dsh-with-pencil

Configuration

visionMode (Settings → Plugins → dsh-with-pencil, default text):

• text — for DeepSeek and other non-multimodal models. Screenshots route to high-resolution rendering so image transcription stays reliable. Image transcription itself is not provided by this plugin: it depends on the deployment's image-capable wrapper (e.g. dsh-vision-proxy or the wrapper / stealth route from dsh-vision-router). A direct text-only Harness route rejects chat image input before this plugin runs and cannot consume image tool results.
• multimodal — native screenshots; the model sees the pixels itself.

The setting card lives in Settings → Plugins → dsh-with-pencil and takes effect immediately after saving. First installs default to text.

For implementation details see docs/ARCHITECTURE.md.

配置

visionMode（设置 → 插件 → dsh-with-pencil，默认 text）：

• text — 适用于 DeepSeek 等非多模态模型。截图自动走高清渲染，保证图片转译的可靠性。 本插件不提供图片转译模块：转译依赖声明图片能力的包装路由（如 dsh-vision-proxy 或 dsh-vision-router 的 wrapper/stealth 路由）。Harness 的直连 纯文本路由会在本插件运行前拒绝聊天图片，也无法消费工具返回的图片结果。
• multimodal — 使用原生截图，模型自己看像素。

配置卡片位于 设置 → 插件 → dsh-with-pencil，保存后立即生效；首次安装默认 text。

实现细节见 docs/ARCHITECTURE.md。