DeepSeek Harness Plugins
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
dsh-system-proxy is an outbound network routing plugin for DeepSeek Harness that manages requests with multiple proxy protocols, intelligent failover, and health memory.
A DeepSeek Harness knowledge graph plugin that exposes a read_graph tool for querying codebase structure, exports, and dependency relationships.
dsh-tool-git provides a structured set of eight Git tools for DeepSeek Harness, running through a shell-free subprocess runner with a safety gate to block destructive operations by default.
The dsh-soul-md plugin is a persona card management tool that supports creating and switching character cards, along with AI long-term memory for persona evolution.
The dsh-survey plugin lets the model send a questionnaire at once, which the user answers in a single card, supporting various modes and question types for efficient interaction.
An installable DeepSeek Harness community plugin for editable, session-isolated multi-person memory supporting up to five individuals with separate information and relationships.
A model switch for DSH that points to any OpenAI-compatible endpoint with curated presets and automatic fallback on rate limits.
The dsh-archive-viewer plugin for DeepSeek Harness Web GUI enables viewing, restoring, and searching archived sessions, along with a one-click shutdown button for dsh.
A behavioral coaching plugin for ADHD users that lowers the cost of starting, provides structured support and instant feedback, breaking down complex tasks into immediately executable minimal actions.
dsh-cad-review is a plugin for DeepSeek Harness that performs evidence-first ASCII DXF inspection and deterministic CAD rule review, emitting structured issues tied to entity handles, layers, and geometric locations.
The dsh-same-mode-sandbox-noop plugin provides compatibility for DeepSeek Harness 0.1.0-rc.6 by automatically removing redundant sandbox permission fields in tool calls when the mode does not escalate, preventing incorrect rejections.
A structured long-term memory system for DeepSeek Harness that teaches agents to organize memories using units, records, index, and auto-injection.