dsh-win-minimal: DeepSeek Harness Minimal Mode on Windows
The `dsh-win-minimal` plugin provides an alternative to the `minimal preset` for DeepSeek Harness (DSH) on Windows. It retains the one-line fixed persona, zero runtime context, and automatic compression, and integrates the three tools: gitbash, str_replace_editor, and web_search. After installation, users can select “Minimal Mode (Windows)” when creating a new session. The plugin can be installed via CLI or DSH Desktop, with files located in the specified directory. It allows users to customize the persona and tools, but requires gitbash to be installed globally.
Read Moredsh-web-search-all-modes: Presetting the web_search Tool for dsh Agent Recovery
dsh-web-search-all-modes is an MIT-licensed plugin within the DeepSeek Harness (DSH) ecosystem, designed to address the issue of inconsistent tool sets across different agent presets. By re-enabling the host-side `tool-web` configuration, this plugin allows DSH agent presets such as minimal mode, which do not natively include `web_search`, to invoke this tool without affecting presets that already have this functionality. Prerequisites include the profile being built on `@deepseek-ai/dsh-base`, along with proper configuration of `DEEPSEEK_API_KEY` and the host search provider. Installation can be performed via local source code or GitHub commands, and a restart is required after installation.
Read Moredsh-minimal-vision: On-Demand Visual Assistance in DSH Minimal Mode
dsh-minimal-vision is a DSH community plugin designed to address the pain point of lacking visual evidence support in minimal mode. It is maintained by Flora233333, licensed under MIT, and requires Node >=20. The plugin enables on-demand visual capabilities by concealing context and Bash calls to `dsh-vision` without registering new chat models or modifying the system prompt, exposing only `persistent-bash` and `str_replace_editor` to the model. It supports processing images in PNG, JPEG, WebP, and GIF formats, as well as documents such as PDF, DOC, and PPT. Images are processed using `sharp` before being sent to the vision interface, while documents are handled by Agen.
Read More