ThreeBody6666/dsh-computer-use 预览 preview

ThreeBody6666/dsh-computer-use

Plugin插件 Native原生 ⭐ 2 MIT Vision & Media视觉与多媒体

DeepSeek Harness 的原生 Windows 电脑使用及可配置视觉工具。

Project Overview项目介绍

This is a native DSH plugin that provides Windows-native computer control functionality built exclusively for DeepSeek Harness. It integrates directly with DSH’s Cordis tool registry and does not require MCP to work, and has been verified to work correctly with DSH version 0.1.1-rc.1. It offers tools for screen capture, pointer control, text and keyboard input, waiting for apps, and extra visual analysis of the desktop. The recommended install method is via npm: run dsh plugin --profile web add @crazy_th/dsh-computer-use, replacing web with your actual DSH profile name. After installation completes, you must restart DSH for the plugin to load.

The plugin includes two independent security gates to prevent unwanted computer actions. A master toggle switch in the DSH UI lets you completely disable all computer-use tools with one click. By default, every computer action requires explicit user approval before execution, and this default setting is enabled by the bundled cordis.patch.yml file. If approval is set to never for the session, all tools will automatically reject calls, and you can fix this by running /permission workspace-write in DSH chat to restore approval prompts.

If you want to use the computer_observe tool, you will need to configure an OpenAI-compatible vision API after installing the plugin. To do this, open DSH Settings > Plugins > Plugin configuration, expand Computer Use Vision, then enter the endpoint URL, API key, vision model name, and max response token limit. The API key is stored as a write-only secret in DSH, so it is not exposed in plaintext configuration. The plugin is Windows-only, uses user32.dll and System.Drawing, and is released under the open source MIT license.

这是一款专为DeepSeek Harness打造的原生Windows平台电脑操控插件,直接调用DSH的Cordis工具注册,不依赖MCP协议,已在DSH 0.1.1-rc.1版本验证通过。它提供截屏、鼠标点击拖拽滚动、键盘输入与快捷键发送、等待应用加载、额外视觉观测等多项能力,默认所有操作都需要经过DSH原生的用户确认流程。

这款插件适合需要让DeepSeek Harness直接操控本地Windows桌面,完成可视化操作、图形界面交互任务的用户,常见场景包括自动化GUI操作、图形界面程序调试、截图视觉分析等。使用时,当DSH自带模型不支持图像识别,可调用配置好的第三方视觉模型对桌面进行观测分析,提升复杂界面识别能力。

插件当前仅支持Windows平台,安装后需要重启DSH才能生效,支持通过npm或源码两种方式安装,推荐使用npm命令,安装时需要指定对应的DSH配置文件。使用观测功能需要配置OpenAI兼容的视觉API地址、密钥和模型名称,密钥以密文存储,整款插件采用MIT许可开源。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 1 warning1 项注意
  • Only 2 stars - very few users, little community feedback星标只有 2,几乎没人在用,遇到问题缺少社区反馈
DSH walks through these 9 checksDSH 会逐条核对这 9 项

Compatibility兼容性

  • DSH, Node, OS and profile requirementsDSH 版本 / Node 版本 / 操作系统 / profile 是否满足要求
  • External dependencies and runtimes (Electron / Python / Docker, ...)外部依赖与运行时(Electron / Python / Docker 等)是否齐备
  • Conflicts with installed plugins: command names, skill / tool names, ports, duplicate MCP registration与已装插件是否冲突:命令名、skill / tool 重名、端口占用、重复 MCP 注册

Security安全性

  • Repo matches the facts registered here; archived or abandoned?仓库是否与页面登记一致,是否归档或长期停更
  • Safety of preinstall / install / postinstall and install.sh / setup.ps1preinstall / install / postinstall 与 install.sh、setup.ps1 是否安全
  • curl|bash, download-then-execute, obfuscation, unrelated domains → stop immediatelycurl|bash、下载即执行、混淆代码、无关域名 → 立刻停止
  • Typosquatting or unmaintained packages among the new dependencies新增依赖里有没有 typosquatting 或无人维护的包
  • Requested permissions vs. what the feature actually needs申请了哪些权限、是否超出功能所需(filesystem / network / shell / clipboard)
  • Any sudo / admin requirement, plus uninstall and rollback是否要求 sudo / 管理员权限,以及卸载与回滚方式

Anything uncertain must be marked unknown with a note on how to confirm it. This site's signal screen is a static snapshot, not a security audit.拿不准的必须标「未知」并说明要我怎么确认。本站的信号筛查是静态快照,不能替代安全审计。

Or use CLI install (for developers)或使用命令行安装(适合开发者)

CLI Install命令行安装

dsh plugin --profile web add @crazy_th/dsh-computer-use

把 ThreeBody6666/dsh-computer-use 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

DSH Computer Use

中文

Native Windows Computer Use bundle for DeepSeek Harness. It uses the DSH Cordis tool registry directly and does not use MCP. Verified against DSH 0.1.1-rc.1.

Capabilities

  • computer_screenshot: captures the virtual desktop and returns a DSH image attachment.
  • computer_click, computer_drag, computer_scroll: pointer control using bounded virtual-desktop coordinates.
  • computer_type, computer_keypress: Unicode text and supported keyboard shortcuts sent to the active control.
  • computer_wait: allows an application to settle before the next observation.
  • computer_observe: captures the desktop and sends it to a user-selected vision model when the active DSH model cannot inspect images.

Every Computer Use tool asks through DSH's native approval service by default. CTRL+ALT+DELETE, empty text, unsupported keys, invalid durations, and coordinates outside the virtual desktop are rejected before input is sent.

Permissions and approvals

If a computer_* tool reports the user rejected tool "computer_screenshot" without any prompt: this usually is not a real user refusal. When the session runs under approval: never (for example a danger-full-access preset), approval-required tools are fail-closed and auto-rejected, and the error text is misleading. Type /permission workspace-write (approval = ask) in the conversation input box to restore the approval prompts.

Two independent gates guard every computer action:

  • Master switch (since 0.2.2): the Computer Use: On/Off pill on the right side of the composer toggles the computer-use-control setting. When disabled, all computer_* tools are denied with an explicit reason.
  • Approval gate: tools/pre-execute returns an ask request for every computer_* call while requireApproval is true (the bundled cordis.patch.yml keeps it enabled). Turn it off only if you accept running computer actions without confirmation; changing it requires a DSH restart.

Vision model settings

After installing and restarting DSH, open Settings > Plugins > Plugin configuration, then expand Computer Use Vision. Enter an OpenAI-compatible HTTPS base URL, API key, vision-capable model name, and response-token limit, then enable the model. The API key is stored in DSH's settings secret field and is not included in Cordis configuration or tool responses.

Showing the opening section of the README — the full document lives in the repository以上为 README 开头摘要,完整文档在仓库内 · View the full README on GitHub →在 GitHub 查看完整 README →

← 上一个 Prev dsh-gmail-plugins 下一个 Next dsh-high-fee-alert →