視覚ツール · DeepSeek Harness プラグイン — 38 ページ

dsh 插件 · 设置界面自定义：调整原生设置窗口大小（全屏 / 预置大尺寸 / 自定义宽高）、背景透明度与背景（主题 / 自定义颜色 / 图片），悬浮球即开即调，存本机浏览器

DSH Web GUI 新闻插件：侧边栏入口 + 弹窗聚合浏览国际新闻、知识科普、历史科普、AI 大模型新闻。Host 端 RSS 抓取/正文提取/图片代理（SSRF 防护 + 缓存），Client 端弹窗三视图（列表/阅读/设置）。

Unified OpenAI Codex subscription plugin for DeepSeek Harness: GPT models, OAuth, quota, image generation, and web search.

AI 生图 (image generation) plugin for the dsh web GUI: text-to-image and image-to-image through a configurable OpenAI-compatible endpoint, with multi-provider support, model discovery, and agent tool registration. 与 @xiaoyuink/dsh-image-vision 同系列的图像插件。

图片识别插件：自动判断当前模型是否具备视觉能力，有则用当前模型并按插件预设提示词分析，无则调用插件配置的视觉模型。文本模型可直接在对话框粘贴/上传/拖拽图片，发送时自动写入 DSH 附件存储（永久），消息区渲染缩略图，模型自动调用 image_vision / ocr / ground / crop 系列工具识别与精读；模型选择器保持简洁。

DSH（DeepSeek Harness）文件树侧边栏插件：右侧栏展示当前工作区文件树（主题半透明、emoji 图标、可折叠），双击文件在主窗口「文件预览」tab 内预览（代码语法高亮 / 图片 / 大文件提示），多文件二级 tab 栏 + 关闭按钮。Durable profile plugin for DeepSeek Harness.

Authenticated image paste, workspace attachment storage, and DSH session injection

自包含的 DeepSeek Harness 视觉系统：DeepSeek 视觉孪生路由（原生图片体验 + 视觉桥请求层拦截）+ 本地 Ollama Agentic Vision 工具（describe/OCR/结构化扫描/区域查询/元素定位/双图对比/剪贴板）+ 粘贴分流（paste-to-path）。安装即用，无本机路径依赖。

Model-facing view_image tool over the DeepSeek Harness vision seam (ctx.vision): reads an image file, commits its bytes durably, and returns a text description from an external vision provider.

Vision capability seam for DeepSeek Harness: a provider registry with auto-selecting describe() execution, letting a text-only model 'understand' an image.

OpenAI-compatible chat-completions vision provider for the DeepSeek Harness vision seam (ctx.vision).

Hover-to-download buttons for images in DeepSeek Harness conversations.

DeepSeek Harness (dsh) plugin for wallpaper theming — upload images from the Settings page, pick one, and the whole dsh web UI renders over it with an adjustable readability mask and blur.

VS Code-style file explorer plugin for the dsh web GUI: file panel beside the sidebar, multi-tab preview panel (code/images/PDF/audio/video/Markdown/CSV/Excel/Word/zip), filename search, sort/filter, copy-paste, swap sides and drag-resize — zero patches to official packages

A headless-first macOS Computer Use driver for DeepSeek Harness with AX-first observations, window-bound multimodal evidence, host-owned approval, and evidence-backed action receipts

Let text-only models see images in DeepSeek Harness: intercepts the llm/stream waterfall and transparently substitutes each image with a vision model's description.

Permanent 4D4Y forum browser: fetches the real Discuz 7.2 interfaces (index / forumdisplay / viewthread), decodes GBK, parses, and renders a browser panel beside the sidebar with a sidebar entry button.

A vision bridge for text-only DeepSeek Harness models: registers an `analyze_image` tool that answers questions about images via ANY OpenAI-compatible vision/multimodal endpoint (SiliconFlow, DashScope, Zhipu, OpenRouter, Ollama, ...).

A vision bridge for text-only DeepSeek Harness models: registers an `analyze_image` tool that answers questions about images via ANY OpenAI-compatible vision/multimodal endpoint (SiliconFlow, DashScope, Zhipu, OpenRouter, Ollama, ...).

DSH（DeepSeek Harness）图片一体化插件：粘贴图片只进对话不进模型请求（llm/stream 清洗），并注册 see_image 工具让文本模型通过任意 OpenAI 兼容视觉 API 按需看图

Per-model capability declaration for DeepSeek Harness: reasoning-effort levels (with wire spellings) and request modalities (text/image) for OpenAI-compatible providers — one settings section, no YAML hand-editing.

面向 AI Agent 的 Markdown 转 Word 工具，提供开箱即用的中文排版、Mermaid 图表转图片和可编辑数学公式，提供Skill、独立 CLI 和 DSH 插件。

Windows computer use for DeepSeek Harness: an MCP stdio server + PowerShell UIA backend, bridged into DSH via @deepseek-ai/dsh-mcp-client. Read/act on real Windows desktop apps (UI Automation tree, screenshots, typed input, OCR, window management).

Codex-style computer use for DeepSeek Harness (DSH): read the screen and drive the mouse & keyboard. Pairs with picturereader (image_scan/image_ocr) to close the look-act-verify loop. Windows.