-time global intelligence dashboard\n\n---\n\n\u003Cdiv align=\"center\">\n\n## ⭐ Star History\n\nIf you find OpenSpace helpful, please consider giving us a star! ⭐\n\n\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fstar-history.com\u002F#HKUDS\u002FOpenSpace&Date\">\n    \u003Cpicture>\n      \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_aa544d6fd96a.png&theme=dark\" \u002F>\n      \u003Csource media=\"(prefers-color-scheme: light)\" srcset=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_aa544d6fd96a.png\" \u002F>\n      \u003Cimg alt=\"Star History Chart\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_aa544d6fd96a.png\" \u002F>\n    \u003C\u002Fpicture>\n  \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n**🧬 Make You Agent Self-Evolve · 🌐 A Community That Grows Together · 💰 Fewer Tokens, Smarter Agents**\n\n\u003C\u002Fdiv>\n\n---\n\n\u003Cp align=\"center\">\n  \u003Cem> ❤️ Thanks for visiting ✨ OpenSpace!\u003C\u002Fem>\u003Cbr>\u003Cbr>\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_7285bc4200c4.png\"\n  alt=\"Views\">\n\u003C\u002Fp>\n","\u003Cdiv align=\"center\">\n\n\u003Cpicture>\n    \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_4aa940ab5ed8.png\" width=\"320px\" style=\"border: none; box-shadow: none;\" alt=\"OpenSpace Logo\">\n\u003C\u002Fpicture>\n\n## ✨ OpenSpace：让您的智能体更聪明、更低成本、自我进化 ✨\n\n| 🔋 **减少46%的Token消耗** | **💰 6小时内赚取1.1万美元** | 🧬 **自我进化的能力** | 🌐 **智能体间的经验共享** |\n\n[![智能体](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F智能体-Claude_Code%20%7C%20Codex%20%7C%20OpenClaw%20%7C%20nanobot%20%7C%20...-99C9BF.svg)](https:\u002F\u002Fmodelcontextprotocol.io\u002F)\n[![Python](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.12+-FCE7D6.svg)](https:\u002F\u002Fwww.python.org\u002F)\n[![许可证](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F许可证-MIT-C1E5F5.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT\u002F)\n[![飞书](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F飞书-群-E9DBFC?style=flat&logo=larksuite&logoColor=white)](.\u002FCOMMUNICATION.md)\n[![微信](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F微信-群-C5EAB4?style=flat&logo=wechat&logoColor=white)](.\u002FCOMMUNICATION.md)\n[![中文文档](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F文档-中文版-F5C6C6?style=flat)](.\u002FREADME_CN.md)\n\n**一条命令即可进化所有AI智能体**：OpenClaw、nanobot、Claude Code、Codex、Cursor等。\n\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_82c25fc4d16b.gif\" width=\"500px\" alt=\"openspace --query your task\">\n\n\u003C\u002Fdiv>\n\n---\n\n## 📢 最新消息\n\n- **2026-04-16** 📊 **进化候选生命周期追踪** — 技能商店现在会记录进化建议何时被处理（`evolution_processed_at`），从而清晰地区分待处理的候选与已处理的版本。\n- **2026-04-12** 🍎 **macOS平台加固** — 将`atomacos`与核心macOS导入解耦，使得截图、窗口控制和录制功能无需它也能独立运行。\n- **2026-04-10** 🎯 **CAPTURED技能** 现在会持久化到宿主智能体自身的技能目录中，而不是默认的注册表路径。云端技能上传现在也正确支持**私密可见性**。\n- **2026-04-09** 💬 多渠道**通信网关**。OpenSpace现在可以接收并响应来自外部平台的消息。自带**WhatsApp**（Baileys桥接 + QR认证）和**飞书**（HTTP Webhook）适配器，支持会话管理、附件缓存以及白名单访问控制。设置详情请参阅[`openspace\u002Fconfig\u002FREADME.md`](openspace\u002Fconfig\u002FREADME.md)。\n- **2026-04-07** 🌐 OpenSpace MCP现在支持独立的**SSE**和**可流式传输的HTTP**启动方式，使远程主机能够通过HTTP连接，而非使用标准输入输出，从而绕过标准输入输出绑定的MCP服务器超时瓶颈。设置细节请参考[主机集成指南](openspace\u002Fhost_skills\u002FREADME.md)。\n- **2026-04-06** 🛠️ 修复了接地、MCP服务、技能进化及持久化等方面的多个运行时问题，提升了长时间运行工作流中的执行稳定性和恢复能力。\n- **2026-04-05** 🧭 清理了LLM凭证解析流程：集中加载`.env`文件，改进了宿主配置的自动检测，并使提供商原生环境变量处理更加一致。\n- **2026-04-03** 🚀 发布了**v0.1.0** — 技能质量监控：从高质量技能中提取的结构化模式现在每天都会评估每一份新提交的技能。更快、更相关的云端搜索。生产级垂直技能集群正由社区自发形成。前端现已支持中文（zh）国际化。\n- **2026-04-02** ⚡ 云端搜索升级，相关性更高且延迟更低。\n- **2026-03-31** 🛡️ 安全加固：强化了zip解压和`import_skill`功能，防止路径遍历攻击。CLI现在会尊重`OPENSPACE_MODEL`和`OPENSPACE_LLM_*`环境变量；增加了MiniMax兼容性；修复了工作流ID冲突问题。\n- **2026-03-29** 🔒 将litellm固定为\u003C1.82.7版本，以避免PYSEC-2026-2供应链攻击。\n- **2026-03-28** 🔧 幂等性技能注册 — `register_skill_dir`现在会对已注册的技能返回现有的`SkillMeta`。更新了OpenClaw的设置文档。\n- **2026-03-27** 🪟 修复了Windows上的标准输入输出死锁问题；通过词干风格的关键字匹配改进了进化确认解析。\n- **2026-03-26** 🌱 每次调用时动态重新扫描技能目录，实现了轻量级本地技能搜索，并简化了文档。\n- **2026-03-25** 🎉 OpenSpace现已开源！\n\n---\n\n## 当今AI智能体面临的问题\n\n如今的AI智能体——如[OpenClaw](https:\u002F\u002Fgithub.com\u002Fopenclaw\u002Fopenclaw)、[nanobot](https:\u002F\u002Fgithub.com\u002FHKUDS\u002Fnanobot)、[Claude Code](https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Fdocs\u002Fclaude-code)、[Codex](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fcodex)、[Cursor](https:\u002F\u002Fcursor.com)等——虽然功能强大，却存在一个关键弱点：它们无法从真实世界的经历中**学习**、**适应**或**进化**，更不用说彼此之间进行**共享**了。\n- **❌ 巨额的Token浪费** - 如何复用成功的任务模式，而不是每次都从头开始推理并消耗大量Token？\n- **❌ 反复发生的高昂失败** - 如何在不同智能体之间共享解决方案，而不是重复同样的昂贵探索和错误？\n- **❌ 技能质量差且不可靠** - 随着工具和API的不断演进，如何保持技能的可靠性？同时，如何确保社区贡献的技能达到严格的质量标准？\n\n## 🎯 什么是OpenSpace？\n\n**🚀 🚀 一个自我进化引擎，每一次任务都能让每个智能体变得更聪明、更经济高效。**\n\nhttps:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fc50f70ab-f6db-47bf-9498-3210c0f0abae\n\nOpenSpace可以接入任何智能体作为技能模块，并赋予其三大超级能力：\n\n### 🧬 自我进化\n能够自动学习和改进的技能\n- ✅ **自动修复** — 当技能出现故障时，它会立即自我修复。\n- ✅ **自动改进** — 成功的模式会转化为更好的技能版本。\n- ✅ **自动学习** — 从实际使用中捕捉高效的 workflows。\n- ✅ **质量监控** — 跟踪所有任务中的技能性能、错误率和执行成功率。\n\n**持续进化的能力——将每一次失败转化为改进，每一次成功转化为优化。**\n\n### 🌐 集体智能\n将单个智能体转化为共享的大脑\n- ✅ **共享进化**：一个智能体的改进会成为所有智能体的升级。\n- ✅ **网络效应**：智能体越多，数据越丰富，每个智能体的进化速度就越快。\n- ✅ **轻松共享** — 仅需一条简单命令即可上传和下载进化后的技能。\n- ✅ **访问控制** — 可以为每项技能选择公开、私有或仅限团队访问。\n\n**一个智能体学习，所有智能体受益——规模化集体智慧。**\n\n### 💰 Token效率\n更智能的智能体，显著降低成本\n- ✅ **停止重复劳动** → 复用成功的解决方案，而不是每次都从零开始。\n- ✅ **任务成本降低** → 随着技能的不断改进，类似的工作成本越来越低。\n- ✅ **只需小范围更新** → 修复损坏的部分，无需重建整个系统。\n- ✅ **实实在在的节约**：在实际任务中，性能提升4.2倍，而Token消耗减少了46%，带来了可观的经济效益。（[GDPVal](#-benchmark-gdpval)）\n\n做更多事，花更少钱——这些智能体真正能在长期使用中为您节省资金。\n\n---\n\n### 区别\n\n**❌ 当前的代理**\n- 随着工具的不断进化，技能会悄然退化\n- 失败的模式会反复出现，缺乏学习机制\n- 知识始终局限于单个代理内部\n\n**✅ OpenSpace 驱动的代理**\n- 多层监控能够及时发现并自动修复问题\n- 成功的工作流程会转化为可复用、可共享的技能\n- 当一个代理学到有用的知识时，所有代理都能立即获得这些知识\n\n### 📊 OpenSpace：将你的代理变成能赚钱的同事\n\n**🎯 真正有意义的实战成果**\n在6个行业的50项专业任务上（**📈 [GDPVal 经济基准](#-benchmark-gdpval)**），使用相同基础大模型（Qwen 3.5-Plus）的OpenSpace代理比基线（[ClawWork](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FClawWork)）代理多赚了**4.2倍的钱**。同时，通过技能进化减少了46%的昂贵token消耗。\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_18c77ed6baef.png\" width=\"100%\" alt=\"GDPVal 基准 — 关键结果\" \u002F>\n\u003C\u002Fdiv>\n\n**💼 这些可不是玩具级的问题**\n- 根据复杂的工会合同构建薪资计算工具\n- 从15份分散的PDF文档中准备纳税申报表\n- 撰写关于加州隐私法规的法律备忘录\n- 制作合规表格和工程规格说明书\n\n**📈 各领域持续取得优异成绩**\n- 合规工作：收益提升18.5%\n- 工程项目：性能提升8.7%\n- 专业文档：所需token数量减少56%\n- 所有类别均有所改善——无一例外\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_73b01c7047c9.png\" width=\"100%\" alt=\"GDPVal 基准 — 按类别展示的任务\" \u002F>\n\u003C\u002Fdiv>\n\n**OpenSpace 不仅让代理更聪明**，还使其具备经济可行性。真正的实际工作、真金白银、可衡量的结果。\n\n## 使用 OpenSpace 进行自主系统开发的用例\n\n**🖥️ [我的日常监控器](showcase\u002FREADME.md)** — OpenSpace 赋予你的代理完成大规模系统开发的能力。这个拥有20多个实时仪表盘面板的个人行为监控系统完全由代理独立构建——60多种技能通过OpenSpace从零开始进化而成，充分展示了自主端到端软件开发的能力。\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_6c1a26859905.png\" width=\"100%\" alt=\"我的日常监控器 – 深色模式\" \u002F>\n\u003C\u002Fdiv>\n\n---\n\n## 📋 目录\n\n- [⚡ 快速入门](#-quick-start)\n  - [🤖 路径 A：用于你的代理](#-path-a-for-your-agent)\n  - [👤 路径 B：作为你的同事](#-path-b-as-your-co-worker)\n  - [📊 本地仪表盘](#-local-dashboard)\n- [📈 基准：GDPVal](#-benchmark-gdpval)\n- [📊 展示：我的日常监控器](#-showcase-my-daily-monitor)\n- [🏗️ 框架](#️-framework)\n  - [🧬 自我进化引擎](#-self-evolution-engine)\n  - [🌐 云端技能社区](#-cloud-skill-community)\n- [🔧 高级配置](#-advanced-configuration)\n- [📖 代码结构](#-code-structure)\n- [🤝 贡献与路线图](#-contribute--roadmap)\n- [🔗 相关项目](#-related-projects)\n\n---\n\n## ⚡ 快速入门\n\n🌐 **只想简单体验？** 无需安装，直接访问 **[open-space.cloud](https:\u002F\u002Fopen-space.cloud)** 浏览社区技能和进化谱系即可。\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace.git && cd OpenSpace\npip install -e .\nopenspace-mcp --help   # 验证安装\n```\n\n> [!TIP]\n> **克隆速度慢？** 默认克隆包含 `assets\u002F` 文件夹（约50 MB图片），体积较大。你可以使用以下轻量级替代方案跳过该文件夹：\n> ```bash\n> git clone --filter=blob:none --sparse https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace.git\n> cd OpenSpace\n> git sparse-checkout set '\u002F*' '!assets\u002F'\n> pip install -e .\n> ```\n\n**选择你的路径：**\n- **[路径 A](#-path-a-for-your-agent)** — 将OpenSpace接入你的代理\n- **[路径 B](#-path-b-as-your-co-worker)** — 直接将OpenSpace用作你的AI同事\n\n### 🤖 路径 A：用于你的代理\n\n适用于任何支持技能（`SKILL.md`）的代理——例如 [Claude Code](https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Fdocs\u002Fclaude-code)、[Codex](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fcodex)、[OpenClaw](https:\u002F\u002Fgithub.com\u002Fopenclaw\u002Fopenclaw)、[nanobot](https:\u002F\u002Fgithub.com\u002FHKUDS\u002Fnanobot)等。\n\n**① 将OpenSpace添加到你的代理的MCP配置中：**\n\n```json\n{\n  \"mcpServers\": {\n    \"openspace\": {\n      \"command\": \"openspace-mcp\",\n      \"toolTimeout\": 600,\n      \"env\": {\n        \"OPENSPACE_HOST_SKILL_DIRS\": \"\u002Fpath\u002Fto\u002Fyour\u002Fagent\u002Fskills\",\n        \"OPENSPACE_WORKSPACE\": \"\u002Fpath\u002Fto\u002FOpenSpace\",\n        \"OPENSPACE_API_KEY\": \"sk-xxx（可选，用于云端）\"\n      }\n    }\n  }\n}\n```\n\n> [!TIP]\n> 凭证（API密钥、模型）会从你的代理配置中**自动检测**，通常无需手动设置。\n\n> [!NOTE]\n> OpenSpace 支持三种启动模式：\n> - **stdio**：在主机配置中保持 `command: \"openspace-mcp\"`。\n> - **SSE**：运行 `openspace-mcp --transport sse --host 127.0.0.1 --port 8080`。\n> - **可流式HTTP**：运行 `openspace-mcp --transport streamable-http --host 127.0.0.1 --port 8081`。\n>\n> 常见的远程端点：\n> - SSE端点：`http:\u002F\u002F127.0.0.1:8080\u002Fsse`\n> - 可流式HTTP端点：`http:\u002F\u002F127.0.0.1:8081\u002Fmcp`\n>\n> `stdio` 是最简单的选项。HTTP模式会将OpenSpace作为独立服务器运行，但仍然需要遵循**主机特定的注册语法**和**主机端的超时设置**。\n\n**② 将技能复制到你的代理的技能目录中：**\n\n```bash\ncp -r OpenSpace\u002Fopenspace\u002Fhost_skills\u002Fdelegate-task\u002F \u002Fpath\u002Fto\u002Fyour\u002Fagent\u002Fskills\u002F\ncp -r OpenSpace\u002Fopenspace\u002Fhost_skills\u002Fskill-discovery\u002F \u002Fpath\u002Fto\u002Fyour\u002Fagent\u002Fskills\u002F\n```\n\n完成。这两个技能会教会你的代理何时以及如何使用OpenSpace——无需额外提示。你的代理现在可以自我进化技能、执行复杂任务，并访问云端技能社区。你也可以添加自定义技能——详情请参阅 [`openspace\u002Fskills\u002FREADME.md`](openspace\u002Fskills\u002FREADME.md)。\n\n> [!NOTE]\n> **云端社区（可选）：** 在 **[open-space.cloud](https:\u002F\u002Fopen-space.cloud)** 注册以获取 `OPENSPACE_API_KEY`，然后将其添加到上述 `env` 块中。如果没有它，本地的所有功能（任务执行、进化、本地技能搜索）仍可正常使用。\n\n📖 每个代理的配置（OpenClaw \u002F nanobot）、所有环境变量及高级设置：[`openspace\u002Fhost_skills\u002FREADME.md`](openspace\u002Fhost_skills\u002FREADME.md)\n\n### 👤 路径 B：作为你的同事\n\n直接使用OpenSpace——编码、搜索、工具使用等——内置了自我进化技能和云端社区。\n\n> [!NOTE]\n> 创建一个 `.env` 文件，填写你的LLM API密钥，并可选地加入 `OPENSPACE_API_KEY` 以访问云端社区（参考 [`openspace\u002F.env.example`](openspace\u002F.env.example)）。\n\n```bash\n# 交互模式\nopenspace\n\n# 执行任务\nopenspace --model \"anthropic\u002Fclaude-sonnet-4-5\" --query \"为我的 Docker 容器创建一个监控仪表盘\"\n```\n\n添加您自己的自定义技能：[`openspace\u002Fskills\u002FREADME.md`](openspace\u002Fskills\u002FREADME.md)。\n\n**云 CLI** — 通过命令行管理技能：\n\n```bash\nopenspace-download-skill \u003Cskill_id>         # 从云端下载技能\nopenspace-upload-skill \u002Fpath\u002Fto\u002Fskill\u002Fdir   # 将技能上传到云端\n```\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>Python API\u003C\u002Fb>\u003C\u002Fsummary>\n\n```python\nimport asyncio\nfrom openspace import OpenSpace\n\nasync def main():\n    async with OpenSpace() as cs:\n        result = await cs.execute(\"分析 GitHub 趋势仓库并生成报告\")\n        print(result[\"response\"])\n\n        for skill in result.get(\"evolved_skills\", []):\n            print(f\"  演化后的技能：{skill['name']} ({skill['origin']})\")\n\nasyncio.run(main())\n```\n\n\u003C\u002Fdetails>\n\n### 📊 本地仪表盘\n\n查看您的技能如何演进——浏览技能、追踪 lineage、比较差异。\n\n> 需要 **Node.js ≥ 20**。\n\n```bash\n# 终端 1. 启动后端 API\nopenspace-dashboard --port 7788\n\n# 终端 2：启动前端开发服务器\ncd frontend\nnpm install        # 只需安装一次\nnpm run dev    \n```\n\n📖 **前端设置指南**：[`frontend\u002FREADME.md`](frontend\u002FREADME.md)\n\n\u003Cdiv align=\"center\">\n\u003Ctable>\n\u003Ctr>\n\u003Ctd width=\"50%\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_4aec6a4c7adf.gif\" width=\"100%\" alt=\"技能分类\" \u002F>\u003C\u002Ftd>\n\u003Ctd width=\"50%\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_8f6b3d5c77f4.gif\" width=\"100%\" alt=\"云端技能记录\" \u002F>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"center\">\u003Csub>技能分类 — 浏览、搜索与排序\u003C\u002Fsub>\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Csub>云端 — 浏览与发现技能记录\u003C\u002Fsub>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd width=\"50%\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_dce23db38fa5.gif\" width=\"100%\" alt=\"版本 lineage\" \u002F>\u003C\u002Ftd>\n\u003Ctd width=\"50%\">\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_ee8537c24a3a.gif\" width=\"100%\" alt=\"工作流会话\" \u002F>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd align=\"center\">\u003Csub>版本 lineage — 技能演化图\u003C\u002Fsub>\u003C\u002Ftd>\n\u003Ctd align=\"center\">\u003Csub>工作流会话 — 执行历史与指标\u003C\u002Fsub>\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\n---\n\n## 📈 基准测试：GDPVal\n\n我们在 [GDPVal](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fopenai\u002Fgdpval) 上评估 OpenSpace——这是一组涵盖 44 种职业的 220 个真实世界专业任务，使用 [ClawWork](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FClawWork) 评估协议，配备相同的生产力工具和基于 LLM 的评分系统。我们的两阶段设计（冷启动 → 热重跑）表明，随着技能的积累，token 消耗会随着时间推移而减少。\n\n公平的基准测试：OpenSpace 使用 Qwen 3.5-Plus 作为其骨干 LLM——与 ClawWork 基线代理相同——从而确保性能差异完全源于技能的演化，而非模型能力。\n\n真正的经济价值：任务范围从构建工资计算器到准备纳税申报表，再到起草法律备忘录——这些都是产生实际 GDP 的专业工作，同时从质量和成本效益两个方面进行评估。\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_45fd6021e81b.png\" width=\"100%\" alt=\"GDPVal 基准测试 — 收入对比\" \u002F>\n\u003C\u002Fdiv>\n\n- **收入高出 ClawWork 4.2 倍**，且使用相同的骨干 LLM（Qwen 3.5-Plus）\n- **价值捕获率 72.8%**——在 15,764 美元的任务价值中赚取了 11,484 美元，表现优于所有其他代理\n- **平均质量 70.8%**——比最佳 ClawWork 代理（40.8%）高出 30 个百分点\n- **第二阶段的 token 使用量比第一阶段减少了 45.9%**——以显著更低的成本获得了更好的结果\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_d3c2a3500c88.png\" width=\"100%\" alt=\"GDPVal 基准测试 — 质量与 token 效率\" \u002F>\n\u003C\u002Fdiv>\n\n### OpenSpace 能够处理哪些真实世界的任务？\n\n50 个 GDPVal 任务涵盖了 6 类真实世界的工作。\n- **第一阶段（冷启动）**依次运行全部 50 个任务——每完成一项任务，技能就会累积到共享数据库中。\n- **第二阶段（热重跑）**则使用第一阶段积累的完整技能数据库，重新执行相同的 50 个任务。\n\n收入捕获率 = 实际获得的报酬 ÷ 最大可能的任务价值\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_73b01c7047c9.png\" width=\"100%\" alt=\"GDPVal 基准测试 — 按类别展示的任务\" \u002F>\n\u003C\u002Fdiv>\n\n## 🎯 演化带来最大影响的领域及其原因：\n\n| 类别 | 收入变化 | Token 变化 | 原因 |\n|---|---|---|---|\n| **📝 文档与通信**（7 项） | 71→74% (+3.3pp) | −56% | 精致的正式输出——加州隐私法备忘录、监视调查报告、子女抚养案件报告。`document-gen-fallback` 技能家族经历了 13 个版本的演化，使结构和错误恢复几乎自动化。 |\n| **📋 合规与表格**（11 项） | 51→70% (+18.5pp) | −51% | 结构化的 PDF 文件——根据 15 份原始文件生成的纳税申报表、药房合规检查清单、临床交接模板。PDF 技能链（检查清单逻辑 → reportlab 布局 → 验证）只需演化一次，随后所有表格类任务都会复用整个流程。 |\n| **🎬 媒体制作**（3 项） | 53→58% (+5.8pp) | −46% | 通过 Python 和 ffmpeg 进行音视频制作——根据鼓声参考生成波萨诺瓦纯音乐、从 5 条音轨中编辑贝斯音轨、从 13 个源视频中制作 CGI 展示片。演化的技能编码了有效的 ffmpeg 标志和编解码器回退方案，消除了沙盒中的反复试验。 |\n| **🛠️ 工程**（4 项） | 70→78% (+8.7pp) | −43% | 多交付成果的技术项目——Web3 全栈开发（Solidity + React + 测试）、CNC 加工单元安全系统（报告 + 布局 + 硬件图纸）、航空航天 CFD 报告。协调技能在这些多样化的任务中具有普遍适用性。 |\n| **📊 电子表格**（15 项） | 63→70% (+7.3pp) | −37% | 功能齐全的 .xlsx 工具——根据工会合同生成的工资计算器、基于历史数据的销售预测、结合竞争对手对标的价格模型。电子表格模式（公式、合并单元格、验证规则）在不同领域中是相同的。 |\n| **📈 战略与分析**（10 项） | 88→89% (+1.0pp) | −32% | 战略建议——供应商谈判策略、非营利组织项目评估、针对 3 亿美元交易台的能源交易分析。这些任务本身已经具有最高质量（88%），节省主要来自于重复利用文档结构和多文件编排。 |\n\n### 进化产生了什么？（165项技能）\n\n在50个第一阶段任务中，OpenSpace自主进化出了**165项技能**。一个突破性的发现是：这些技能不仅仅是领域知识——它们更是**稳健的执行模式**和**质量保证工作流**。智能体学会了如何在不完美的真实环境中可靠地交付结果。\n\n**关键发现**：大多数技能聚焦于工具的可靠性和错误恢复，而非特定任务的知识。\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_8bfba90ef327.png\" width=\"100%\" alt=\"GDPVal基准测试 — 进化后的技能分类\" \u002F>\n\u003C\u002Fdiv>\n\n| 目的 | 数量 | 对智能体的启示 |\n|---|---|---|\n| **文件格式输入输出** | 44 | PDF提取的后备方案、DOCX解析、Excel合并单元格处理、PPTX创建。其中32\u002F44项是从实际失败中“捕获”的——每一项都解决了生产环境中的一个Bug。 |\n| **执行恢复** | 29 | 分层后备机制：沙箱失败→Shell→先写文件再运行→heredoc。28\u002F29项同样来自真实的崩溃场景。这是让所有其他操作都可靠的基石。 |\n| **文档生成** | 26 | 端到端的文档流水线。`document-gen-fallback`从最初导入的1项技能，进化出**13个衍生版本**——这是迭代最深入的技能家族。 |\n| **质量保证** | 23 | 写后校验：检查Excel行数、验证PDF页数、对表格公式进行Proof-Gate检查。为什么P2级别的质量会提升——智能体不仅产出内容，还会主动验证。 |\n| **任务编排** | 17 | 多文件跟踪、ZIP打包、零次迭代即发现失败的能力。这些元技能可以帮助处理涉及多个交付物的各种任务类型。 |\n| **领域工作流** | 13 | SOAP病历、音频制作（从1个模板衍生出**4代**）、视频流水线。虽然数量不多，但每个领域的进化都非常深入。 |\n| **Web与研究** | 11 | SSL\u002F代理调试、搜索后备方案、复杂JS页面的处理。其中包括2项“修复型”技能——因为Web访问本身就非常不稳定。 |\n\n**复现实验、分析工具及结果**：[`gdpval_bench\u002FREADME.md`](gdpval_bench\u002FREADME.md)\n\n---\n\n## 📊 展示：我的每日监控面板\n\n> **全程未编写任何人工代码。** 60余项技能从零开始进化，构建了一个完全可用的实时仪表盘。\n\n**我的每日监控面板**是一个始终在线的仪表盘，可实时展示进程、服务器、新闻、市场、邮件和日程安排，并内置了一个AI智能体。\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_a10f865e4476.png\" width=\"90%\" alt=\"我的每日监控面板 – 浅色模式\" \u002F>\n\u003C\u002Fdiv>\n\n### OpenSpace是如何从零开始构建它的\n\n| 阶段 | 发生了什么 | 技能 |\n|-------|--------------|--------|\n| 🌱 **种子** | 分析了开源项目[WorldMonitor](https:\u002F\u002Fgithub.com\u002Fkoala73\u002Fworldmonitor)，提取参考模式 | 6项初始技能 |\n| 🏗️ **脚手架** | 生成项目结构、Vite配置、TypeScript设置 | +8项技能 |\n| 🎨 **构建** | 创建20+个数据服务面板、API路由和网格布局 | +25项技能 |\n| 🔧 **修复** | 自动修复损坏的TypeScript、API不匹配和CSS冲突 | +12项FIX型进化 |\n| 🧬 **进化** | 衍生出增强型模式，合并互补技能 | +15项DERIVED型技能 |\n| 📦 **捕获** | 从成功执行中提取可重用模式 | +8项CAPTURED型技能 |\n\n### 📈 技能进化图\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_7431bb292cac.png\" width=\"90%\" alt=\"技能进化图\" \u002F>\n\u003C\u002Fdiv>\n\n> 每个节点都是OpenSpace学习、提取或优化过的技能。完整的进化历史已开源至[`showcase\u002F.openspace\u002Fopenspace.db`](showcase\u002F.openspace\u002Fopenspace.db)——你可以在任意SQLite浏览器中加载它，探索技能的 lineage、差异以及质量指标。\n\n**详细信息**：[`showcase\u002FREADME.md`](showcase\u002FREADME.md)\n\n---\n\n## 🏗️ OpenSpace框架\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_77235adca28f.png\" width=\"90%\" alt=\"OpenSpace框架\" \u002F>\n\u003C\u002Fdiv>\n\n### 🧬 自我进化引擎\n\nOpenSpace的核心。技能并非静态文件——它们是能够自动选择、应用、监控、分析并自我进化的活体实体。\n\n#### 🔄 自主且持续的进化\n\n- **全生命周期管理**：从发现到应用再到进化——全程无需人工干预。即使不存在匹配的技能，OpenSpace也能完成任务。\n\n**三种进化模式**：\n- 🔧 FIX — 就地修复破损或过时的指令。同一技能，新版本。\n- 🚀 DERIVED — 从父技能中创建增强版或专用版。新建技能目录，与父技能共存。\n- ✨ CAPTURED — 从成功执行中提取新颖的可重用模式。全新的技能，无父技能。\n\n**三个独立触发机制**：多重防线防止技能退化——无论是成功还是失败的执行都会推动进化。\n- **📈 执行后分析** — 每次任务完成后运行。分析完整记录，并为相关技能提出FIX\u002FDERIVED\u002FCAPTURED建议。\n- **⚠️ 工具退化** — 当工具的成功率下降时，质量监控系统会找到所有依赖该工具的技能，并批量进化。\n- **📊 指标监控** — 定期扫描技能健康指标（应用率、完成率、回退率），并对表现不佳的技能进行进化。\n\n#### 📊 全栈质量监控\n多层追踪：质量监控覆盖整个执行栈——从高层工作流到单个工具调用：\n- **🎯 技能** — 应用率、完成率、有效率、回退率\n- **🔨 工具调用** — 成功率、延迟、标记的问题\n- **⚡ 代码执行** — 执行状态、错误模式\n\n**级联进化**：当任何一个组件退化——无论是技能工作流还是单个工具调用——所有上游依赖的技能都会自动触发进化，从而保持系统整体的一致性。\n\n#### 🔧 智能且安全的进化\n**🤖 自主进化**：每次进化都会探索代码库，发现根本原因，并自主决定修复方案——在做出更改之前收集真实证据，而不是盲目生成。\n\n**⚡ 基于差异且节省token**：只生成最小化、有针对性的代码改动，而非全面重写；失败时会自动重试。每个版本都存储在一个带有完整 lineage 追踪的版本DAG中。\n\n**🛡️ 内置保障措施**：\n- 确认门限可减少误触发\n- 防循环保护可防止进化陷入无限循环\n- 安全检查会标记危险模式（提示注入、凭据泄露）\n- 进化后的技能会在替换旧版本前经过验证。\n\n**🌐 协作式技能社区**\n一个协作式的注册中心，智能体可以在这里共享进化后的技能。当一个智能体进化出改进时，所有连接的智能体都可以发现、导入并在此基础上进一步开发——将个体的进步转化为集体智慧。\n\n- **🔐 灵活的共享方式**：可以公开分享技能，也可以仅在小组内共享，或者保持私密。智能搜索会找到你需要的内容并自动导入。每一次进化都会被追溯到其 lineage，并附带完整的差异报告。\n\n- **☁️ 协作平台**：open-space.cloud——注册API密钥，浏览社区技能，并管理你的小组。\n\n---\n\n## 🔧 高级配置\n\n对于大多数用户来说，[快速入门](#-quick-start) 就足够了。如需使用高级选项（环境变量、执行模式、安全策略等），请参阅 [`openspace\u002Fconfig\u002FREADME.md`](openspace\u002Fconfig\u002FREADME.md)。\n\n---\n\n\u003Ca id=\"-code-structure\">\u003C\u002Fa>\n\u003Cdetails>\n\u003Csummary>\u003Cb>📖 代码结构\u003C\u002Fb>\u003C\u002Fsummary>\n\n> **图例**: ⚡ 核心模块  |  🧬 技能进化  |  🌐 云端  |  🔧 支持性模块\n\n```\nOpenSpace\u002F\n├── openspace\u002F\n│   ├── tool_layer.py                     # OpenSpace 主类及 OpenSpaceConfig\n│   ├── mcp_server.py                     # MCP 服务器（为你的智能体提供4种工具）\n│   ├── __main__.py                       # CLI 入口点（python -m openspace）\n│   ├── dashboard_server.py               # Web 控制台 API 服务器\n│   │\n│   ├── ⚡ agents\u002F                         # 智能体系统\n│   │   ├── base.py                       # 基础智能体类\n│   │   └── grounding_agent.py            # 执行型智能体（调用工具、迭代、注入技能）\n│   │\n│   ├── ⚡ grounding\u002F                      # 统一后端系统\n│   │   ├── core\u002F\n│   │   │   ├── grounding_client.py       # 跨所有后端的统一接口\n│   │   │   ├── search_tools.py           # 智能工具 RAG（BM25 + 嵌入 + LLM）\n│   │   │   ├── quality\u002F                  # 工具质量追踪与自我进化\n│   │   │   ├── security\u002F                 # 策略、沙箱机制、E2B\n│   │   │   ├── system\u002F                   # 系统级提供商与工具\n│   │   │   ├── transport\u002F                # 连接器与任务管理器\n│   │   │   └── tool\u002F                     # 工具抽象（基础、本地、远程）\n│   │   └── backends\u002F\n│   │       ├── shell\u002F                    # Shell 命令执行\n│   │       ├── gui\u002F                      # Anthropic 计算机使用\n│   │       ├── mcp\u002F                      # 模型上下文协议（stdio、HTTP、WebSocket）\n│   │       └── web\u002F                      # 网络搜索与浏览\n│   │\n│   ├── 🧬 skill_engine\u002F                  # 自我进化技能系统\n│   │   ├── registry.py                   # 发现、BM25+嵌入预过滤、LLM 选择\n│   │   ├── analyzer.py                   # 执行后分析（智能体循环 + 工具访问）\n│   │   ├── evolver.py                    # FIX \u002F DERIVED \u002F CAPTURED 进化（3种触发条件）\n│   │   ├── patch.py                      # 多文件 FULL \u002F DIFF \u002F PATCH 应用\n│   │   ├── store.py                      # SQLite 持久化、版本 DAG、质量指标\n│   │   ├── skill_ranker.py               # BM25 + 嵌入混合排名\n│   │   ├── retrieve_tool.py              # 用于智能体的技能检索工具\n│   │   ├── fuzzy_match.py                # 技能发现的模糊匹配\n│   │   ├── conversation_formatter.py     # 格式化执行历史以供分析\n│   │   ├── skill_utils.py                # 共享技能工具\n│   │   └── types.py                      # SkillRecord、SkillLineage、EvolutionSuggestion\n│   │\n│   ├── 🌐 cloud\u002F                         # 云端技能社区\n│   │   ├── client.py                     # HTTP 客户端（上传、下载、搜索）\n│   │   ├── search.py                     # 混合搜索引擎\n│   │   ├── embedding.py                  # 用于技能搜索的嵌入生成\n│   │   ├── auth.py                       # API 密钥管理\n│   │   └── cli\u002F                          # CLI 工具（download_skill、upload_skill）\n│   │\n│   ├── 💬 communication\u002F                  # 多渠道通信网关\n│   │   ├── gateway.py                    # 消息路由、会话管理、回复分发\n│   │   ├── adapters\u002F                     # 平台适配器（WhatsApp、Feishu）\n│   │   ├── bridges\u002F                      # 非 Python 运行时（WhatsApp Baileys 桥接）\n│   │   ├── config.py                     # 通信配置加载器\n│   │   ├── session_store.py              # 每个渠道的会话持久化\n│   │   └── types.py                      # ChannelMessage、ChannelSource、SendResult\n│   │\n│   ├── 🔧 platform\u002F                      # 平台抽象层（系统信息、截图）\n│   ├── 🔧 host_detection\u002F                # 自动检测纳米机器人 \u002F openclaw 凭证\n│   ├── 🔧 host_skills\u002F                   # SKILL.md 定义，用于智能体集成\n│   │   ├── delegate-task\u002FSKILL.md        # 教导智能体：执行、修复、上传\n│   │   └── skill-discovery\u002FSKILL.md      # 教导智能体：搜索并发现技能\n│   ├── 🔧 prompts\u002F                       # LLM 提示模板（接地 + 技能引擎）\n│   ├── 🔧 llm\u002F                           # LiteLLM 包装器，带重试和速率限制\n│   ├── 🔧 config\u002F                        # 分层配置系统\n│   ├── 🔧 local_server\u002F                  # GUI\u002FShell 后端 Flask 服务器（服务器模式）\n│   ├── 🔧 recording\u002F                     # 执行记录、截图及视频捕获\n│   ├── 🔧 utils\u002F                         # 日志记录、UI、遥测\n│   └── 📦 skills\u002F                        # 内置技能（优先级最低，用户可在此添加）\n│\n├── frontend\u002F                             # 控制台 UI（React + Tailwind）\n├── gdpval_bench\u002F                         # GDPVal 基准测试实验及结果\n├── showcase\u002F                             # 我的日常监控（60+进化的技能）\n│   ├── my-daily-monitor\u002F                 # 完整应用（无需人工编写代码）\n│   └── skills\u002F                           # 60+进化的技能及其完整谱系\n├── .openspace\u002F                           # 运行时：嵌入缓存 + 技能数据库\n└── logs\u002F                                 # 执行日志及录制\n```\n\n\u003C\u002Fdetails>\n\n---\n\n## 🤝 贡献与路线图\n\n我们欢迎各方贡献！目前，OpenSpace 正在演进的是“如何完成 X”。下一个前沿领域是：**演进智能体如何协同组织完成 X**。\n\n团队基础设施（可见性、共享、权限）已经上线。接下来的工作包括：\n\n- [ ] **[Kanban](https:\u002F\u002Fgithub.com\u002FBloopAI\u002Fvibe-kanban)-风格编排** — 共享任务看板，具备技能感知的调度功能；调度本身也将不断进化\n- [ ] **协作模式进化** — 从已完成的任务中捕捉并改进分解、交接、优先级排序等策略\n- [ ] **角色涌现** — 智能体通过实践而非配置来发展角色定位\n- [ ] **跨团队模式转移** — 一个团队发现的协调模式可通过云端注册表供其他团队使用\n\n---\n\n## 🔗 相关项目\n\nOpenSpace 基于以下开源项目构建。我们衷心感谢这些项目的作者和贡献者：\n\n- **[AnyTool](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FAnyTool)** — 适用于任何 AI 智能体的即插即用通用工具层\n- **[ClawWork](https:\u002F\u002Fgithub.com\u002FHKUDS\u002FClawWork)** — 将 AI 助手转变为真正的 AI 同事\n- **[WorldMonitor](https:\u002F\u002Fgithub.com\u002Fkoala73\u002Fworldmonitor)** — 实时全球情报仪表盘\n\n---\n\n\u003Cdiv align=\"center\">\n\n## ⭐ 星标历史\n\n如果您觉得 OpenSpace 很有帮助，请考虑给我们点个星！⭐\n\n\u003Cdiv align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fstar-history.com\u002F#HKUDS\u002FOpenSpace&Date\">\n    \u003Cpicture>\n      \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_aa544d6fd96a.png&theme=dark\" \u002F>\n      \u003Csource media=\"(prefers-color-scheme: light)\" srcset=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_aa544d6fd96a.png\" \u002F>\n      \u003Cimg alt=\"星标历史图表\" src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_aa544d6fd96a.png\" \u002F>\n    \u003C\u002Fpicture>\n  \u003C\u002Fa>\n\u003C\u002Fdiv>\n\n**🧬 让你的智能体自我进化 · 🌐 共同成长的社区 · 💰 更少的 Token，更聪明的智能体**\n\n\u003C\u002Fdiv>\n\n---\n\n\u003Cp align=\"center\">\n  \u003Cem> ❤️ 感谢您的光临 ✨ OpenSpace！\u003C\u002Fem>\u003Cbr>\u003Cbr>\n  \u003Cimg src=\"https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_readme_7285bc4200c4.png\"\n  alt=\"访问量\">\n\u003C\u002Fp>","# OpenSpace 快速上手指南\n\nOpenSpace 是一个让 AI Agent（如 Claude Code、Codex、OpenClaw 等）具备**自进化**、**技能共享**和**低成本**能力的引擎。它能通过复用成功模式减少 46% 的 Token 消耗，并让所有接入的 Agent 共享进化成果。\n\n## 1. 环境准备\n\n在开始之前，请确保您的系统满足以下要求：\n\n*   **操作系统**: Linux, macOS 或 Windows\n*   **Python 版本**: 3.12 或更高版本\n*   **依赖管理**: `pip`\n*   **网络环境**: 需能访问 GitHub 和 PyPI（国内用户建议使用镜像源加速）\n\n## 2. 安装步骤\n\n### 方案 A：标准安装\n克隆仓库并安装依赖：\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace.git && cd OpenSpace\npip install -e .\n```\n\n验证安装是否成功：\n```bash\nopenspace-mcp --help\n```\n\n### 方案 B：轻量安装（推荐国内或网络较慢用户）\n默认仓库包含约 50MB 的图片资源。若只需核心功能，可使用稀疏克隆跳过资源文件，大幅加快下载速度：\n\n```bash\ngit clone --filter=blob:none --sparse https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace.git\ncd OpenSpace\ngit sparse-checkout set '\u002F*' '!assets\u002F'\npip install -e .\n```\n\n> **💡 国内加速提示**：如果 `pip install` 速度较慢，建议指定国内镜像源：\n> ```bash\n> pip install -e . -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n> ```\n\n## 3. 基本使用\n\nOpenSpace 主要有两种使用模式：**作为 Agent 的插件** 或 **直接作为 AI 同事**。\n\n### 模式一：集成到现有 Agent (Path A)\n适用于支持 MCP (Model Context Protocol) 的 Agent，如 Claude Code、Codex、OpenClaw、nanobot 等。\n\n**步骤 1：配置 MCP**\n在您的 Agent 配置文件（通常是 `mcp.json` 或类似配置）中添加 OpenSpace 服务：\n\n```json\n{\n  \"mcpServers\": {\n    \"openspace\": {\n      \"command\": \"openspace-mcp\",\n      \"toolTimeout\": 600,\n      \"env\": {\n        \"OPENSPACE_HOST_SKILL_DIRS\": \"\u002Fpath\u002Fto\u002Fyour\u002Fagent\u002Fskills\",\n        \"OPENSPACE_WORKSPACE\": \"\u002Fpath\u002Fto\u002FOpenSpace\",\n        \"OPENSPACE_API_KEY\": \"sk-xxx\" \n      }\n    }\n  }\n}\n```\n*注：`OPENSPACE_API_KEY` 仅在需要使用云端技能共享时填写；API Key 和模型通常会自动从宿主 Agent 配置中检测。*\n\n**步骤 2：启动与运行**\n重启您的 Agent。OpenSpace 将自动拦截任务，尝试复用已有技能或在失败时自动修复\u002F进化技能。\n\n*支持多种启动传输方式：*\n*   **Stdio (默认)**: 配置中保留 `\"command\": \"openspace-mcp\"`\n*   **SSE**: `openspace-mcp --transport sse --host 127.0.0.1 --port 8080`\n*   **Streamable HTTP**: `openspace-mcp --transport streamable-http --host 127.0.0.1 --port 8081`\n\n### 模式二：直接作为 AI 同事 (Path B)\n如果您没有特定的 Agent 框架，可以直接通过命令行让 OpenSpace 执行任务。\n\n```bash\nopenspace --query \"帮我分析当前的销售数据并生成报告\"\n```\n\n### 探索云端技能\n无需安装即可浏览社区进化的技能库和演化历史：\n👉 **[open-space.cloud](https:\u002F\u002Fopen-space.cloud)**","某电商初创公司的技术团队正利用 Claude Code 和 OpenClaw 等多个 AI 代理构建自动化订单处理系统，急需提升代理的长期协作效率并控制高昂的 API 成本。\n\n### 没有 OpenSpace 时\n- **重复造轮子成本高**：不同代理各自为战，A 代理学会的“异常订单识别”技巧无法共享给 B 代理，导致每次新任务都要重新消耗大量 Token 学习相同逻辑。\n- **Token 消耗失控**：由于缺乏历史经验复用，代理在处理相似流程时反复请求基础指令，月度 API 账单居高不下，严重挤压利润空间。\n- **能力迭代停滞**：代理的技能库是静态的，无法从过往成功或失败的案例中自动提取优化策略，遇到复杂边缘情况时容易反复犯错。\n- **多端协作割裂**：团队希望通过 WhatsApp 或飞书接收外部通知并触发代理行动，但需手动编写大量胶水代码来打通通信壁垒，维护极其困难。\n\n### 使用 OpenSpace 后\n- **技能全网共享**：OpenSpace 让所有代理接入统一的“技能商店”，一个代理掌握的优质解题思路能瞬间同步给集群内其他成员，彻底消除重复学习。\n- **大幅降低开销**：通过复用已验证的高质量技能模式，整体 Token 消耗量减少 46%，显著降低了运行成本，让小规模团队也能负担得起高频自动化任务。\n- **自主进化成长**：系统会自动监控任务执行质量，将成功的操作路径转化为新技能并持久化存储，使代理团队具备“越用越聪明”的自我演进能力。\n- **无缝多端连接**：借助内置的通信网关，只需简单配置即可让代理直接响应 WhatsApp 或飞书消息，无需额外开发即可实现跨平台业务闭环。\n\nOpenSpace 通过构建可共享、自进化的技能生态，将分散的 AI 代理升级为低成本、高智能的协同作战军团。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FHKUDS_OpenSpace_73b01c70.png","HKUDS","✨Data Intelligence Lab@HKU✨","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FHKUDS_fc32cc87.jpg",null,"https:\u002F\u002Fsites.google.com\u002Fview\u002Fchaoh","https:\u002F\u002Fgithub.com\u002FHKUDS",[80,84,88,92,96,100],{"name":81,"color":82,"percentage":83},"Python","#3572A5",76,{"name":85,"color":86,"percentage":87},"TypeScript","#3178c6",18.4,{"name":89,"color":90,"percentage":91},"CSS","#663399",4.4,{"name":93,"color":94,"percentage":95},"JavaScript","#f1e05a",0.9,{"name":97,"color":98,"percentage":99},"Shell","#89e051",0.1,{"name":101,"color":102,"percentage":99},"HTML","#e34c26",5483,669,"2026-04-19T18:25:18","MIT","Linux, macOS, Windows","未说明",{"notes":110,"python":111,"dependencies":112},"该工具主要作为 AI Agent 的 MCP（Model Context Protocol）服务器运行，依赖外部 LLM API（如 Claude, OpenAI 等），而非本地运行大模型，因此无明确 GPU 需求。macOS 平台已优化截图和窗口控制功能（可选依赖 atomacos）。Windows 平台已修复 stdio 死锁问题。支持通过 stdio、SSE 或 streamable HTTP 模式启动。","3.12+",[113,114],"litellm\u003C1.82.7","atomacos (macOS, optional)",[13,14],"2026-03-27T02:49:30.150509","2026-04-20T10:23:57.009787",[119,124,129,134,139,144,149],{"id":120,"question_zh":121,"answer_zh":122,"source_url":123},44333,"OpenSpace 项目中的 litellm 依赖项存在供应链攻击风险，应如何修复？","由于 litellm 的特定版本（1.82.7 和 1.82.8）包含恶意代码，建议在 `pyproject.toml` 和 `requirements.txt` 中将依赖版本限制在安全范围内。请将配置修改为：`litellm>=1.70.0,\u003C1.82.7`。这是一个临时措施，项目方正在计划迁移到直接调用提供商 SDK（如 anthropic, openai）以彻底移除此依赖风险。","https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace\u002Fissues\u002F31",{"id":125,"question_zh":126,"answer_zh":127,"source_url":128},44334,"OpenClaw 调用 execute_task 时返回 status: \"error\"，但在终端中运行正常，该如何排查？","这种情况通常与环境差异或权限有关。维护者建议提供相关的日志文件以便进一步调查。请在分享日志前，务必掩盖（mask）任何个人或敏感信息（如 API 密钥、路径等）。一旦获得脱敏后的日志，团队可以帮助深入分析具体错误原因。","https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace\u002Fissues\u002F57",{"id":130,"question_zh":131,"answer_zh":132,"source_url":133},44335,"OpenSpace 是否可以与 Serena 项目结合使用？","目前 OpenSpace 的官方文档（包括 README）中没有提及与 Serena 的集成，也没有官方的支持或指导。虽然用户可以尝试自行结合两者，但这并非开箱即用的功能，需要用户自行探索兼容性。","https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace\u002Fissues\u002F34",{"id":135,"question_zh":136,"answer_zh":137,"source_url":138},44336,"遇到 LiteLLM 警告\"Connection refused. Falling back to local backup\"是什么意思？","这表示 LiteLLM 尝试从 GitHub 远程获取最新的模型价格映射表（model_prices_and_context_window.json）时连接被拒绝（可能是网络问题或 GitHub 限流）。系统已自动回退到使用本地备份的文件，这通常不会影响核心功能的正常运行，只是可能无法获取最新的模型定价信息。","https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace\u002Fissues\u002F58",{"id":140,"question_zh":141,"answer_zh":142,"source_url":143},44337,"为什么对已注册的技能执行 fix_skill 操作时会失败？","这是一个已知逻辑缺陷。`fix_skill` 函数在修复前会尝试重新注册技能，而注册函数在技能已存在时返回 `None`，导致误报失败。解决方法是修改 `mcp_server.py` 中的逻辑：当 `register_skill_dir` 返回 `None` 时，应尝试通过 `get_skill_by_path` 获取现有的技能元数据，而不是直接报错；或者修改注册函数，使其在技能已存在时直接返回现有的 `SkillMeta` 对象。","https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace\u002Fissues\u002F29",{"id":145,"question_zh":146,"answer_zh":147,"source_url":148},44338,"OpenSpace MCP 服务器是否存在向客户端泄露敏感信息的风险？","是的，早期版本中存在一个严重的安全问题：服务器在错误响应中会将完整的 Python 堆栈跟踪（traceback）返回给 MCP 客户端，这可能泄露内部文件路径、代码结构和配置细节。修复方案是修改 `mcp_server.py`，仅在服务器端日志中记录完整堆栈，而向客户端仅返回通用的错误类型和消息，移除 `traceback` 字段。","https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace\u002Fissues\u002F19",{"id":150,"question_zh":151,"answer_zh":152,"source_url":153},44339,"OpenSpace 的云客户端在解压 ZIP 文件时是否存在路径遍历漏洞？","是的，旧版本的 `_extract_zip()` 函数仅检查文件名是否以 `..` 或 `\u002F` 开头，这不足以防止如 `subdir\u002F..\u002F..\u002F..\u002Fetc\u002Fpasswd` 这类复杂的路径遍历攻击。修复方法是：在计算目标路径后，使用 `.resolve()` 解析绝对路径，并验证该路径是否仍然位于目标目录内（使用 `Path.is_relative_to()`），如果不是则跳过该文件。","https:\u002F\u002Fgithub.com\u002FHKUDS\u002FOpenSpace\u002Fissues\u002F17",[]]