[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-81521":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":12,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":15,"stars30d":15,"stars90d":14,"forks30d":14,"starsTrendScore":14,"compositeScore":16,"rankGlobal":9,"rankLanguage":9,"license":17,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":20,"hasPages":18,"topics":21,"createdAt":9,"pushedAt":9,"updatedAt":22,"readmeContent":23,"aiSummary":24,"trendingCount":14,"starSnapshotCount":14,"syncStatus":25,"lastSyncTime":26,"discoverSource":27},81521,"MicroClaw","microsofthackathons\u002FMicroClaw","microsofthackathons","One click to install OpenClaw on Windows, one click to run it securely and seamlessly.",null,"TypeScript",28,3,27,0,1,1.81,"MIT License",false,"main",true,[],"2026-06-12 02:04:16","# 🦞 MicroClaw\n\n[中文版](README.zh-CN.md)\n\n**MicroClaw** makes [OpenClaw](https:\u002F\u002Fgithub.com\u002Fopenclaw) instantly available on Windows through a familiar, low-friction install experience. It packages the desktop client, local Gateway, managed runtime, preloaded skills, and permission-controlled sandbox into one product so users can get to real tasks quickly. You bring the LLM connection; MicroClaw brings the Windows app, local runtime, and trust boundary.\n\n> [!WARNING]\n> **AI & Security Notice**\n> MicroClaw is an experimental open-source project initiated by Microsoft that provides a secure execution environment and user interface for the open-source [OpenClaw](https:\u002F\u002Fgithub.com\u002Fopenclaw) software that users have installed on their devices. MicroClaw is **not** an AI service, it doesn't contain an AI model, nor does it generate or alter user prompts, responses, or any AI-generated contents on behalf of users.\n>\n> All AI tasks including inferencing and content generation are solely performed by OpenClaw software with the AI model of your choice, accessed via your own API key and credentials. AI-generated content may be incorrect, AI is not a human, and it has no personality. The sandbox restricts what OpenClaw can access but does not prevent OpenClaw or the LLM from performing actions within the permissions you have granted. You are responsible for reviewing all prompts before submission and for not running prompts from untrusted sources.\n>\n> 📄 See [DISCLAIMER.md](DISCLAIMER.md) for the full disclaimer, including prompt injection risks, sandbox scope, and limitation of liability.\n\n## Why MicroClaw\n\nMicroClaw is designed to remove the usual Windows setup friction around OpenClaw. Instead of asking users to assemble Node.js, OpenClaw, local configuration, skills, and sandboxing by hand, it installs and wires those pieces together in a single flow.\n\n### Instant Availability\n\n- **Familiar Windows install flow**: packaged installer with desktop shortcut, Start menu entry, and one-click uninstall\n- **One run sets up the runtime**: Git, managed Node.js, OpenClaw Gateway, the MicroClaw desktop app, managed skills, and AppContainer provisioning\n- **Ready after install**: launch the app immediately after setup instead of building a local OpenClaw environment by hand\n\n### Ready-to-Use Experience\n\n- **Only the LLM is bring-your-own**: the user supplies the model endpoint, API key, and model name; the rest of the stack is installed and configured by MicroClaw\n- **Fast path on first launch**: if no provider is configured, the app opens a setup wizard that asks only for model credentials; if `MODEL_*` values already exist in `.env`, MicroClaw can auto-configure them\n- **Recommended tasks on day one**: the home screen ships with starter task cards and prompt suggestions so users can begin with concrete tasks instead of a blank chat box\n- **Preloaded capability surface**: 52 bundled skills plus 7 managed skills are available as part of the default Windows experience\n\n### Built-In Trust\n\n- **Transparent, permission-controlled actions**: file and tool access requests are surfaced to the user instead of being silently granted\n- **Sandboxed with Windows AppContainer**: tool execution runs inside an OS-enforced AppContainer boundary on supported systems\n- **Hooks plus sandbox, defense in depth**: hook-based prechecks improve UX while AppContainer ACL enforcement remains the actual security boundary\n- **Hard blocks for sensitive paths**: credential-heavy locations such as `.ssh` and cloud config folders are denied rather than merely warned about\n\n---\n\n## Architecture\n\n```mermaid\ngraph TB\n    User((\"👤 User\"))\n\n    Desktop[\"Desktop App\"]\n    Gateway[\"OpenClaw Gateway\"]\n    Skills[\"Skills\u003Cbr\u002F>\u003Ci>52 built-in + 7 managed\u003C\u002Fi>\"]\n    Plugins[\"Plugins\u003Cbr\u002F>\u003Ci>WeChat · Teams\u003C\u002Fi>\"]\n    PermMgr[\"Permission Manager\"]\n    Installer[\"Installer\"]\n\n    subgraph External[\"External Services\"]\n        LLM[\"LLM API\"]\n        ThirdParty[\"WeChat · Bing · M365\"]\n    end\n\n    User --> Desktop\n    User --> PermMgr\n    Desktop \u003C--> Gateway\n    Gateway --> Skills\n    Gateway --> Plugins\n    Gateway --> LLM\n    Skills --> ThirdParty\n    Plugins --> ThirdParty\n    Installer -.->|\"sets up\"| Desktop\n    Installer -.->|\"sets up\"| Gateway\n\n    classDef app fill:#4A90D9,stroke:#2C5F8A,color:#fff\n    classDef gw fill:#3AAA6D,stroke:#2B7F52,color:#fff\n    classDef ext fill:#E8744F,stroke:#B85636,color:#fff\n    classDef tool fill:#78C0E0,stroke:#5090B0,color:#333\n\n    class Desktop app\n    class Gateway,Skills,Plugins gw\n    class LLM,ThirdParty ext\n    class PermMgr,Installer tool\n```\n\n### Key Interactions\n\n| Path | Protocol | Description |\n|---|---|---|\n| Renderer ↔ Main Process | Electron IPC | `namespace:action` pattern (e.g. `chat:send-message`, `gateway:restart`) |\n| Main Process ↔ Gateway | WebSocket JSON-RPC | Port 18789 loopback, Ed25519 device auth, streaming chat events |\n| Main Process → Gateway | HTTP | `\u002Fhealth` polled every 10s; process spawn with auto-restart |\n| Gateway → LLM | HTTPS | Streaming API calls to Claude \u002F OpenAI \u002F Gemini |\n| Gateway → Skills | In-process | Agent loop invokes tool calls; results fed back to LLM |\n| WeChat Plugin → WeChat | HTTPS | Long-poll `getUpdates` + `sendMessage` |\n| Permission Manager → AppContainer | Win32 API | AppContainer security capabilities + Job Object resource limits |\n\n---\n\n## Components\n\n| Component | Path | Stack | Description |\n|---|---|---|---|\n| **Desktop App** | `desktop\u002F` | Electron 33 + TypeScript + Vue 3 + Element Plus | Chat UI, Gateway lifecycle management, system tray |\n| **AppContainer Sandbox** | `appcontainer\u002F` | .NET 9 + Node.js preload hooks | Windows AppContainer launcher + permission hooks for tool isolation |\n| **Installer** | `deploy.py` + `deployer\u002F` | Python 3 + Tkinter | Wizard-style graphical installer (can be packaged as a single exe) |\n| **Skill Packs** | `skills\u002F` | Markdown + JSON + Python\u002FNode | Office, search, browser automation managed skills |\n| **WeChat Plugin** | `plugins\u002Fopenclaw-weixin\u002F` | TypeScript + OpenClaw Plugin SDK | WeChat channel integration |\n\n---\n\n## Quick Start\n\nMicroClaw targets **Windows 10\u002F11**. For most users, the only thing you need to bring is an LLM endpoint and API key.\n\n### 1. Install\n\n> **Note on releases:** The MicroClaw GitHub release page does not ship a pre-built installer binary. Users build the installer from source themselves using the steps below. This is intentional — see the [Releases](#releases) section.\n\nBuild the installer from source, then run it:\n\n> **Build prerequisites** (only needed to run `build.ps1`): **Node.js 22+** and **Python 3.10+**. Install the Python build dependencies with `pip install -r requirements.txt` (this includes PyInstaller, used to package the installer exe). End users running the packaged `MicroClawInstaller.exe` do **not** need Python.\n\n```powershell\n.\\build.ps1                                        # produces dist\u002FMicroClawInstaller\u002FMicroClawInstaller.exe\n.\\dist\\MicroClawInstaller\\MicroClawInstaller.exe   # run the installer wizard\n```\n\nThe installer handles the Windows-side setup in a single run:\n\n- Git for Windows (PortableGit → `~\u002F.openclaw-git`)\n- Node.js 22+ via the official signed `.msi` (per-machine install to `%ProgramFiles%\\nodejs\\`, UAC-elevated; an existing system Node ≥22.16 at that path is reused as-is)\n- OpenClaw Gateway (`npm install -g openclaw`)\n- Configures the npm registry mirror and V8 compile cache\n- Installs the MicroClaw desktop client, managed skills, AppContainer sandbox, WeChat plugin\n- Adds Windows Defender exclusions, creates desktop shortcuts (including one-click Uninstall)\n\nAfter install, launch **MicroClaw** from the Start menu or the desktop shortcut. The desktop app auto-starts the Gateway.\n\n### 2. Add your model credentials\n\nOn first launch, the desktop app opens a SetupWizard only when model configuration is missing. The required user input is your model connection:\n\n- Base URL\n- API Key\n- Model name\n\nYou can add or edit models any time under **Settings → Models**. Credentials are stored locally by the desktop app. If `.env` already contains `MODEL_BASE_URL`, `MODEL_API_KEY`, and `MODEL_NAME`, MicroClaw can prefill or auto-apply that configuration.\n\n### 3. Start working immediately\n\nThe default home screen is not empty. It ships with starter task cards such as daily news, desktop organization, and travel help, plus prompt suggestions for common work. The bundled and managed skills are already installed, so optional service-specific credentials can be added later only when a particular workflow needs them.\n\n---\n\n## Uninstall\n\n**One-click uninstall** — double-click the **Uninstall MicroClaw** shortcut on your desktop (or run `MicroClawInstaller.exe --uninstall`). This removes the desktop app, skills, sandbox config, shortcuts, and the Node.js \u002F Git \u002F OpenClaw dependencies that the installer set up.\n\n---\n\n## Desktop App\n\nThe Electron desktop app is the primary user interface:\n\n- **Chat Interface**: Built with Vue 3 + Element Plus, supports multiple sessions\n- **Gateway Management**: Automatically starts\u002Frestarts the local OpenClaw Gateway\n- **WebSocket Communication**: JSON RPC protocol with Ed25519 device authentication\n- **Skill Integrity Detection**: SHA-256 checksums on all skill files at startup, Ed25519 signature verification\n- **System Tray**: Runs in the background with status indicators\n\nDevelopment mode:\n\n```bash\ncd desktop\nnpm install\nnpm run dev\n```\n\n### Linting\n\nESLint is configured for both the main process and renderer. **Run lint before committing** to avoid CI failures:\n\n```bash\n# Desktop main process\ncd desktop\nnpm run lint\n\n# Renderer (Vue 3)\ncd desktop\u002Frenderer\nnpm run lint\n```\n\nTo generate SARIF output for CI integration:\n\n```bash\nnpm run lint:sarif    # outputs eslint-results.sarif\n```\n\n---\n\n## Sandbox Isolation\n\n`appcontainer\u002F` provides a Windows AppContainer-based sandbox for isolating tool execution:\n\n- **.NET launcher** (`AppContainerLauncher.exe`) runs child processes inside an AppContainer with restricted ACLs\n- **Node.js preload hooks** (`sandbox-preload.js`, `sandbox-fs-hooks.js`, …) intercept `fs.*` \u002F `child_process.*` calls and prompt the user for permission\n- **Sensitive-path shield** hard-denies access to `~\u002F.ssh`, `~\u002F.azure`, and other credential directories — no override\n\nSee [appcontainer\u002FREADME.md](appcontainer\u002FREADME.md) for details.\n\n---\n\n## Skill System\n\n### Built-in Skills (52)\n\nInstalled alongside OpenClaw and controlled via the `skills.allowBundled` allowlist. Categories include:\n\n| Category | Example Skills |\n|---|---|\n| Productivity | obsidian, notion, trello, slack, discord, things-mac |\n| AI \u002F Coding | coding-agent, gh-issues, oracle, skill-creator |\n| Communication | imsg, bluebubbles, wacli, voice-call |\n| Smart Home | openhue, blucli, sonoscli, eightctl |\n| Media | spotify-player, songsee, video-frames |\n| Utilities | weather, healthcheck, session-logs, model-usage, tmux |\n| Voice | openai-whisper, sherpa-onnx-tts, sag |\n\n### Managed Skills (7)\n\nInstalled to `~\u002F.openclaw\u002Fskills\u002F`, these are custom advanced skills included in this project:\n\n| Skill | Description |\n|---|---|\n| excel-xlsx | Create and edit Excel workbooks |\n| powerpoint-pptx | Create and edit PowerPoint presentations |\n| word-docx | Create and edit Word documents |\n| officecli | Office document CLI tool (create\u002Fedit .docx\u002F.xlsx\u002F.pptx) |\n| desktop-organizer | Scan and organize files on the Windows desktop |\n| desktop-beautify | Tidy desktop files and create sticky notes |\n| security-practice | AI agent safety practices (red\u002Fyellow line rules, install audit protocol) |\n\n---\n\n## WeChat Plugin\n\n`plugins\u002Fopenclaw-weixin\u002F` — Connects OpenClaw to WeChat:\n\n- QR code login, no username\u002Fpassword needed\n- Multi-account + sender isolation\n- Supports text, image, video, and file messages\n- Long-polling message updates\n\n---\n\n## Security Features\n\n| Mechanism | Description |\n|---|---|\n| **Skill Integrity Checks** | SHA-256 hashes + Ed25519 signatures — detects tampering of all skill files at startup |\n| **Device Authentication** | Each device generates an Ed25519 key pair; Gateway connections are signature-authenticated |\n| **Skill Allowlist** | `allowBundled` \u002F `allowManaged` control the available skill scope |\n| **Sandbox Isolation** | OS-native AppContainer + Job Object isolation for file system, network, and resource limits |\n| **Local Gateway** | Binds to loopback only — does not accept remote connections |\n\n---\n\n## Build\n\nFull build pipeline (PowerShell):\n\n```powershell\n.\\build.ps1\n```\n\nThis script sequentially:\n\n1. Builds the desktop app (`desktop\u002F` → Electron Builder)\n2. Creates portable zip package (`dist\u002Fmicroclaw-portable.zip`)\n3. Packages the installer exe (PyInstaller → `dist\u002FMicroClawInstaller.exe`)\n\n### Prerequisites\n\n- Node.js 22+\n- Python 3.10+ — install build deps with `pip install -r requirements.txt` (includes PyInstaller)\n- .NET 9 SDK (for the AppContainer launcher)\n- npm dependencies installed (`cd desktop && npm install`)\n\n---\n\n## Project Structure\n\nOperational Windows scripts now live under `scripts\u002Fwindows\u002F`. Root `.bat` and `.ps1` files are retained as compatibility wrappers so existing commands keep working.\n\n```\n├── deploy.py                    # Stable installer entry point (Tkinter GUI)\n├── deployer\u002F\n│   ├── config.py                # Configuration management (.env + YAML)\n│   ├── logger.py                # Thread-safe logger + in-memory ring buffer\n│   ├── skill_catalog.py         # 52 built-in + 7 managed skill catalog\n│   ├── skill_manager_ui.py      # Skill selector dialog\n│   └── windows_setup.py         # Windows install logic (Node\u002Fnpm\u002FOpenClaw)\n├── desktop\u002F                     # Electron desktop app\n│   ├── src\u002F                     # Main process (TypeScript)\n│   └── renderer\u002F                # Vue 3 renderer process\n├── appcontainer\u002F                # Windows AppContainer sandbox (.NET + preload hooks)\n├── skills\u002F                      # Managed skill definitions\n├── plugins\u002Fopenclaw-weixin\u002F     # WeChat channel plugin\n├── scripts\u002F\n│   ├── windows\u002F                 # Canonical Windows helper scripts\n│   ├── generate-skill-snapshot.js\n│   └── sync-studio-assets.js\n├── docs\u002F\n│   ├── architecture\u002F            # Repo structure and architectural decisions\n│   ├── plans\u002F                   # Planning docs\n│   └── reference\u002F               # Design and setup reference material\n├── build.ps1                    # One-click build script\n├── launch.bat                   # Compatibility wrapper -> scripts\u002Fwindows\u002Flaunch.bat\n├── setup.bat                    # Compatibility wrapper -> scripts\u002Fwindows\u002Fsetup.bat\n├── setup-dependencies.ps1       # Compatibility wrapper -> scripts\u002Fwindows\u002Fsetup-dependencies.ps1\n├── uninstall.bat                # Compatibility wrapper -> scripts\u002Fwindows\u002Funinstall.bat\n├── uninstall-dependencies.ps1   # Compatibility wrapper -> scripts\u002Fwindows\u002Funinstall-dependencies.ps1\n├── start-gateway.cmd            # Manual Gateway start\n├── MicroClawDeployer.spec       # PyInstaller packaging config\n└── requirements.txt             # Python dependencies\n```\n\nSee [docs\u002Farchitecture\u002Frepository-layout.md](docs\u002Farchitecture\u002Frepository-layout.md) for the structural rules and the long-term migration target.\n\n## Configuration\n\nMost users can do everything from **Settings** inside the desktop app — models, Brave API key, skill allowlist, etc. Runtime settings live in `~\u002F.openclaw\u002Fopenclaw.json`; you can hand-edit it if needed.\n\nFor developers \u002F unattended installs, the installer also accepts a `.env` file (`MODEL_BASE_URL`, `MODEL_API_KEY`, `MODEL_NAME`, `MODEL_API_FORMAT`, `MODEL_REASONING_EFFORT`, `BRAVE_API_KEY`) — see [.env.example](.env.example). `python deploy.py` reads `.env` from the repo root; the packaged `MicroClawInstaller.exe` reads `.env` from next to the exe.\n\n## System Requirements\n\n- Windows 10\u002F11\n- Python 3.10+ (only required to run the installer)\n- Network connection (mainland China mirror support included)\n- Optional: Microsoft Edge (browser skills)\n\n## Releases\n\nThe MicroClaw GitHub release page is intentionally **source-only**: it does **not** publish prebuilt installer binaries (`MicroClawInstaller.exe`, portable zips, or any other packaged artifacts). Users obtain MicroClaw by cloning this repository and running `build.ps1` themselves. This keeps the distribution channel transparent and auditable — you only ever run binaries you produced from source you can inspect.\n\nIf you fork this project, please honor the same policy unless you intend to take on responsibility for code signing, malware scanning, and supply-chain attestation of any binaries you publish.\n\n## Contributing\n\nThis project welcomes contributions and suggestions. Most contributions require you to agree to a\nContributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us\nthe rights to use your contribution. For details, visit https:\u002F\u002Fcla.opensource.microsoft.com.\n\nWhen you submit a pull request, a CLA bot will automatically determine whether you need to provide\na CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the\ninstructions provided by the bot. You will only need to do this once across all repos using our CLA.\n\nThis project has adopted the [Microsoft Open Source Code of Conduct](https:\u002F\u002Fopensource.microsoft.com\u002Fcodeofconduct\u002F).\nFor more information see the [Code of Conduct FAQ](https:\u002F\u002Fopensource.microsoft.com\u002Fcodeofconduct\u002Ffaq\u002F) or\ncontact [opencode@microsoft.com](mailto:opencode@microsoft.com) with any additional questions or comments.\n\nSee [CONTRIBUTING.md](CONTRIBUTING.md) for detailed contribution guidelines.\n\n## Security\n\nPlease see [SECURITY.md](SECURITY.md) for information on reporting security vulnerabilities.\n\n## License\n\nMIT — see [LICENSE](LICENSE) for the full license text.\n","MicroClaw 是一个让 OpenClaw 在 Windows 系统上一键安装和运行的工具。它通过将桌面客户端、本地网关、托管运行时、预加载技能及权限控制沙箱整合到一个产品中，简化了用户的配置过程。用户只需提供大语言模型（LLM）连接所需的API密钥等信息，MicroClaw 就能处理其余的安装与配置工作。该项目使用 TypeScript 编写，并采用 MIT 许可证开放源代码。特别适合需要在 Windows 平台上快速搭建并安全地使用基于 LLM 的应用程序或服务的场景，如开发测试环境下的 AI 功能集成实验。",2,"2026-06-11 04:05:22","CREATED_QUERY"]