DSHEssentialTools
在终端中运行以下命令:
dsh plugin install LLYlab/DSHEssentialTools
将以下提示词粘贴到 DeepSeek Harness 对话框中:
在 DeepSeek Harness 终端中执行 dsh plugin install LLYlab/DSHEssentialTools,插件源码位于 https://github.com/LLYlab/DSHEssentialTools ,安装后重启 DSH 即可在 Settings → Plugin inventory 中看到该插件。
插件介绍
Working on real code with an AI agent in the loop, four things grind you down: one wrong message and you lose the whole thread; the AI cannot see your logged-in dashboards or internal pages; build-run-inspect means Alt-Tab forty times an hour; and the bill lands at month-end as a surprise. DSHEssentialTools attacks all four at once.
Two axes carry the plugin. On the conversation side, the VTD (Virtual Tree Dialogue) lets you edit or retry any user message into a hidden branch, hop between branches at will, and the workspace code snapshots and restores in lockstep so your original thread stays pristine. On the browser side, an MV3 extension bridges the model to your already-authenticated Edge or Chrome over a localhost-only WebSocket, exposing read and write tools for internal tools, webmail, and file:// pages, while a four-position dial in the extension popup keeps the keys firmly in your hand and the DSH host can only read it, never override it.
The right-hand toolbar folds build, file tree, preview, and program-level snapshots into the chat window. A global plugin manager offers five enablement tiers, GitHub direct import or an AI rewrite that never executes third-party code, and a Boot Guard that auto-disables all globals after repeated startup failures so a bad plugin can never brick your session. MDA, CDM, and TCT layers add cross-conversation memory, workspace-scoped retrieval, and one-shot scratch chats that destroy themselves on close. A corner status card tracks your DeepSeek balance, per-model unit price with peak and off-peak pricing, and per-turn cost in real time, with the API key living only in a request header, never written to disk or sent to the client. A single master switch restores pure stock DSH in one click, unloading every DET tool, injection, and bound plugin, and is fully reversible because every registration is held as a disposer.
Built for Windows developers running DeepSeek Harness on real projects who want the AI to operate their live browser sessions, iterate conversations without losing context, juggle a plugin library, and keep a running eye on token spend. Zero-config out of the box for branching and cost tracking; add a project path to unlock the run toolbar and file tools.
使用场景
- 让 AI 直接操控已登录的浏览器完成内部系统操作
- 对某条消息不满意时编辑重试,无需从头开始对话
- 在聊天窗口内直接运行项目、查看文件树和回滚代码
适合人员
- 使用 DeepSeek Harness 进行实际项目开发的 Windows 开发者
- 需要 AI 协助操作内部系统或已登录网页的工程师
- 重视 token 成本可控性和对话上下文管理的重度用户