DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Provides reproducible and deterministic benchmark evidence for DeepSeek Harness tools and plugins, using fixed cases and versioned scoring to ensure repeatable and objective execution results.
dsh-switch is the evidence-first model control plane for DeepSeek Harness, providing model route inspection, validation, and standalone CLI endpoint probing tools.
dsh-music-plugin adds local music playback control to DeepSeek Harness, allowing the agent to search, play, pause, and skip songs with a web panel.
dsh-zh-output provides Chinese presets for DeepSeek Harness that enforce Chinese thinking and output, with a dedicated Chinese compaction engine.
A local offline voice input plugin using FunASR engine for accurate transcription, with optional LLM polishing and automatic fallback to browser Web Speech API.
dsh-vision enables pure text models to automatically see images by bridging them to text descriptions generated by a vision model.
Provides capability-aware reasoning profiles and provider-native Fast mode for the DeepSeek Harness web composer, enabling multi-provider fast switching and reasoning configuration.
A DeepSeek Harness plugin that translates reasoning and final answers in the Web UI, supporting Chrome's built-in Translator or OpenAI-compatible APIs.
A DeepSeek Harness skill plugin that provides Qian Xuesen's cybernetics methodology to help agents control complex projects.
This plugin converts UI screenshots into implementation-oriented structured specifications to help a text-only DeepSeek model recreate responsive web interfaces.
A minimal native vision enhancement plugin that supports the free Zhipu API and the local Qwen-VL 2B model.
A plugin for DeepSeek Harness that automatically generates a suggested next prompt after each AI turn, displayed as ghost placeholder text in the input field, and adopted by pressing Tab.