DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Independently assign models for the main conversation, subagents, and plan execution in DeepSeek Harness, eliminating frequent switching and automatically restoring the main model afterward.
dsh-spend is a floating token usage and cost monitor for DeepSeek Harness with zero-config billing plan detection, real-time multi-dimensional stats and quota display.
An AI entertainment skill distilled from over 22,000 real live-stream comments, generating CS/Dota commentary, player roasts, and signature meme replies in the distinctive Wan Jiqi style.
A deterministic BaZi and Zi Wei Dou Shu chart calculator for DeepSeek Harness that supports dual-chart generation, handles uncertain birth hours via candidate sets, and prompts for missing data to deliver structured insights to AI agents.
An intelligent prompt optimizer that automatically rewrites casual inputs into professional, structured, and ready-to-use prompts with context awareness and iterative refinement.
An unofficial DeepSeek Harness plugin integrating the Google Cloud Code Assist API to enable multi-model inference, seamless OAuth authentication, and real-time usage quota tracking.
YinChao AI Music is an agent skill that generates playable songs and BGM from prompts, lyrics, or reference audio, supporting composition, continuation, and lyric-only creation.
A DSH plugin for Chinese retail investors that orchestrates multiple subagents for financial data collection, cleaning, analysis, and self-contained chart generation.
Real-time translate the thinking chain, task cards and answers in DeepSeek Harness into 8 languages — display-only, zero context cost, with local Ollama and multi-provider support.
A safe mathematical expression evaluator with arithmetic operations and common math functions, zero dependencies and zero processes, designed for DeepSeek Agent.
snapgrep is an in-process trigram index plugin that accelerates code search, typically 40–70× faster than ripgrep, with no sidecar process.
Integrate ChatGPT/Codex subscriptions into DeepSeek Harness as model inference providers via OAuth device code login.