DeepSeek Harness Plugins
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
dsh-token-usage is a DSH plugin that provides a token usage statistics panel, mimicking Codex style, to display total consumption, per-session peaks, activity heatmaps, and top plugins for clear insight into token usage.
Annotates Markdown items in preview and generates structured modification instructions for precise content editing.
A DeepSeek Harness plugin offering semi-automatic and fully automatic approval modes on the workspace-write sandbox through layered decisions with deterministic rules and LLM review.
dsh-notion connects DeepSeek Harness to Notion via OAuth 2.0, enabling agents to search, read, and write Notion pages, databases, and comments.
A DeepSeek Harness plugin for managing conversations with archive, delete, move to workspace, and per-conversation Agent preset migration features.
Adds a file browser to the DeepSeek Harness Web interface, enabling viewing, editing, and saving project files with a VSCode-style editor.
DSH Git Memory is a Git-backed memory plugin for DeepSeek Harness that provides reviewable, revertible long-term context, currently in pre-alpha status.
A green interrupt button that silently halts the agent, which then summarizes its progress and asks for new requirements, with customizable interrupt prompts.
A plugin that adds free image understanding capabilities to DeepSeek Harness, using OpenRouter vision models to automatically process images and enable pure-text models to understand visual content.
dsh-run-guard is a DeepSeek Harness plugin that prevents LLM reasoning loops and automatically continues unfinished tasks using guard and continue mechanisms to stabilize Agent runtime rhythm.
dsh-recap is a session recap summary plugin that automatically generates summaries for each model request and displays them chronologically to help users quickly review conversation history.
A DSH plugin for AI role-playing and multi-role discussions using @ mentions, where each role responds independently and a moderator converges consensus, disagreements, and action suggestions.