Vision tools · DeepSeek Harness plugins — Page 54

Frosted-glass skin for the DeepSeek Harness web UI: pick any image as background, theme colors extracted from it, glassmorphism surfaces.

DSH 自定义皮肤插件：照片/视频壁纸（含 Steam 壁纸引擎导入）+ 从壁纸自动取色 + 分组半透明毛玻璃 + 壁纸文件化轮询热更新。改进自 dsh-ui-appearance。

Create, manage, and share portable DeepSeek Harness skins.

《犹格索托斯的庭院》特莉波卡 (Tlipoca) GAL 视觉小说沉浸式皮肤插件 for DeepSeek Harness

Automatic model routing for DeepSeek Harness: a virtual `smart` model classifies every request by difficulty (hard / normal / easy) and by vision need, then delegates to the models you already configured in Settings → Models. 内置智能路由：三级难度 + 视觉自动路由，模型从「设置-模型」已配置模型中选取，虚拟 smart 模型零配置接入。

Sprite Sheet Generator with AI Image Generation

Remote SSH workspace and closed-loop deployment Agent for DeepSeek Harness: SSH Files, terminal, remote editing, zero-to-one Bootstrap, Runbooks, automation, Vision/OCR and safe recovery.

给 dsh 的聊天表情包库：收图去重、视觉打标、按含义挑图；删了真删（哈希黑名单），脏话与烂梗在入库那一刻拦掉。

DeepSeek Harness hybrid attachment plugin: native image attachments plus bounded-memory non-image streaming into a verified session workspace, with quotas, trust fencing and TTL cleanup.

dsh bundle: subagent_vision — delegate image reading to a vision-capable model from a text-only session, plus paste-to-path so pasted images reach the subagent as file paths.

Connect an official Grok subscription to DeepSeek Harness as an ACP subagent, with multimodal media tools.

Native DSH audio/video cards and a compliant ACP bridge for subscription-backed multimodal work.

Paste or drag non-image files (xlsx/csv/pdf/zip…) into the DSH web composer: files are saved into the session workspace and an @reference is inserted.

DSH client plugin — flip to the back of the page for a floating glass task console: live background jobs, subagents, session overview and workspace of the current session, with mouse-draggable and mouse-resizable cards (bottom-right handle, size persisted per card). Maximizing a card unrolls its controls onto the card's own canvas, where every control is a draggable, editable item and the board card stays a live thumbnail whose control order follows that arrangement; built-in cards open as a fullscreen view. The top-right card conversation calls a temporary agent session to refactor each card's style and internal controls from natural language, and any conversation can author cards via [taskcard] blocks (text/heading/note/stats/progress/trend/kv/links/chips/checklist/counter/button/table/code/toggle/countdown/bars/image/embed). Embedded web apps can be marked 受控 and routed through the bundled local page bridge (bridge/server.mjs), which injects an agent into the page so the workbench — and the agent — can click, type, scroll and read inside it through [page] action blocks; the app height is dragged on its bottom edge, accurate on the zoomed canvas too. Cards can be published as copies onto a fullscreen infinite canvas with named boards, share strings and a zero-dependency WebSocket relay for multi-user editing, and a right-hand dock renders the project's archify architecture diagram (embedded, sandboxed, resizable).

测试工作台 Profile for DeepSeek Harness — 组装 UI 验证、会话分析、视觉路由为一体的 QA Agent。

把长文本渲染成图片发送来压缩 LLM 输入 token（DeepSeek 视觉模型、每图 384 token 封顶）— text-as-image token compression for DeepSeek Harness

Configurable skin for the DeepSeek Harness web UI

DeepSeek Harness (DSH) Web 界面自定义主题插件：七区域背景（主界面/侧边栏/输入区/设置界面/浮窗/Cordis，纯色/图片/选区裁剪/透明度）、文字颜色 5 类、框线颜色、标志调色、对话区细节定制、输入区统计条接管、浮动主题面板、预设导入导出（.tczp 含图）、localStorage 持久化。

The vn-harness Web GUI's conversation-header package. It owns three controls on that header. The Screenshot control (left of the Themes control) captures the whole window - the app is a fixed-viewport shell, so its 100% width and 100% height are exactly the tab's box - and saves the PNG to the Desktop of the machine running the app: the browser takes the pixels (getDisplayMedia on the current tab, one frame into a canvas) and this package's own host route writes the file, falling back to an ordinary browser download when that route is absent. The Themes control switches the app between the light, dark and system appearance, the same preference Settings > General > Appearance owns, showing which one is active. And the Session-log download seat replaces the shipped three-dot 'more actions' button (its only menu item was Download session log) with one plain download icon button that starts the export on the first click while the shipped sessionLogDownload controller keeps doing the work, its preparing/success/error dialog included. It also carries the pack's appearance overrides: the always-light paper the rendered Markdown view is drawn on, the chrome rule that hides the preview header's viewer menu on Markdown tabs so the page has exactly one renderer, the left column's top bar (the branding row becomes the same 76px band, ending in the same hairline, that the middle and right columns open with), the pack's VN branding on that row - the shipped mark and wordmark are replaced by a 24px black disc and the text VN Harness - and the header ring, so the right bar's own collapse/expand toggle in the header corner wears the same round hairline outline as this package's own header buttons. Alpha.

Reasoning layer for the official DeepSeek Harness. Every outgoing request stays pinned to native max effort, with isolated Flash/Vision/Pro depth controls, stability axes, reasoning toolboxes and a metacognition loop built above it. Delivered build only; commercial operation is allowed, reverse analysis, modification, extraction and re-upload are not - see LICENSE.

Automatic tier-based model routing for DeepSeek Harness: a virtual `smart` model classifies each request by difficulty (hard / normal / easy) and by vision need, then delegates it to the models you already configured. 三级难度 + 视觉自动路由，虚拟 smart 模型零配置接入。

DSH profile bundle: control your real, logged-in browser through TMWebDriver (Chrome extension bridge)

Model-facing accurate_vision tool: precise image spatial reasoning via a vision model

DSH plugin: describe_image tool + paste-image-to-text browser half, bridging image understanding to text-only models via DashScope qwen-vl