deepseek-harness-minimal-omni
Run the following command in DeepSeek Harness:
dsh plugin install Rainflowers686/deepseek-harness-minimal-omni
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install Rainflowers686/deepseek-harness-minimal-omni inside your DeepSeek Harness environment to install this plugin; source code is available at https://github.com/Rainflowers686/deepseek-harness-minimal-omni .
About this plugin
Many teams pile Harness tool panels high with browser, document parsing, and code navigation all loaded at once. Idle sessions burn resources and concurrent tasks interfere with each other. Minimal Omni takes the opposite approach: at idle, only the smallest tool shell is present. When a task genuinely needs a browser, document extraction, or repository-level code navigation, the session issues a single capability request, and releases it the moment the task finishes. The runtime injects no planner and no review loop; the model brain remains the official Minimal, and the only change lives in the tool supply layer.
The capability matrix covers the most common day-to-day needs: on-demand Browser and Web fetch, PDF/DOCX/ZIP document extraction, Code/LSP repository navigation, a Jobs queue, native Goal and manual /compact, session hot-reload, and an integrated coding path. XLSX/PPTX model-facing workflows, Goal continuity across compaction, automatic compaction tuning, and optional media extraction are marked Experimental. Vision is conditional on whether the selected route supports image input, while Web Search and authenticated GitHub actions require their own provider or authorization. The capability boundary is explicit: anything not done is clearly labeled Deferred or Conditional, eliminating the grey zone of silent failures.
Who is it for: if you run DSH v0.1.2-rc.1 on Windows and want daily sessions to stay lightweight, pulling heavy tools only when needed, without touching existing provider settings, model selection, or credential management, Minimal Omni is the independent developer preview built for that scenario. One command layers it onto your current Web profile, and one command removes it cleanly with no residual state.
Use Cases
- Lightweight coding sessions that occasionally need to fetch a page or parse a PDF document
- Concurrent multi-task workloads where per-session tool isolation keeps resource use predictable
- Teams with a shared provider setup that still want task-specific tool panels
Best For
- Developers on Windows running DSH v0.1.2-rc.1
- Sole developers or small teams that prefer a lean default Harness profile without tool bloat
- Users who want to extend tool capabilities without touching provider, credential, or model settings
Related Plugins
A unified suite combining hot runtime injection, task-aware thinking-mode routing, and a graded session protocol with red-team gates to sustain model diligence across long-horizon inference.
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.