DeepSeek Harness Plugins
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
dshx is a CLI tool for managing MCP servers, skills, and memory in DeepSeek Harness, featuring one-command add/remove, connection tests, secure configuration, and migration from other environments.
This plugin adds visual capabilities to text-only models in DeepSeek Harness by using free Gemini and GLM vision APIs for image description, OCR, and visual Q&A.
dsh-approval-gate is a DeepSeek Harness plugin that uses Flash model for risk prediction to implement automatic approval gate, auto-passing safe operations and routing dangerous ones to human confirmation.
dsh-browser is a Playwright-powered browser automation tool family for DeepSeek Harness, enabling agents to drive real web pages, fill forms, click through flows, scrape content, and take screenshots.
A visual Git workbench for inspecting repositories, running common actions, and safely executing AI-generated Git workflows in DeepSeek Harness Web.
The dsh-balance plugin queries the DeepSeek account balance and estimates the current session spend, with balance display and settings features.
A session manager plugin for DeepSeek Harness that lists all sessions and offers continue, outline, and permanent deletion.
A pure client-side CSS overlay that repairs mobile layout issues for DeepSeek Harness Web UI on narrow screens.
A DeepSeek Harness frontend plugin that replaces the model selection widget with a model list and a three-tier reasoning effort slider, featuring smooth interactions like continuous tracking, elastic overshoot, and snap-to-position.
A VS Code-style workspace keyword search plugin for the DeepSeek Harness web GUI, supporting file name and content search with integration into the sidebar.
A Feishu channel bridge for DeepSeek Harness that enables interaction with dsh agents through Feishu bots, supporting messaging, session management, and tool approval.
dsh-llm-fallback is a DeepSeek Harness plugin that provides automatic fallback to alternative providers when the primary one fails, ensuring uninterrupted model inference.