DeepSeek Harness Plugins · Model Inference
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Adds voice input, tool-call collapsing, and custom CSS to the DeepSeek Harness Web interface as optional enhancements.
Local-first plugin that turns growth data into AARRR funnel diagnostics, HADI experiments, and execution SOPs.
A system-wide voice input plugin that transcribes speech to text at any cursor position using a local Whisper model, no network required.
A zero-dependency local repository analyzer that detects tech stacks, parses multi-language dependencies, and maps module references, outputting structured JSON for agent-driven architecture understanding.
Group multiple project folders and work across them in a single conversation—read and write files, list directories, and run commands without switching workspaces.
Gives text-only models three multimodal skills via Qwen APIs: image understanding, speech-to-text, and text-to-image, with a built-in generate-then-verify loop.
A code health check plugin for DSH that monitors syntax, type, and lint issues, displays a diagnostic card beside the chat input, and shares findings with the AI assistant.
Vision bridge for text-only DeepSeek routes: auto-describes image attachments via a vision model and feeds them back as text to prevent UNSUPPORTED_CONTENT failures.
Registers a workspace_digest tool that lists files by directory for model-inference retrieval with configurable entry limits.
Show provider name next to model name in the selector to distinguish same-named models across providers. Deprecated; functionality moved to dsh-chat-tweaks.
Model-facing tools to read, create, and edit PDF, DOCX, XLSX, PPTX, CSV, and Markdown files without shelling out to external utilities.
A port of superpowers-zh into DeepSeek Harness, offering 20 ready-to-use Chinese AI-coding skills auto-discovered by the agent for testing, code review, planning, and more.