← 返回 Skills 市场
127
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install compuse
功能描述
Top-level cross-platform computer-use skill that bundles standalone macOS, Windows, and Linux runtimes with zero local Claude dependency and selects the corr...
使用说明 (SKILL.md)
Cross-Platform Computer-Use Skill
Use this skill when the task needs a top-level portable computer-use skill that works across macOS, Windows, and Linux without depending on any local Claude installation, private native modules, or extracted app assets.
What this skill does
- bundles standalone platform runtimes for
macOS,Windows, andLinux - installs one top-level skill that contains all three platform payloads
- selects the correct project for the current host platform
- preserves the standalone bootstrap model: each platform runtime creates its own virtualenv and installs public Python dependencies on first real run
- keeps platform-specific limitations explicit instead of pretending the hosts behave identically
Installed layout
After installation, assume the top-level skill lives at:
~/.codex/skills/compuse
The bundled projects are stored under:
~/.codex/skills/compuse/project/platforms/macos
~/.codex/skills/compuse/project/platforms/windows
~/.codex/skills/compuse/project/platforms/linux
Platform selection
Use the helper script from the installed skill root to resolve the active platform project:
bash ~/.codex/skills/compuse/scripts/current-project.sh
On PowerShell:
powershell -ExecutionPolicy Bypass -File $HOME/.codex/skills/compuse/scripts/current-project.ps1
Build
Always build the selected platform project, not all three at once:
cd "$(bash ~/.codex/skills/compuse/scripts/current-project.sh)"
npm install
npm run build
Run
cd "$(bash ~/.codex/skills/compuse/scripts/current-project.sh)"
node dist/cli.js
Validation notes
macOShas been real-device validated in this workspace, including GUI typing round-trip through the MCPtypetool.WindowsandLinuxare implemented, built, packaged, and published, but still need end-to-end runtime validation on real hosts.Linuxcurrently targetsX11first; Wayland can restrict screenshots, focus inspection, clipboard, and synthetic input.
Guardrails
- Treat this host as trusted-local only.
- Do not tell the user to search a local Claude install for binaries or hidden assets.
- Be explicit about the current host platform and its validation status before claiming something is verified.
- Mention that current runtimes report
screenshotFiltering: none, so action gating is handled at the MCP layer.
安全使用建议
What this means for you:
- This package appears to be what it says: a local, self‑contained computer‑use runtime that can take screenshots, list running/installed apps, and control mouse/keyboard. Those are powerful local operations — they can capture sensitive screen content and drive input devices — but they are consistent with the stated purpose.
- Before installing or enabling it: (1) only install on a machine you trust or a test machine; (2) review the included scripts (scripts/install.sh) and entry points (dist/cli.js, runtime/*_helper.py) to ensure no unexpected network endpoints or post-install steps; (3) be aware first run will create a Python virtualenv and pip-install packages from PyPI (network fetch); (4) check whether your deployment allows the agent to invoke the skill autonomously — if so, consider restricting autonomous invocation or gating actions in the MCP config; (5) if you need the least privilege, run it in an isolated VM or sandbox first and validate behavior (especially on Wayland/X11 differences).
- Additional checks that would increase confidence: verifying the upstream GitHub/ClawHub repos referenced in project/manifest.json, scanning the codebase for any outbound network calls or HTTP clients, and confirming there are no hard-coded remote endpoints or credentials in any of the truncated files.
功能分析
Type: OpenClaw Skill
Name: compuse
Version: 0.1.1
The skill bundle provides a cross-platform implementation of 'computer-use' capabilities (mouse/keyboard control, screenshots, and clipboard access) via a Model Context Protocol (MCP) server. It uses a TypeScript core to manage tool calls and a Python bridge (e.g., linux_helper.py, mac_helper.py, windows_helper.py) to perform OS-level automation using standard libraries like pyautogui, mss, and psutil. The pythonBridge.ts script handles the automatic creation of a virtual environment and installation of public dependencies. The code and instructions in SKILL.md are consistent with the stated goal of providing a standalone, portable runtime that does not depend on a local Claude installation, and no evidence of malicious intent or data exfiltration was found.
能力评估
Purpose & Capability
Name/description claim a cross-platform computer-use skill. The bundle actually contains macOS, Windows, and Linux platform projects (source and runtime helpers) and implements screenshot, mouse/keyboard, window/process enumeration, and app-listing logic — all expected for a computer‑use MCP runtime. No unrelated credentials or external services are required by the manifest.
Instruction Scope
SKILL.md instructs building and running the selected platform project and documents first-run behavior (creating a virtualenv and installing Python deps). The runtime code will access local displays, clipboard, processes, .desktop/installed-app metadata, and will drive input devices — these are within the stated purpose but are high‑sensitivity local operations. The instructions do not direct data to external endpoints, but the runtime will obtain desktop contents which an agent could forward if misused.
Install Mechanism
Registry shows no explicit install spec (skill is distributed with code included), so nothing is downloaded at install time by the platform. On first run the runtime bootstraps a Python virtualenv and pip-installs public packages from PyPI (mss, pyautogui, Pillow, psutil, python-xlib). That network activity is normal for a self-bootstrapping runtime but should be expected and reviewed. No obscure external download URLs or archive extraction from personal servers were observed in the provided metadata.
Credentials
The skill requires no credentials or special environment variables. The code reads optional orchestration env vars (e.g., CODEX_THREAD_ID, CLAUDE_COMPUTER_USE_* flags) which are appropriate for session/verbosity control. There are no unrelated API keys or secrets declared.
Persistence & Privilege
The skill will install files under the skill path (~/.codex/skills/compuse or similar) and the selected platform runtime will create a .runtime/venv on first run. always:true is not set. The skill can be invoked autonomously (default), which combined with desktop-control capabilities increases potential impact — this is expected for a computer-use MCP skill but worth user attention.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install compuse - 安装完成后，直接呼叫该 Skill 的名称或使用
/compuse触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v0.1.1
Fix macOS raw typing stability and add a cross-process bootstrap lock for runtime dependency setup.
v0.1.0
Publish the top-level cross-platform computer-use skill under the shorter public package name compuse.
元数据
常见问题
Compuse 是什么？
Top-level cross-platform computer-use skill that bundles standalone macOS, Windows, and Linux runtimes with zero local Claude dependency and selects the corr... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 127 次。
如何安装 Compuse？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install compuse」即可一键安装，无需额外配置。
Compuse 是免费的吗？
是的，Compuse 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Compuse 支持哪些平台？
Compuse 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Compuse？
由 wimi321（@wimi321）开发并维护，当前版本 v0.1.1。
推荐 Skills