dsh-teacher: DSH plugin, turning agents into Socratic teachers
dsh-teacher is a plugin-based teaching tool within the DSH ecosystem, designed to transform traditional "direct instruction" into "Socratic questioning guidance." The plugin loads courses from Markdown question sets and persists data in SQLite. Its core strategy is "ask without answering," guiding user thinking by hiding answer keys while recording "knowledge gaps" in real-time during sessions. It integrates the FSRS-5 spaced repetition algorithm to perform on-demand retesting of overdue knowledge gaps, achieving cross-session long-term memory consolidation. Furthermore, it provides immediate quiz popups (supporting multiple choice and fill-in-the-blank) without requiring LLM involvement, followed by an LLM-based scoring and Socratic review of error points after the quiz. The Web client supports viewing the question bank, the knowledge gap panel, and status. Installation requires DSH rc.6+ and Node >= 22.
Read Moredsh-cost-tracker: Add configurable Token cost tracking for DeepSeek Harness
`yflmq001/dsh-cost-tracker` is a developer preview plugin for DeepSeek Harness (DSH) designed to solve the complexity of session cost calculation. It prices completed LLM calls based on model-configured price tables and exposes session-level costs to the Web UI or the `/cost` command. Its core features include support for configuring prices for cache-hit/miss inputs and outputs, optional peak/off-peak rate periods, and the ability to distinguish between total, peak, off-peak, and per-model cost readings. For models without configured prices, the plugin marks them as "unconfigured" rather than silently guessing, ensuring billing transparency.
Read Moredsh-plugin-audiolib: Play AudioLib.ai ambient music based on DeepSeek Harness session state
dsh-plugin-audiolib is a plugin for DeepSeek Harness (DSH) designed to automatically control ambient background sounds based on the agent's session state. Its audio source comes from AudioLib.ai; a core feature is that it does not abruptly switch tracks during state changes, stopping playback only when explicitly calling `music_stop()`. The plugin provides three tools, `music_play`, `music_stop`, and `music_status`, supporting streaming via `mpv` or `ffplay`, with a fallback to macOS' `afplay` (requiring pre-downloaded tracks) if these players are unavailable. API keys are securely stored in the DSH credential store and feature a network failure backoff retry mechanism. Installation requires pnpm.
Read MoreDeepSeek Harness MCP Service and Skills Manager @xxxyz/dsh-mcp-manager
@xxxyz/dsh-mcp-manager is an MIT-licensed plugin for DeepSeek Harness (DSH) designed to centrally manage MCP servers and Skills. It displays the MCP server list, connection status, Loader stage, and number of tools via a web settings page, supporting the addition, deletion, enabling, disabling, and restarting of servers, as well as JSON configuration import and export. Configuration is persisted to the project or global `cordis.patch.yml`, ensuring persistence after restart. The plugin also provides a Skills management interface, allowing users to browse, search by source, and toggle Skill status with a single click. Additionally, it exposes an HTTP API (`POST /dsh-mcp-manager/api`) for script automation and registers 4...
Read Moredsh-craft-your-textbook: A DSH plugin to transform textbook PDFs into AI teaching blueprints.
`dsh-craft-your-textbook` is an MIT-licensed plugin package for DeepSeek Harness (dsh) designed to transform textbook PDFs into "AI teaching blueprints." This tool provides an automated pipeline comprising six stages, fully supporting Chinese and requiring no deep technical background. The core workflow includes uploading and parsing multi-type teaching materials (such as student books, syllabi, etc.) to generate knowledge maps and chapter structures; establishing "approval" checkpoints at critical decision points to ensure human control; generating a best-practice chapter for line-by-line review and comparison; writing the remaining chapters through multi-role AI collaboration, with support for chapter-by-chapter preview and modification; and finally, performing a full chapter review, targeted modifications, and rollback operations. Installation requires ensuring the dsh environment is ready; simply run `npx dsh-craft-your-textbook`.
Read Moredsh-mediacrawler: DeepSeek Harness and MediaCrawler Controlled Crawling Adapter
`dsh-mediacrawler` is an MIT-licensed DSH profile bundle maintained by `xwh-01`. Acting as a bounded stdio MCP adapter, it integrates independently installed MediaCrawler with DeepSeek Harness (DSH). It does not modify MediaCrawler's source code but provides an adapter layer to handle tasks that search providers cannot, such as DSH agents needing to access logged-in platform records, creator feeds, and explicitly enable comments. The tool supports searching, details, and comment scraping for Xiaohongshu, Douyin, Kuaishou, Bilibili, Weibo, Tieba, and Zhihu, exposing 12 MCP tools for starting, monitoring, persisting, and exporting data with redacted credentials.
Read Moredsh-grok-geo: Execute grok-geo brand GEO audit in DSH
`xuboboo/dsh-grok-geo` is an MIT-licensed DSH bundle plugin designed to provide AI search and GEO (Generative Engine Optimization) audit capabilities for a single brand. The plugin supports 17+ mainstream AI search engines (including 8 international and 9 Chinese engines), covering brand visibility, citations/recommendations, competitor comparison, factual accuracy, and content gap analysis. Its workflow includes input collection, validation, search execution, metrics calculation, and report rendering, and it can automatically switch to an offline import mode when `web_search` is unavailable. The plugin itself does not require Python, but script execution relies on the Python 3.11+ standard library. Users can quickly install it via the `dsh plugin` command, and it takes effect after restarting DSH.
Read Moredsh-vision-bridge: Add a description bridging layer for DeepSeek V4 / V4-Flash image messages
dsh-vision-bridge is a third-party image processing plugin for DSH, designed to address the limitation where text-only models, such as DeepSeek V4/V4-Flash, cannot natively process image context. By invoking an OpenAI-compatible Vision Language Model (VLM) to automatically convert user input images into text descriptions before passing them to DeepSeek, the plugin achieves image understanding without altering the original text processing workflow. Core features include: automatically recognizing and converting image content; dynamically declaring model support for image input; intercepting and replacing image blocks before requests; allowing models that natively support images to pass through directly; caching descriptions based on attachmentId to prevent double billing for repeated sends; and graceful degradation without interrupting the conversation when recognition fails. Installation requires the dsh CLI and pnpm as prerequisites.
Read Moredsh-MusicPlayer: Control music while chatting in DSH Web
`dsh-MusicPlayer` is a plugin for DSH Web, maintained by xiekai886 under the MIT license. It features a dark frosted glass floating player in the bottom right corner, supporting drag-and-drop position memory, expand/collapse animations, and playback controls such as play/pause, track switching, progress bar, volume, loop/random modes, and a playlist. It utilizes custom frontend module loading and inline SVG icons without relying on CDNs. Core capabilities include importing NetEase Cloud Music playlists and searching for individual tracks, proxying search results, playlists, and audio streams on the host side to bypass CORS and hotlink protection, and supporting Meting same-origin parsing. Default playlists can be configured via the environment variable `DSH_MUSIC_PLAYLIST` and are automatically loaded upon startup. It provides a `music` tool that enables the agent to control playback, import playlists, search for songs, switch tracks, and more using natural language.
Read Morexiaoxianyu-office/dsh-image-tools: Add image recognition capability to DeepSeek Harness text sessions
`xiaoxianyu-office/dsh-image-tools` is a DSH plugin designed to equip text-only models like `deepseek-v4-pro` with image recognition capabilities. The plugin works by automatically saving chat images to the workspace, dynamically disabling the native `read_image` tool, and guiding the model to use the `image_recognize` tool to delegate visual sub-agents (such as `xiaomi/mimo-v2.5`) for image recognition. It supports various preset sessions like `standard` and `code`, dynamically determines interception or bridging strategies based on current model capabilities, and includes a first-time setup wizard and fault self-diagnosis functions. Image recognition output follows strict specifications, requiring pixel coordinates and `#RRGGBB` color values, and addresses items not appearing in the image or...
Read Moredsh-vision: A plugin to add visual capabilities to DSH
`dsh-vision` is a native DSH plugin maintained by `xiaoshihou514`, aiming to supplement visual capabilities for DeepSeek Harness. The plugin features a minimalist design, with core functions including adding visual support, utilizing free Zhipu interfaces, and employing a local Qwen-VL 2B visual small model. Installation requires deploying the DSH core first (requiring Node engine ^22.19.0 || >=24 and specific peer dependencies), and the plugin is released under the MIT license. Please note that the installation command provided in the README references the incorrect plugin address `dsh-weixin#main`; the command should not be executed directly to install `dsh-vision` until the maintainer corrects it. The plugin is suitable...
Read Moredsh-keepalive: Add an optional watchdog for the DSH Web process
dsh-keepalive is an MIT-licensed plugin maintained by xiaohj233, designed specifically for the DeepSeek Harness (DSH) Web process. Its core functionality is to run a detached watchdog to monitor the status of the DSH Web process and automatically restart it with the same parameters after the process exits. The plugin supports manual restarts via API or the UI, and provides an optional repair agent that performs repairs, validation, and rollback based on snapshots within a controlled workspace when repeated startup failures occur. Usage requires specific Node.js and pnpm versions and can be installed via a command. After installation, restart the Web profile and explicitly enable `enable` in the configuration.
Read Moredsh-plugin-uisfx: Add semantic UI sound effects for DeepSeek Harness
dsh-plugin-uisfx is a plugin designed for the DeepSeek Harness (DSH) Web environment, maintained by XanthanL and released under the MIT License. By embedding the uisfx 0.4.0 Web Audio synthesis runtime, the plugin provides semantic sound feedback for task statuses and button actions without requiring built-in audio files. Its core features include 12 global sound effect packages, 11 scene mappings, real-time preview, automatic sound feedback for task statuses (e.g., success, failure, pending) and button types, as well as persisting settings to the Host configuration. Additionally, it exposes the `ctx.uisfx` API for other DSH plugins to call for playing sounds, switching styles, or adjusting volume. Installation requires DSH 0.1.0-rc.6.
Read Moredsh-writing-remote: DSH Writing Panel Host-side Data Channel
dsh-writing-remote is a host-side data channel plugin in the DSH writing ecosystem, designed to connect the `dsh-tool-writing` engine running on the host side with the `dsh-client-ui-writing` panel running on the client side. It adapts engine capabilities to Remote methods via the Typert protocol, ensuring that only read-only data and orchestration trigger interfaces are exposed to the client, while keeping generation capabilities and full write operations on the host side to avoid security risks. The plugin's core functions include data reading and orchestration triggering. Data reading covers project index, status, chapter list, body preview, setting browsing, project and library search, evolution version chain, and multi-line narrative thread graphs, etc.; orchestration triggering provides chapter-end automatic orchestration (such as summary, extraction, etc.) and non-destructive...
Read Moredsh-client-ui-writing: DSH Client Writing Panel
`dsh-client-ui-writing` is a Web client plugin within the DeepSeek Harness ecosystem, version 0.6.2, licensed under the MIT License. Designed specifically for writing preset sessions, it provides a sidebar panel and a full-book index view within the session. Its core responsibilities are reading data, displaying information, and triggering orchestration, without directly generating the main text. The plugin comprises four main functional modules: the Project tab for viewing the table of contents, statistics, and initializing projects; the Library tab for displaying imported novel information; the Search tab supporting combined full-text search across projects and libraries; and the Evolution tab for presenting setting and character evolution version chains and graphs. Furthermore, users can trigger actions such as `novel_sync` for chapter-end synchronization, summary generation, and plot twist reminders. Within the session, it supports jumping to chapters via numbers and quickly focusing the search box using shortcuts.
Read Moredsh-equip-engine: Equip DSH plugins based on tasks
dsh-equip-engine is a task-driven DSH plugin configuration engine designed to address issues such as mutual exclusivity, cost, and compatibility within plugin combinations. It decomposes task capabilities using the "Five-slot" (Perception, Decision, Action, Memory, Output) model and introduces a combination scoring mechanism that incorporates synergy bonuses, conflict penalties, cost control, and trustworthiness. This mechanism evaluates the overall compatibility of plugin combinations rather than the quality of individual plugins. The engine employs a dual retrieval strategy that combines exact matching from a rule-based curated library with LLM semantic understanding. It rapidly filters down to fewer than 48 candidates from over 2000 plugins, balancing efficiency and accuracy. In terms of data, it includes 19 manually curated entries and 2033 LLM-generated manifests, and possesses the capability to automatically sync with the ecosystem on a weekly basis. Users can perform rule-based retrieval via the `/equip` command, or use...
Read MoreKnowLP-RAG: A Plugin for DSH Providing Dual Knowledge Graph Retrieval for Markdown Notes
KnowLP-RAG is a plugin for DeepSeek Harness (dSH) designed to organize Markdown notes into a retrievable knowledge structure. It addresses the pain point of developers struggling to determine reading order and relevance among numerous notes by providing retrieval results with paths that clearly recommend reading sequences and similar alternatives. Built upon dual-graph retrieval, this plugin combines vector retrieval, full-text retrieval, and feedback weighting mechanisms. Its core capabilities include: adopting a four-engine fan-out retrieval approach; supporting paragraph-level matching and Chinese note libraries (including time-anchored queries); featuring a decay-based forgetting mechanism (strengthening when used and weakening when not); and operating in n-gram mode without a GPU. The main tool interfaces include `knowlp_search` and `knowlp_get_note`.
Read More@winterchenhuan/dsh-skills-viewer: DeepSeek Harness Web Read-only Skills Settings Page
`@winterchenhuan/dsh-skills-viewer` is an MIT-licensed pure UI plugin developed for DeepSeek Harness Web, aiming to address the lack of a read-only skill list viewing feature in the Web settings panel. The plugin retrieves the catalog of skills available for the current normal session via the `skill.list` wire RPC and adds a new page named `Skills` to the Settings panel. The page presents skill names, `model: off` badges, `description`, and `whenToUse` fields in an alphabetically sorted read-only list. It supports keyword-based search and filtering but does not provide a preview of the skill content or a manual refresh function. When the session changes or points to a subagent...
Read Moredsh-turn-fold: Collapsing DSH's Thinking and Tool Calls into Steps and Turn Bars
dsh-turn-fold is a standalone plugin developed specifically for the DSH (DeepSeek Harness) Web interface, maintained by Winter-And-You-Gone and licensed under the MIT License. It does not modify the DSH core source code. Its core functionality is to collapse long Think, tool calls, and command outputs within agent replies into interactive step bars and turn bars, addressing the issues of interface clutter caused by long replies and the difficulty in locating summaries. Key features include: 1. **Automatic Step Folding**: It groups and collapses tool calls and thinking processes between text segments. It displays actions and shimmer effects in real-time during execution. Upon completion, it groups titles by type (e.g., execution, reading, searching), and highlights failed items in red. 2. **Real-time Metric Monitoring**: It displays execution time and Token consumption in real-time.
Read Moredsh-prompt-stash: Local input stash plugin for DSH Web
dsh-prompt-stash is a local input stash plugin for DeepSeek Harness Web, maintained by Wine-Red and licensed under the MIT License. It addresses the pain point where users need to handle other issues temporarily while writing unsent plain text, without losing their current content. Core features include a session-isolated LIFO stash stack (up to 10 entries), a collapsible preview bar above the input box, and the secure clearing and restoration of drafts via the official API. The plugin supports Chinese and English, light and dark themes, and keyboard shortcuts. Data is stored locally in browser localStorage and does not send network requests. It supports custom hotkeys (default: Ctrl+S), which can be configured in settings. Installation requires DSH versions 0.1.0-rc.6 to rc.8 and a Web profile, via command.
Read Moredsh-browser-fs: Enable the dsh agent to read and write local files on the browser's machine
dsh-browser-fs is a dsh client plugin designed to address the issue where remotely deployed dsh agents cannot directly access the user's local browser files. This plugin utilizes the File System Access API, allowing users to authorize local directories and forward tool calls to the browser for execution via a WebSocket channel. Core features include registering three model tools, `browser_fs_list`, `browser_fs_read`, and `browser_fs_write`, which support directory listing, UTF-8 text reading (default limit of 256 KiB), and writing. The browser interface provides a floating card to display connection and authorization status, supporting directory tree browsing, file search, and image/text preview (text preview limit 64K).
Read Moredsh-plugin-doctor: Diagnostic and Decision Tool for the DSH Plugin Ecosystem
`dsh-plugin-doctor` is a diagnostic and decision-making tool for the DSH plugin ecosystem, maintained by white-sand-grand, designed to address plugin discovery, duplicate detection, conflict checking, and version synchronization issues. The tool requires Node.js >=22.19, operates in read-only mode by default, and does not automatically execute installation, uninstallation, or the launching of the Web UI. Its core capabilities include: searching for community plugins and identifying functional duplicates; generating similarity scores and Mermaid relationship overview diagrams; checking package names, dependencies, and potential conflicts prior to bulk installation, employing a fail-close strategy if metadata is missing or APIs are restricted; identifying sub-plugins already provided within aggregated bundles to avoid redundancy; auditing actual plugin usage based on locally persisted logs; and comparing differences between the local DSH and official versions to evaluate the impact of upgrades on installed plugins.
Read Moredsh-law: Local Legal Workbench Plugin for DeepSeek Harness
`dsh-law` is a local legal workbench plugin for DeepSeek Harness designed to consolidate to-do reminders, monthly calendar scheduling, case overviews, and DSH native legal mode sessions into a single workspace. It addresses the high costs associated with verifying information scattered across different legal workflows. Licensed under the MIT License, the plugin requires Node.js >= 18, has zero npm dependencies, and stores all data locally. Its core features include an on-screen case management homepage (supporting marking of approaching due dates and overdue cases), a six-column drag-and-drop case Kanban board with SSE real-time synchronization, DSH native session binding, a skill command panel, and four sets of legal themes. The system supports consultation intake, automatic generation of eight-level local working directories, AI material onboarding, and access control management (AI application and lawyer approval), while also providing a remote entry point for mobile access. Users can install it via GitHub or npm commands and restart the DSH web service.
Read Moredsh-zotero-wiki: Convert Zotero library to searchable Obsidian Wiki
dsh-zotero-wiki is an MIT-licensed DeepSeekHarness (DSH) plugin designed to integrate Zotero literature management, PDF parsing, full-text note generation, and Agent retrieval into a unified workflow. It automatically syncs Zotero records from local or web sources, utilizes MinerU to parse PDF original text, and invokes DeepSeek to generate structured Chinese notes, ultimately compiling them into a searchable Obsidian Wiki. Core features include dual-source synchronization, interactive range selection, parsing quota control (default 50 items per run), Wiki index compilation, LINT self-checking, and Agent queries based on in-memory full-text retrieval to ensure answers include traceable literature citations. The plugin supports CLI and DSH native tools.
Read MoreWayneJin0918/dsh-wm: World Model Check, Comparison, and RSI Plugin for DeepSeek Harness
`WayneJin0918/dsh-wm` is a pure JavaScript world model debugging toolkit for DeepSeek Harness (DSH), designed to simplify rollout frame inspection, pred/gt comparison, and research iteration workflows. This plugin supports Node.js 18+, allowing quick start via `fixtures/sunset` without requiring a GPU; video input relies on ffmpeg. Core functions include `wm_inspect` for sampling keyframes to generate contact sheets and analyzing brightness and color; `wm_view` for generating self-contained HTML comparison pages containing SSIM timelines, difference heatmaps, etc.; and `wm_rollout_diff` for calculating SSIM/MSE metrics and...
Read More