yuqingsh/dsh-image-subagent
50Last commit Aug 14, 2026
dsh-image-subagent DSH plugin
dsh-image-subagent projects image attachments into explicit text placeholders, allowing a text-only main model to delegate image reading to a multi-modal sub-agent. It integrates with DSH's plugin system and requires a configured visual sub-agent with read_attachment/read_image tools.
How to install the dsh-image-subagent DSH plugin
dsh plugin --profile web add github:yuqingsh/dsh-image-subagent#v0.1.1Copying does not run this command. Review the repository and version before installing the dsh-image-subagent DSH plugin.
dsh-image-subagent DSH plugin data source
dsh-image-subagent DSH plugin snapshot date: Aug 16, 2026
discovered
What the dsh-image-subagent DSH plugin can do
- Makes text-only models capable of handling image attachments
- Automatically replaces image attachments with text placeholders containing attachment IDs
- Delegates image reading to a visual sub-agent (e.g., observer) with multi-modal capabilities
- Supports pixel-level image description via read_attachment/read_image tools
- Seamlessly integrates with DSH plugin system and profile-based activation
Where the dsh-image-subagent DSH plugin fits
- Analyze image content using a text-only base model by leveraging a visual sub-agent
- Enable non-image-supporting models to answer questions about uploaded photos
- Build multi-modal workflows where a text model orchestrates a vision specialist
- Automate image captioning or description generation in DSH sessions
Who the dsh-image-subagent DSH plugin is for
- DSH users who need to process images with text-only main models
- Developers building multi-modal agent pipelines with DSH
- Users who want to extend deepseek-v4-pro or similar models with vision capabilities
dsh-image-subagent DSH plugin limitations
- Requires a pre-configured visual sub-agent with image input declaration (e.g., MiniMax-M3)
- The sub-agent must have read_attachment/read_image tools in its toolset
- Sub-agent must be spawned/forked in the same session for attachment reading authorization
- Currently in early version (v0.1.1) and relies on pnpm for installation
- Only works in DSH web environment and requires restart after installation
dsh-image-subagent DSH plugin: from the repository README
Quoted from the yuqingsh/dsh-image-subagent README, the upstream source of the dsh-image-subagent DSH plugin. Copyright remains with the original authors.
让纯文本主模型（如 deepseek-v4-pro）也能接收图片附件：图片进入会话后投影为显式文本占位符，由主模型委托视觉子代理（多模态模型）读取。 ## 安装 ### 常规安装 ```sh dsh plugin --profile web add github:yuqingsh/dsh-image-subagent#v0.1.1 ``` 发布到 npm 后：`dsh plugin --profile web add dsh-image-subagent`。本地 checkout：`dsh plugin --profile web add ./dsh-image-subagent`。 需要 pnpm（`brew install pnpm`）。安装后重启 `dsh web` 并刷新页面。 ### 用 Prompt 安装 把下面这段发给你的 DSH 会话，让 Agent 代装： > 请安装 dsh-image-subagent 插件：运行 `dsh plugin --profile web add github:yuqingsh/dsh-image-subagent#v0.1.1`。完成后重启 dsh web 进程，刷新浏览器页面。 ## 使用 ### 前置条件（缺一不可） - 预设里有一个视觉子代理（如 observer），其模型声明 image 输入（如 MiniMax-M3）； - 该子代理的工具集包含 `read_attachment` / `read_image`； - 子代理须在本会话内派生（spawn / fork）——附件读取按会话日志授权； - 主模型委托时，把占位符中的 `attachmentId` 一并传给子代理。 ### 日常使用 贴图并附上问题 → 主模型收到占位符（含 `id=sha256:…`）→ 委托 observer 读图 → observer 用 `read_attachment` 返回像素级描述 → 主模型作答。 ### 验证安装 ```sh curl -s -X POST http://127.0.0.1:3080/image-subagent/status \ -H 'content-type: application/json' \ -d '{"type":"client-request","rpcId":"s1","method":"status","payload":{}}' ``` 预期 `bridged` 含 `"image"`，`real` 为路由真实声明。 ## License MIT
Read the full READMERepository license: MIT
dsh-image-subagent DSH plugin questions
How do I install dsh-image-subagent?
Run `dsh plugin --profile web add github:yuqingsh/dsh-image-subagent#v0.1.1` in your terminal. Make sure you have pnpm installed (`brew install pnpm`). After installation, restart the `dsh web` process and refresh your browser. You can also use the prompt method described in the README.
What are the prerequisites for using this plugin?
You need a visual sub-agent (e.g., observer) in your preset that declares image input (e.g., MiniMax-M3). The sub-agent must have `read_attachment` or `read_image` tools in its toolset. The sub-agent must be spawned/forked in the same session, and the main model must pass the attachmentId from the placeholder to the sub-agent when delegating.
Why is my image not being processed by the main model?
First, ensure the visual sub-agent is properly configured and spawned in the same session. Check that the sub-agent's model supports image input and has the required tools. Also verify that the main model is delegating the image placeholder to the sub-agent. If the issue persists, restart DSH web after installation.
Can I use any visual sub-agent with this plugin?
The sub-agent must declare image input capability and have `read_attachment` or `read_image` tools. The README mentions MiniMax-M3 as an example, but any multi-modal model meeting these requirements should work. Ensure the sub-agent is configured in your preset and spawned in the same session.
How do I verify the plugin is working?
Run the curl command: `curl -s -X POST http://127.0.0.1:3080/image-subagent/status -H 'content-type: application/json' -d '{"type":"client-request","rpcId":"s1","method":"status","payload":{}}'`. The response should show `bridged` containing `"image"` and `real` as the route's actual declaration.