視覚ツール · DeepSeek Harness プラグイン — 45 ページ

Let image-incapable LLM models receive image messages in DeepSeek Harness (dsh): replace image blocks with text placeholders + local paths, then describe via a vision script (qwen)

DSH 图片理解桥接插件:挂 agent/pre-step,把消息中的图片块经视觉模型(默认智谱 GLM-4V-Flash 免费档)转成文字描述,让纯文本模型(DeepSeek)也能看图

Codex-style generated-image gallery and lightbox for DeepSeek Harness

Giving text models eyes to see in DeepSeek Harness.

Convert model-bound images to JPEG/PNG/WebP for LM Studio compatibility

DeepSeek Harness plugin: generate images via Aliyun Bailian Qwen-Image (bl CLI). Adds a generate_image tool that returns images as conversation attachments.

DSH 生图插件：在设置中配置多个生图供应商/模型，并在对话中通过 image_generate 调用当前选中的模型。

DeepSeek Harness Cordis plugin for Responses API image_generation: durable attachments and inline conversation rendering.

Session image payload governor for DeepSeek Harness: audit which images a session still ships, pick the ones to keep, and move the rest out of the model context.

DeepSeek Harness plugin: trims the oldest images from outgoing chat requests and, when the provider rejects the image count with HTTP 400, parses the limit from the error and retries with fewer images. · DSH 插件：裁剪即将发出请求中的历史图片，并在上游因图片数量返回 400 时解析该上限并按更少的图片重试。

Let the model render an image into the DSH web chat flow (QQ/WeChat style) via a show_image tool: the message content keeps only the path text, the image never enters model context, and the browser shows the picture inline, scrollable with history.

让文本模型（如 DeepSeek）在 DSH 里直接收发图片：图片落盘工作区、模型可见面自动替换为 mmx（MiniMax VLM）识别结果，用户侧照常显示缩略图。

DeepSeek Harness plugin: let text-only models receive pasted images, and analyze them with a built-in OpenAI-compatible vision tool

DeepSeek Harness Web GUI 输入框 📎 选图按钮:通过系统文件选择器添加参考图,绕开拖拽环境问题,复用官方附件管线(缩略图/上限校验/随消息上传)。Image-picker button for the conversation composer.

Multimodal plugin for DeepSeek Harness: understand images and generate images through configurable OpenAI-compatible or DashScope endpoints.

DSH 会话内图片预览服务：把工作区图片通过 HTTP 暴露给会话渲染、文本→图片渲染（双字体防乱码）、SVG→PNG 转换。AI 在回复里贴 markdown 图片 URL，用户直接在对话中看图。

Inline previews and clipboard copying for DeepSeek Harness read_image tool calls, including nested Code Mode calls.

Local image hosting for DeepSeek Harness: serves images from a preview directory over the web profile's own HTTP server so chat messages can render them inline, plus a preview_image tool that stages an existing local image and returns its URL.

Inline previews for the images an agent already produced in DeepSeek Harness: read_image attachments and local paths in any format (HEIC/TIFF/PSD/SVG/BMP/ICO/TGA/PNM/JP2 …), rendered through the official gallery and lightbox.

在 dsh Web 对话正文中直接渲染本地图片路径的插件（host 改写路径 + 极简客户端拦截点击）

DeepSeek Harness plugin: structured image analysis via vision APIs (read_image_mimo tool) with Web UI config card. Attachment input moved to dsh-input-enhancement.

Give DeepSeek Harness agents the ability to read images directly: a model-facing read_image tool that answers questions about an image through any OpenAI-compatible vision endpoint.

DeepSeek Harness plugin: analyse images out of band with a vision model — pasted images are digested into text before admission and a describe_image tool covers image paths, all without ever changing the session's model.

User-tunable inline image display for dsh web sessions: a larger, auto-unrolling read_image tool view behind an image-settings section.