dsh-computer-use：DeepSeek Harness 桌面操作插件
dsh-computer-use（v0.5.6）是 orangeofcarl0-sys 维护的 DeepSeek Harness 客户端插件，通过 cua-driver 实现观察→定位→自动升级投递→验证的桌面自动化。默认折叠仅暴露 computer_do 与 screen_observe，展开后提供 21 个工具；后台优先三级投递、窗口本地像素坐标、UIA 验证与子 agent 委派。安装依赖 Node≥22.18.0、DSH≥0.1.5-rc.2，使用 install.sh 链接 profile。设计为无感自治：零审批，密码框为唯一硬拒绝。
Read More电脑全自动控制：用 Rebind MCP 让 Agent 操作本机桌面
SkillHub 技能「电脑全自动控制」（@user_6be3c359/rebind-pc-auto-control，v1.0.0）基于 Rebind 可编程输入层与 @rebind.gg/mcp-server，面向 Windows/macOS/Linux。当 Agent 需要看屏幕、移动鼠标、点击、输入、滚动并操作窗口，以完成自动填表、批量处理文件、无 API 的 GUI 或跨应用流转时，该 Skill 提供 MCP 接线、环境自检与安全相关的全自动工作流说明，无需 OpenAI Codex 账号、本机运行。安装可遵循 skillhub.cn/install/skillhub.md，使用 skillhub install 并指定 Agent 的 skills 目录。
Read More电脑管家：让 Agent 直接操控本机音量、电源与系统设置
电脑管家（@user_ad872d55/pc-butler，v2.0.3）是 SkillHub 社区 ai-agent 类 Skill，无需 API Key。它将音量、亮度、锁屏、应用与网络、截屏、剪贴板、媒体、内存清理、关机重启等能力封装给 Cursor、Claude Code、Codex 等 Agent，并支持审计日志、可逆快照、电源计划、外接显示器 DDC/CI、合盖保活与健康巡检。文章说明定位、已核实功能、skillhub CLI 安装路径与典型对话触发方式，并提醒高权限与机型差异风险。
Read Moredsh-codex-computer-use：本地 Codex Computer Use 集成
Han-1413141 开发的 DSH 插件，通过本地 Codex Computer Use 运行时实现 Windows 桌面自动化。支持读取窗口、截图、点击、输入、滚动及启动应用等核心功能。安装需满足 Windows 系统、DSH 0.2.0-rc.2 及 Codex 依赖版本，首次操作需手动授权。插件在首次调用时自动启动进程，操作主体由 DSH 当前模型决定。
Read Moredsh-photocraft: PhotoCraft Image Editing and Desktop Automation for DeepSeek Harness
`dsh-photocraft` is a community plugin that bridges DeepSeek Harness (DSH) with the PhotoCraft image editor via the MCP protocol, enabling agents with image editing, layer management, and desktop automation capabilities. The plugin requires a PowerShell 7 environment; during installation, the official runtime must be downloaded first and its path configured. It supports Windows and macOS and uses secure rendering by default. It provides two toolsets, for headless and desktop interaction, and requires a unified workspace path configuration so that agents can process, preview, and export images.
Read Moredsh-windows-mcp
`dsh-windows-mcp` is a Windows desktop automation plugin for DeepSeek Harness that connects to the CursorTouch/Windows-MCP protocol, allowing agents to control the screen, mouse, keyboard, and system processes. It uses DSH’s built-in bridge and uvx to launch services; after installing dependencies, the “Computer Use” hard toggle must be enabled in the interface. It supports tools such as screenshots, clicks, clipboard read/write, and process management, excluding app launching and file system features by default.
Read Moredsh-codex-style-computer-use: Providing Desktop-Level Operation Capabilities for DeepSeek Harness
DeepSeek Harness has released the dsh-codex-style-computer-use plugin, designed to address the need for agents to write automation scripts when operating Windows desktop applications. Through native C# Workers, the plugin enables agents to “see” and “operate” the desktop, supporting window listing, screenshots, UI Automation tree observation, as well as click, keyboard input, and drag-and-drop actions. The plugin also supports batch execution sequences, interactive prompts, and manual stop controls, lowering the barrier to interaction. It is available only on Windows, can be installed via dsh-market, does not read user files, and writes only binaries and logs to a specific directory.
Read MoreBlue-Haired Little Maid: Transparent Windows Desktop Pet
Blue-Haired Maid is an open-source visual plugin for the Windows desktop, designed to provide visual feedback for DeepSeek Harness users. It features a transparent window, offers 106 hand-drawn animations, and supports local AI chat. Core capabilities include intelligent dialogue via Ollama, offline voice interaction, screen monitoring, and file operations. It is easy to install, allows behavior modes to be configured from the console, and can run without an internet connection. Licensed under the MIT License, it is well suited for users seeking visual feedback for automated workflows.
Read MoreDSH Desktop Operator: Windows Desktop Automation and Security Computer Use Plugin
`dsh-desktop-operator` is a Windows desktop automation plugin designed specifically for DeepSeek Harness. It packages an adapted Open Computer Use native runtime to support operations such as clicks, text input, and window management, and it exposes an MCP interface. The plugin prioritizes accessibility elements and supports confirmation for high-risk operations to ensure safety. Users must install the `.tgz` package using PowerShell and configure permission policies in the Agent Preset (for example, `highRiskActionPolicy: confirm`). It currently supports Windows only and depends on specific versions of Harness and DSH.
Read Moredsh-appshots: Capturing the Foreground Window and Injecting Context
dsh-appshots is an automation plugin designed for DeepSeek Harness Desktop, addressing the challenge of large language models understanding application interfaces. It supports macOS and Windows, accurately captures the frontmost window, adds the screenshot as a draft attachment, and extracts window text to inject into the model context. After installation, users can trigger screenshots with a keyboard shortcut; DSH automatically brings the window to the front, requiring no manual operation. It provides rich configuration options, including skipping its own window and controlling text injection. When using it, users should grant the system Screen Recording and Accessibility permissions. Text extraction depends on the system accessibility tree, so some terminals or protected content may not be readable.
Read Morewushi2333/dsh-computer-use_codex-style: Codex-Style Desktop Control Plugin for DeepSeek Harness
The `dsh-computer-use_codex-style` plugin in DeepSeek Harness is based on the Codex `window2` tool surface and provides DSH with a desktop automation solution without relying on Codex itself. It uses `SendInput`, UI Automation, and screenshot technology to support 13 window and application control operations. Installation requires the Bundle method, with the tool line configured in the Agent preset. The plugin adopts a two-layer architecture for security, supports Windows 10/11, is provided under the MIT license, and is suitable for implementing complex desktop interactions in DSH environments.
Read Morecomputer-user: DeepSeek Harness Computer Control Plugin
computer-user is a plugin within the DeepSeek Harness (DSH) ecosystem, designed to address the limited visual capabilities of text-only models. It enables text-only LLMs to read the screen via local image analysis tools and directly control the operating system using Win32 SendInput for mouse clicks, keyboard input, and window operations. The plugin supports core features such as screenshots, multi-monitor support, UTF-16 text input, and keyboard shortcuts. All operations are performed locally without external API calls, ensuring strong privacy protection. It is particularly suited for models such as DeepSeek V4 Flash, enabling a local observe–act–verify loop. With easy installation and headless mode support, it is an ideal choice for desktop automation and privacy-conscious use cases.
Read Moredsh-pc-pilot: A Windows Desktop and Browser Automation Plugin
`dsh-pc-pilot` is a DeepSeek Harness plugin running on Windows 10/11 that provides agents with integrated local desktop and browser operation capabilities. By leveraging the UIA accessibility tree, background synthetic input, and persistent Chromium CDP sessions, it addresses the issues of traditional automation stealing focus and obscuring windows. The plugin supports application and window management, basic interactions, and browser navigation, with advanced features such as identity recognition, state restoration, and batch execution. It runs with the current process permissions, does not support virtual machines or hidden sessions, and is designed to provide a secure and stable foundation for agent environment operations.
Read MoreIntroducing the dsh-desktop-automation Plugin: macOS Desktop Automation and Visual Closed-Loop
dsh-desktop-automation is a plugin that enables a dsh agent to directly control macOS desktop applications. It communicates with the Accessibility service over a local socket, providing mouse and keyboard control as well as window management capabilities, and integrates screenshot, image recognition, and visual localization features to form a closed-loop "screenshot–image recognition–localization–action" workflow. Installation requires granting Accessibility permissions and supports installation from a package, GitHub, or a local link. This plugin is suitable for automating non-browser applications (such as design software), GUI testing, and remote desktop control.
Read MoreComputer-Control DSH Plugin: Desktop Control and Security Protection
The computer-control plugin for DeepSeek Harness (DSH) is designed to provide agents with secure desktop control capabilities, addressing the high risks and limited semantic support of traditional pixel-level operations. Based on the accessibility tree, it provides screenshot, pointer, and keyboard control, and supports semantic actions and batch execution. It includes built-in safety mechanisms and supports dry-run mode. The plugin primarily supports Windows 10/11, interacts with DSH via the JSON-RPC protocol, and allows users to safely take over desktop tasks in a controlled environment.
Read Moredsh_omnivision: An OmniParser-Driven GUI Agent Plugin
OmniVision is a GUI agent plugin built on OmniParser, designed to address the challenge that models without native multimodal vision capabilities have difficulty understanding screens and operating interfaces. The plugin converts screen content into structured data containing text, icons, and coordinates, supporting screen recognition, desktop automation (clicking, typing, etc.), real-time Dock recognition, image parsing, and tool call logging. Installation requires conditions such as a Windows platform and a running OmniParser service, and it is primarily targeted at desktop task development scenarios for DeepSeek Harness.
Read MoreDeepSeek Harness Plugin: dsh-computer-use
`dsh-computer-use` is a native Windows desktop automation plugin for DeepSeek Harness (DSH), designed to provide a lightweight control solution. It directly calls the DSH Cordis tool registry and relies on `user32.dll` to perform operations such as taking screenshots, clicking, dragging, and typing, without relying on the MCP protocol, with the native approval service enabled by default to ensure security. The plugin supports sending screenshots to external vision models for analysis. Installation requires specifying the DSH configuration file and restarting; it is currently supported only on Windows, and it is recommended to use `/permission workspace-write` in conversations to prevent operations from being denied.
Read Moredsh-computer-use-windows: Windows Computer Use Plugin with Security Access Control
The difficulty in enabling agents to operate on Windows desktops lies not in a simple "click," but in three specific challenges: the screenshot captured by the model might be of the wrong window; custom-rendered interfaces like WeChat and Electron lack accessibility trees, rendering the `element_index` path unusable; and operations such as deletion or payment are irreversible without a confirmation step. Common approaches either rely entirely on guessing coordinates from screenshots or require the target application to provide a complete UI Automation tree; neither is stable in real-world desktop environments. Below, I introduce `dsh-computer-use-windows`, a plugin for DeepSeek Harness (DSH). DSH's philosophy is "everything is a plugin," and this plugin fills the gap for Windows.
Read Moredsh-computer-use: Teaching DeepSeek Harness to Operate Computers
dsh-computer-use is an MIT-licensed DSH plugin maintained by JasonWei04 that encapsulates the entire pipeline of "screenshot/recording → Xiaomi MIMO visual analysis → real keyboard/mouse/window/program control → autonomous loop" into a tool, enabling the pure text models of DeepSeek Harness to operate real Windows desktops to complete GUI tasks. The workflow is fixed into three steps: observe, analyze, and execute: `computer_screenshot` for screenshots (delegable to MIMO for static analysis), `computer_record` for recording frame extraction (to observe dynamic processes), `computer_action` for single-step operations (supporting clicks, drags, input, launching programs, etc.), while `computer_use` then autonomously loops...
Read Moredsh-computer-use-win: Add Windows Desktop Observation and Operation Capabilities to DSH
dsh-computer-use-win is a DSH plugin maintained by Yu-tao-Li, designed to provide computer operation capabilities for Windows desktops to AI agents. Based on the MCP stdio server and the PowerShell/UIA engine, this plugin offers 22 desktop tools to address the needs of agents for reading window structure, executing input, and managing window states. Its core capabilities include visual observation via the UIA accessibility tree and a three-level screenshot chain (PrintWindow, WGC, screen area), supporting control finding and coordinate-based OCR fallback; executing mouse and keyboard operations and window management; and featuring safety mechanisms such as coordinate homing and emergency stop failsafe. Installation requires a Windows system and Node.
Read Moredsh-computer-use: Adding Virtual Mouse for Human-Like Operations to DeepSeek Harness
dsh-computer-use is a DSH workflow plugin maintained by 988hj7tczd-oss, providing 11 model-friendly tools such as screen_observe and computer_click for harness-desktop based on cua-driver. It first observes the screen using the AX tree, then performs pixel-level operations like clicking, typing, and scrolling through an independent virtual cursor across macOS/Windows/Linux. This article introduces the core capabilities, install.sh installation steps, security configurations, and typical usage.
Read More