dsh-soundscape: Session Sound Effects Plugin for DeepSeek Harness Web UI
dsh-soundscape is a community plugin (MIT License) for the DeepSeek Harness Web UI maintained by Blaczz, designed to enhance session state feedback through synthesized sound effects via WebAudio, avoiding the limitations of relying solely on text prompts. The plugin plays specific synthesized sound effects at the start of turns, upon successful completion (configurable confetti), during error interruptions, and while waiting for user approval, and supports optional session greeting sounds and streaming output ambient sounds. Its core advantages lie in zero audio files, zero network requests, and zero core code modifications, utilizing pure oscillator synthesis technology. The plugin provides an independent settings namespace, supporting the saving of the main toggle, volume, and individual sound effect preferences, and retains configuration functionality across restarts. Furthermore, it exposes the `ctx.soundscape` service to other client plugins, allowing calls to `play`, `celebrate`...
Read Moredsh-deck-builder: Generate self-contained HTML presentations from Markdown
dsh-deck-builder is an independent community plugin for DeepSeek Harness (DSH) maintained by Blaczz and released under the MIT License. It is designed to address the need for directly displaying Markdown content (such as reports and meeting minutes) within agent tasks. By registering the `deck_build` tool, it converts Markdown into self-contained, single-file HTML presentations. The plugin features zero dependencies and requires no core modifications, with generated outputs containing inline CSS/JS and no network access. It supports three themes (`deepseek`, `light`, `dark`) and keyboard navigation (arrow key switching and fullscreen). Its splitting logic prioritizes separators such as `---`, falling back to H1/H2 headers if they are absent. Markdown rendering covers...
Read Moredsh-math-copy: Add accessible copy interaction for DeepSeek Harness Web formulas
`Blackspace2/dsh-math-copy` is an MIT-licensed plugin designed for the DeepSeek Harness (DSH) Web client, aiming to address the difficulty of directly copying LaTeX source text from rendered formulas. The plugin automatically identifies inline, display, and malformed KaTeX formulas within DSH, retrieving the original LaTeX by reading `application/x-tex` annotations rather than inferring from the visual appearance. Users can directly copy to `$$...$$` format via a left-click, or right-click to open a menu and select `$...$`, `$$...$$`, and `\[...\]` block formats. Keyboard operations are supported (Enter/Space to copy, Shift+F10 to invoke the menu). The plugin features hover...
Read MoreFabric: DSH Runtime Plugin Kernel and Downstream SDK
Fabric is the Runtime Plugin kernel and downstream SDK of DSH, designed to solve the challenges of business plugin loading, synchronization, and maintenance, realizing the philosophy of "everything is a plugin." It supports installing, starting, stopping, upgrading, and uninstalling plugins in the form of Runtime Packages according to Profiles, and synchronizes them to all browser tabs without requiring a DSH restart after installation. Core functions include Runtime Package management, author commands (such as create, build, test, hot-reload development), Runtime manifest configuration, collaboration interfaces, and credential management. Fabric provides a typed Config and Resource exchange mechanism to ensure data transmission through interfaces with identity identifiers, ensuring the Client cannot read plain credentials. In addition, it supports converting ordinary DSH plugins via `fabric mi...
Read Moredsh-dir-tree: DSH Web Session Floating Workspace Directory Tree
dsh-dir-tree is a floating directory tree plugin for the DeepSeek Harness (DSH) web client, designed to address the issue of errors associated with manual file path entry. The plugin displays the workspace directory of the current session in the bottom-right corner of the interface, supporting lazy loading of subdirectories, filename search and filtering, file type icon display, and directory highlighting. Users can click to select, double-click to copy the path, or directly drag files or folders into the chat interface to automatically fill in the full path. The directory tree automatically follows workspace or session changes, and its position is draggable. For installation, users must first prepare the DSH web environment, install the plugin via `dsh plugin add` from npm or a git repository (configuring build authorization is required for git installation), and finally start `dsh web` to use it. Developed by bentong
Read Moredsh-claude-marketplace: Load Claude Code assets as DSH Skills
`ben7am1n/dsh-claude-marketplace` is a Claude Code compatible layer plugin for DeepSeek Harness (DSH), designed to enable DSH to directly reuse skills, commands, and agents from the project's `.claude` directory or the Claude Code marketplace, eliminating the need to maintain separate invocation methods. The plugin loads components from various sources—including local paths, GitHub, and URLs—via tool entry points such as `cc_scan` and `cc_marketplace_add`, registering them as native DSH skills with a `cc-` prefix. It is installed using the `dsh plugin` command and supports configuration of a cache directory.
Read MoreHermes-to-DSH: Browse and inject local Hermes assets in the DeepSeek Harness Web sidebar
hermes-to-dsh is an MIT-licensed DSH plugin designed to integrate local Hermes skills, MCP services, and conversation history into the DeepSeek Harness Web sidebar, addressing the inconvenience of viewing and filtering local data. The plugin provides a draggable two-column panel supporting global search, conversation preview, source, and message type filtering. Its core mechanism includes two injection modes: active and passive. The active mode allows users to select specific skills and MCP services to inject their configuration into the current agent and persistently save it; the passive mode guides the agent to read on demand via prompts. The plugin relies on the local Hermes state database (read-only) and Python environment, requiring a restart of the web service to take effect after installation. It is suitable for those wishing to quickly browse local Hermes
Read Moredsh-usage-ledger: DeepSeek Harness Token Usage and Cost Statistics Plugin
dsh-usage-ledger is a DeepSeek Harness (DSH) plugin maintained by beijingwahw, designed to provide token usage statistics, cost calculation, and budget control features. By monitoring session events, the plugin aggregates and persists model call inputs, outputs, and cache consumption across three dimensions: session, Beijing time, and lifecycle. Its core features include dynamically fetching and hot-reloading official pricing, supporting DeepSeek peak-valley time-of-use pricing, and covering price catalogs of multiple vendors such as DeepSeek, Zhipu, and Kimi, while allowing users to perform highest-priority price overrides via customPrices. In terms of cost control, the plugin supports daily, total, and per-session budget limits, issues alerts when thresholds are reached, and intercepts model calls when exceeding the budget through hooks. Additionally, it is built-in...
Read More@beihaizb/dsh-notebook: Agent-controllable stateful Jupyter workspace for DSH
`beihzb/dsh-notebook` is an MIT-licensed plugin designed for DeepSeek Harness (DSH) to enable agents to operate on stateful Jupyter workspaces. The plugin provides a persistent `ipykernel` runtime, enabling agents to read, edit, and execute cells using tools such as `nb_get`, `nb_edit_cell`, and `nb_run_cell`, while also allowing the viewing of runtime variables and structured error information. Core features include version-snapshot-based safe editing and rollback mechanisms, long output protection (automatic summarization and archiving of full logs), and a VS Code-aligned UI supporting queues, interruptions, and restarts. The plugin ensures that each DSH session has an independent kernel and directory to prevent state leakage, and includes automatic process cleanup.
Read Moredsh-stt-input: Add speech input for DSH Web GUI
`dsh-stt-input` is a voice input plugin for the DeepSeek Harness (DSH) Web GUI, designed to address the need for quickly inputting long prompts via voice within DSH. The plugin allows users to record by clicking the microphone button and fills the recognized text directly into the input box upon stopping. It provides two recognition engines: one is browser-based local recognition using the Web Speech API, supporting zero-configuration usage in Chrome/Edge with real-time intermediate results; the other is API recognition based on MediaRecorder and OpenAI-compatible interfaces, supporting OpenAI, Groq, or custom services and multiple Whisper models. The plugin supports configuring service presets, API keys, language, and write mode (append or replace). API
Read Moredsh-code-impact: Read-only TypeScript/JavaScript Change Impact Analysis for DeepSeek Harness
dsh-code-impact is a standalone read-only plugin released by the DeepSeek Harness community, designed to provide impact analysis for TypeScript/JavaScript Git changes. The plugin is maintained by baidd1011, licensed under the MIT License, and is currently at version 0.2.0. By temporarily building a reverse module dependency graph, it outputs changed declarations, affected files, propagation paths, evidence locations, and sorted test candidates, though it does not replace test execution or maintain a long-term index. The plugin supports three comparison modes: workspace, staged, and refs. It utilizes the TypeScript Compiler API to parse various file types and project configurations (such as `tsconfig.json`, path aliases, etc.), and via Sche...
Read Moredsh-llm-volcengine: Integrating DSH with Volcano Ark Agent Plan and Coding Plan
dsh-llm-volcengine is an MIT-licensed DSH plugin maintained by Badakonpro, designed to simplify the integration of Volcengine Ark Agent Plan and Coding Plan. By registering two LLM provider routes, `volcengine-plan` and `volcengine-coding`, this plugin unifies the handling of protocol routing, compatibility toggles, credential parsing, and thinking intensity settings. It adopts a self-contained architecture and does not rely on any specific settings schema. Core features include: Agent Plan routing using the `openai-responses` protocol (with `openai-completions` for certain models), and...
Read Moredsh-ocgo-usage: Display OpenCode Go usage in the DeepSeek Harness Web sidebar
dsh-ocgo-usage is an MIT-licensed plugin for DSH Web designed to address the issue of switching to an external interface to view OpenCode Go usage. It displays a persistent badge in the bottom-left corner of the sidebar, intuitively presenting the remaining usage percentage for the current month along with status colors (orange indicates low remaining quota, and red indicates exhaustion). Hovering over the badge expands a detail card showing usage progress for the rolling period, this week, and the current month, as well as the reset time. Core features include a peak-hours timeline based on Beijing Time, highlighting the current interval and displaying the time until the next switch point. The plugin automatically reads the API key from local configuration files or environment variables in sequence, requiring no additional configuration, and refreshes data every 60 seconds. The Host side uses Node ESM to register a route for proxying the usage interface, while the Client side registers the UI location using a hand-written Bundle, enabling cross-platform integration.
Read Moredsh-mermaid-image-preview: Preview Mermaid diagrams in DSH Web Chat
dsh-mermaid-image-preview is a community plugin for DeepSeek Harness maintained by baconbao, designed to address the pain point of viewing Mermaid charts in DSH Web sessions, which typically requires copying code to external tools. When the plugin detects a Mermaid code block within a chat message, it renders it directly as an image preview. It supports 14 chart types, automatic light/dark mode switching, and allows users to click on charts to zoom in or out. Its core features include a "local rendering first" strategy, where external or self-hosted services can be configured as a fallback rendering engine if local rendering fails or is disabled. Users can adjust runtime parameters via UI setting cards without restarting the service; however, restarting dsh web is required if underlying configurations are modified. Installation involves adding the plugin via the npx command and restarting the service.
Read MoreAzzygoatcoder/agent-useful-skills: Modular AI Research/Engineering Skills Collection
Azzygoatcoder/agent-useful-skills is a modular collection of AI research and engineering skills licensed under the MIT License, designed to address the pain points agent developers face in tasks such as paper reading, plotting, document conversion, and security auditing, where cross-project reuse is difficult. Compatible with Claude Code and DeepSeek Harness (DSH), this tool transforms repetitive workflows into registerable, verifiable, and deployable skill modules. Core features include code security auditing, paper workflows, Office document processing (e.g., Markdown to Word/PPT), publication-quality data plotting, read-only disk analysis, and Git collaborative publishing. Users can install it via DSH plugin commands or links and configure API keys via environment variables to call image recognition features,
Read Moredsh-download-progress: Provides a persistent download progress panel for DeepSeek Harness
`dsh-download-progress` is a download management plugin for DeepSeek Harness (DSH), designed to address the difficulty in distinguishing progress during multi-task downloads. The plugin provides a resident floating panel in the bottom right corner, displaying the filename, percentage, speed, and remaining time in real-time. It supports multi-task concurrency, history retention (up to 20 entries), and file locating, and features panel dragging and resizing. The download functionality supports breakpoint resumption, HTTP redirects, GitHub mirror acceleration, and aria2 multi-threaded downloads (16 connections), with task status written to a JSON file for the panel to read. Installation is divided into Dynamic (temporary) and Fixed (auto-load on restart) modes; the Fixed mode is recommended for long-term use, requiring the copy of the plugin package to a specified directory and the configuration of `cordis.patch.yml`. Typical use...
Read More@az7627/dsh-token-usage: DeepSeek Harness plugin for tracking token usage in the web chat view
`@az7627/dsh-token-usage` is an MIT-licensed plugin for DeepSeek Harness (DSH) designed to provide fine-grained token usage statistics via a Web conversation view. It is available as a standalone package, with its core functionality being to provide a "Usage Statistics" entry in the session header that opens a right-side drawer displaying an ECharts-based stacked bar chart by time slot. It supports statistics for the current, multi-selected, or all historical conversations. Time granularity options include 15 minutes, 1 hour, or 1 day, with time ranges covering the last 7 days, 30 days, or all history. The data presentation precisely breaks down usage into three categories: cache input, input (including cache writes), and output, supporting totals down to the unit and per-conversation details. Data is sourced from session log replay, and historical data is also viewable. Users can...
Read Moredsh-conversation-cost: Display DeepSeek usage cost at the bottom of DSH conversation
dsh-conversation-cost is an MIT-licensed plugin for DSH Web designed to address the complexity of DeepSeek conversation cost calculation. It displays real-time RMB and USD dual-currency costs at the bottom of the conversation statistics bar, eliminating the need for manual price splitting against a reference table. Core features include precise request-level billing, capable of handling mixed models, cache hits (distinguishing between hit and miss inputs), and peak/off-peak pricing (standard, peak, and idle). The plugin supports installation from GitHub or a local directory, with upgrades requiring an overwrite installation and a restart. When the cost ledger is not ready, it displays an approximate value marked with "≈"; unknown models are not billed and will trigger a prompt. This tool is suitable for developers needing to quickly verify complex session costs, but please note that official prices may change and this plugin is not an official app store product.
Read Moredsh-lark-web-auth: Lark Login and Multi-user Session Isolation for DeepSeek Harness Web GUI
`dsh-lark-web-auth` is a community plugin for DeepSeek Harness, designed to address authentication and data isolation issues during multi-user access to the Web GUI. This plugin implements login via Lark OAuth and utilizes `open_id` to ensure that each user can only view, access, and modify their own `session`. Requests from non-owners uniformly return 404, thereby protecting privacy. Its core features include: HTTP-layer isolation achieved through middleware Cookie verification and permission filtering at the RPC and list layers; the use of SQLite to persist login status, ensuring no disconnection upon process restart; zero frontend source code modification with UI injection via API; and support for automatically reassigning ownership of existing `session`s. Installation requires configuring Lark application credentials and configuration.
Read Moredsh-cursor-theme: Customize DeepSeek Harness cursor by status
`dsh-cursor-theme` is a DSH plugin for DeepSeek Harness, designed to independently customize cursors based on UI mouse states. It supports 14 states, including default, pointer, and text, each of which can be individually configured with images, hotspots, and sizes. It supports real-time preview and persistence to the DSH profile, ensuring settings remain effective after a restart. The plugin provides 18 preset themes covering all states and a library of 25 built-in SVG graphics. It supports uploading PNG/CUR images or importing and exporting ZIP image packages (including manifest.json). It features a hotspot editor and size adjustment options (16/24/32/48). On Windows, it can be used via registry + S...
Read Moredshp: Transform DSH profiles into listable, copyable, and shareable files.
dshp is a command-line tool and plugin for managing DeepSeek Harness (DSH) profiles, designed to address the limitations of DSH's native profile management. It supports listing local profiles and their resource usage, viewing detailed bundle loading order, plugins, and patches, and performing profile creation, cloning (preserving node_modules), export to single file, import and rebuild, diff comparison, and deletion. Built on Node 18+, this tool is compatible with dsh 0.1.0-rc.5 and is licensed under the MIT License. Users can utilize it as a CLI via `npx` or install it as a plugin. Typical workflows include using `ls` to check the status and `clone` to create an instance.
Read Moredsh-plan-first-dev: DSH Plugin that Automatically Enters Plan Mode Before Development
`dsh-plan-first-dev` is an MIT-licensed plugin for DeepSeek Harness (DSH) designed to establish "plan first, implement later" as the default behavior for code development tasks. Installed at the profile layer, the plugin listens for session creation events; upon detecting a new session, it automatically enables plan mode while retaining the user's permission to manually disable it. For development tasks, the plugin enforces a workflow where the model must submit a complete plan—encompassing architecture, changed files, API changes, and testing standards—before implementation is permitted, requiring user approval via `exit_plan_mode`. For non-development tasks such as writing and data analysis, the plugin registers a `no_plan_required` tool, allowing the model to identify the task type and exit plan mode directly for execution without requiring plan approval.
Read Moredsh-eval-regression: Deterministic Output Regression Plugin for DeepSeek Harness
`aryswisnu/dsh-eval-regression` is a plugin for DeepSeek Harness (DSH) that provides a deterministic, CI-safe golden output evaluation tool. It registers the `evaluate_golden_output` tool to specifically check whether the candidate output contains required fragments or forbidden fragments. This plugin does not evaluate semantic correctness or model quality; it only generates repeatable pass/fail evidence. Its core functions include fragment-level checks, a deterministic report outputting a total score and case-by-case missing or forbidden fragments, and a CLI tool that supports JSON suite input, prints reports via standard output, and returns different exit codes based on results, facilitating...
Read MoreArtificialwhale/session-reader-plugin: Read content of other sessions within a DSH session
Artificialwhale/session-reader-plugin is a read-only plugin for DSH that allows reading the context of other sessions within the current session, including user/assistant messages, tool call results, reasoning processes, and raw event streams. Independent of the official implementation, it provides core functions `session_list` and `session_read`, supporting direct querying via the `/sessions` slash command without requiring model turns. Reading modes include `io` (default message interaction), `thinking` (reasoning chunks), `surface` (compressed view), and `raw` (raw logs). All outputs are equipped with truncation protection to prevent context pollution. This plugin is implemented purely in JavaScript with no native dependencies and is based on DSH 0.1.0-rc.6.
Read Moredsh-turn-approval: Add task-level authorization for DSH Web approval cards
dsh-turn-approval is a community plugin for DeepSeek Harness (DSH) designed to address the operational costs associated with repeated approvals for high-privilege operations under default permissions. This plugin introduces an "Allow this task" button to the DSH Web approval card, enabling users to temporarily permit specific types of `danger-full-access` escalation within a single `turn`, with the authorization automatically expiring upon `turn/end`. Its core advantage lies in the balance between security and convenience: it does not alter the default permission model or enable global `full access`. Authorization is stored only in memory and bound to specific `session` and `turn` contexts; child agents do not inherit this authorization, and it only automatically allows explicitly matched `danger-full-access`.
Read More