CodeReader: View and edit workspace files in DSH Web
CodeReader is a plugin designed specifically for DeepSeek Harness (DSH) Web, maintained by PathChan and licensed under the MIT License. It aims to integrate a workspace file tree and a multi-tab editing canvas into the conversational interface, enabling users to browse, verify, edit, and write back files in the current workspace without switching windows. Key features include an expandable file tree, a multi-tab editor supporting syntax highlighting and line numbers, and the ability to save and validate content via keyboard shortcuts or the toolbar. The plugin supports agent collaboration by automatically switching to read-only mode when a model generates a file to prevent conflicts, and allows opening files with the system's default program. User preferences are stored locally in the browser and are not transmitted to the server. Installation requires configuring DSH Web first, followed by adding the plugin via the `npx` command from GitHub or local source code.
Read Moredsh-plugin-workbench: Transforming DSH web pages into an editable file workbench
dsh-plugin-workbench is a file workspace plugin developed specifically for DSH Web, designed to resolve workflow interruptions and provide lightweight code editor capabilities within the web interface. Maintained by Pasumao and released under the MIT License, it supports Node >= 18 and is currently at version 0.0.15. Core features include: a left-side file tree sidebar (supporting lazy loading and independent state persistence); multi-tab editable code previews (supporting syntax highlighting for JS/TS, Python, etc., line numbers, and transparent textarea overlay editing); right-click file operations (create, rename, delete, copy/cut/paste, and cross-workspace operations); inline image preview (supporting various formats with a 20MB limit); and Markdown rendering with the ability to @ mention or drag-and-drop workspace file paths into messages. The plugin supports per workspace...
Read Moredsh-worktree: Git worktree isolation and delivery for DSH subagents
`dsh-worktree` is a Git worktree plugin for DSH designed to resolve write conflicts when multiple sub-agents process the same repository in parallel. By providing independent worktrees for sub-agents, it supports a complete delivery workflow including review, verification, commit, merge, push, Pull Request creation, and archive recovery. The plugin is installed within a global DSH profile and does not modify business project configurations. Users can delegate tasks via `subagent_worktree` or manually manage worktrees using the `/worktree` command. The plugin supports various worktree origins and retention modes, includes built-in safety mechanisms to prevent the accidental deletion of modified worktrees, and does not use force push. It is suitable for scenarios requiring parallel development while maintaining result traceability. Users must review the source code and license before use, and uninstalling the plugin will not automatically delete files.
Read Moredsh-balance-stats: Provides balance and usage statistics for DeepSeek Harness Web
dsh-balance-stats is a community plugin for DeepSeek Harness Web, maintained by pangzi499, and is licensed under the MIT License. This plugin aims to address issues regarding scattered balance, session costs, and usage data by centrally displaying DeepSeek account balances (available, top-up, and bonus), real-time session cost estimation, historical spending (today, 7 days, 30 days), and Token usage broken down by model within the interface. The plugin supports two invoice import methods: automatic import, which requires saving the platform's userToken to a local credential file (permission 0600) to periodically retrieve invoices; and manual import of JSON responses. If invoices are not imported, it falls back to local estimation. The plugin features a fault-tolerant mechanism that retains the most recent successful data upon request failures and supports configurable interval refreshing.
Read Moredsh-mobile-control: DSH Plugin for driving iOS Simulator with WDA
`dsh-mobile-control` is a DeepSeek Harness (DSH) plugin designed to drive iOS simulators via WebDriverAgent (WDA) to achieve automated control of mobile devices. This plugin does not rely on macOS accessibility features or idb, but instead leverages WDA's HTTP service to provide features such as screenshots, UI element discovery, clicking, URL opening, accessibility tree export, and privacy pre-authorization. Upon installation, it registers eight `mobile_*` tools with the agent and supports natural language invocation. Before use, ensure the environment meets Node >= 20.11 and specific dependencies, and install it via command from GitHub. Core operations include starting it using the `wda-up.sh` script.
Read MoreDSH Marketplace: Community Marketplace Plugin for DeepSeek Harness
DSH Marketplace is a non-official community plugin maintained by ouyangyipeng. It aims to integrate the real-time community repository tagged with GitHub `topic:dsh-plugin` into DeepSeek Harness, supporting plugin discovery, search, verification, installation, updates, and uninstallation. A built-in offline version is available in DS-Harness Desktop v0.2.0; the standard Web profile requires Node.js v22.19.0+ and pnpm, installed via the `dsh plugin add` command. The plugin provides HTTP APIs such as Bootstrap, Catalog, Install, and Remove, with write interfaces restricted to same-origin loopback and requiring a process.
Read Moreorriduck/dsh-tui: A session-aware DeepSeek Harness terminal UI
orriduck/dsh-tui is a lightweight Terminal User Interface (TUI) released under the MIT License, designed for solo daily use scenarios. It aims to optimize the interaction experience of DeepSeek Harness (DSH) within local projects. The tool utilizes a "session-aware" mechanism, allowing users to initiate sessions within the current directory, ask follow-up questions, and monitor tool activities, approval statuses, context window usage, and permission badges in real-time. It handles credentials, model settings, sandboxing, and approvals via official DSH commands and does not store sensitive keys locally. Key features include streaming text and reasoning output, interactive approvals, skill discovery and invocation, and directory-based persistent session management (supporting listing, creating, and resuming sessions). Installation requires Node.js 22+, pnpm, and the DSH CLI, and can be installed globally via `npm install`.
Read Moredsh-read-image: Plug-and-play Image Reading for Text-only DeepSeek Harness
`dsh-read-image` is an MIT-licensed open-source plugin for DeepSeek Harness (DSH) designed to address the pain point where pure text models cannot directly read image pixels. This plugin enables text routing to accept image inputs, projects image chunks into `[Image #N]` text placeholders, and automatically registers the `read_image` tool. Through this tool, developers can invoke a configurable vision model to read specified images or file paths within a session, converting their content into text for model inference. If the routing itself supports native multimodality, the plugin intelligently falls back to the built-in tool to avoid conflicts. The plugin provides a "Read image" settings block, supporting the configuration of parameters such as the vision model and prompts. Installation requires DSH version 0.1.0-rc.6 or higher; simply add it via the command line and restart. The plugin features environment self-check and failure-safe mechanisms.
Read Moredsh-voice-funasr: DSH Web UI local offline speech input plugin
dsh-voice-funasr is a local offline speech input plugin for DSH Web UI, designed to provide a speech transcription solution that does not rely on cloud speech services. Its core mechanism is "local-first," prioritizing local audio transcription using the FunASR int8 ONNX engine (including Paraformer, FSMN-VAD, and ct-punc models). If the local engine is unavailable, it automatically falls back to the browser's Web Speech API. The plugin supports a "hold to record, release to send" operation and optionally invokes a user-configured LLM endpoint for two-step polishing to correct stutters or filler words; this is the only component that may send data over the network, while all other data is processed locally in memory. Installation requires installing funasr-onnx and other dependencies via pip, and downloading...
Read Moredsh-tool-schema: DSH JSON Schema Validation Tool Plugin
`dsh-tool-schema` is a DSH local JSON Schema validation plugin released under the MIT License, designed to address data validation, error localization, and Schema explanation issues within complex nested structures via a single tool call. The plugin provides four core actions—`validate`, `paths`, `explain`, and `normalize`—used for comprehensive validation, rapid failure path localization, static constraint explanation, and safe default value application prior to validation. Emphasizing security and stability, the plugin strictly adheres to zero network access and zero dynamic code execution, operating within fixed resource limits such as data size (256 KiB), nesting depth (64 layers), and node count to prevent performance abuse. Its validation kernel adopts a pure function style and supports local `$ref`.
Read Moredsh-tool-regex: Deterministic regex tool for DSH
`dsh-tool-regex` is a zero-dependency regex tool plugin for the DSH workflow, designed to address the challenges of verifying regex results directly performed by models and the risks associated with introducing temporary scripts. This plugin adopts a pure function design and outputs JSON uniformly. It provides four core operations: `test` to check for matches, `find` to extract capture groups and indices, `replace` to perform safe global replacements with backreference support, and `explain` to statically parse regex structure without executing matches. The plugin features strict security boundaries, including limits on input length, pattern size, output volume, and the number of matches, as well as a 1000ms hard timeout mechanism to prevent resource exhaustion caused by malicious or complex regex patterns. It requires Node.js ^22.19.0 or higher and is installed to a specified path via the DSH plugin command.
Read Moreomdsh-dev/dsh-tool-json: DSH JSON Query Tool Plugin
`omdsh-dev/dsh-tool-json` is a lightweight, read-only JSON query tool developed for DSH plugins, licensed under the MIT License. It provides a tool named `json` that precisely extracts fields from JSON values or strings using a custom JMESPath-inspired path subset, avoiding the confusion of string searching or the high cost of a full query engine. Core features support dot notation access, array indexing, access to keys with special characters, and array wildcard projection. The plugin intentionally limits query scope, does not support filters, pipes, function calls, or object field enumeration, and multi-level wildcards return nested arrays rather than flattened results. Regarding security, it employs a hand-written parser to avoid `eval`, prevents prototype chain pollution, and sets resource limits on query length, depth, and element count. Installation requires adaptation.
Read Moredsh-tool-encoding: DSH UTF-8 Text Encoding, Hash, and UUID Tool Plugin
dsh-tool-encoding is an encoding and hashing tool plugin for DSH agent workflows, providing base64/base64url/url/hex encoding and decoding for UTF-8 text, hashing (md5/sha1/sha256/sha512), and UUID v4 generation. It is designed as a zero-dependency, zero-process, pure function, aiming to replace shell external commands and reduce process scheduling and concatenation costs. All operations are processed as UTF-8 text, with an input limit of 1 MB and an output limit of 4 MB. The tool features strict validation mechanisms, including RFC 4648 canonical validation, fatal mode UTF-8 decoding, rejection of isolated surrogates, and hex decoding validity checks. This plugin is suitable for web and high-performance applications.
Read Moredsh-auto-chess: Simplified Auto Chess Plugin for DSH Web
`dsh-auto-chess` is an MIT-licensed DSH Web plugin developed by `omdsh-dev`, designed to provide a simplified auto-battler interface for observing the decision logic of large language models under fixed rules. The plugin supports Human-vs-AI (where the user plays the Blue side) and Dual-AI battle modes, allowing users to configure independent models and edit prompts for the Blue and Red sides respectively. It also allows selecting thinking levels such as Off, High, and Max, automatically falling back to default settings if the model does not support them. The system defaults to a single decision timeout of 300,000ms and an output limit of 32,000 tokens. After each AI decision, users can view the full reasoning text on the panel; if an invalid operation or transient failure occurs, the system will indicate the cause and automatically retry within the budget. Building the plugin requires executing `pnpm run build` and running `dsh`...
Read Moredsh-lan-access: Add LAN access switch in DSH Web
`dsh-lan-access` is a DSH Web plugin designed to simplify the configuration management of DSH LAN access. By default, DSH binds only to the loopback address (127.0.0.1) to ensure security and refuses to bind to 0.0.0.0. This plugin provides a "LAN access" toggle in the `Settings → General` interface, allowing users to flexibly switch the binding address within a trusted local network: when enabled, it binds to 0.0.0.0, enabling devices on the same network segment to access it via `http://<LAN-IP>:3080`; when disabled, it reverts to 127.0.0.1. The core functionality of the plugin requires no modification of the DSH core code; it achieves state reading, writing, and configuration synchronization by registering secure JSON routes, and supports access to the remote settings page as well as browser compatibility patches. Installation requires executing...
Read MoreDeepSeek Harness Plugin: unified-agent-memory (Shared Obsidian Vault Agent Memory)
unified-agent-memory is a DSH plugin developed by Noelune, designed to address the memory isolation issues of multiple agents such as dsh, Codex, Claude Code, and Hermes. It provides a shared Obsidian Vault memory repository, utilizing a local SQLite FTS5 index for efficient retrieval, and supports developers in directly editing and manually reviewing facts. Core functions include full lifecycle management covering knowledge ingestion, promotion, conflict adjudication, and reversible forgetting (archiving without deletion). The plugin supports basic BM25 retrieval and optional hybrid retrieval (combining semantic vectors and concept graphs), and automatically distills session content through `memory digest`. In terms of security, it features prompt isolation and credential sanitization mechanisms. Installation requires Node...
Read Moreniyongsheng/free-vision-skill: DSH Local Image Understanding and OCR Plugin
`niyongsheng/free-vision-skill` is a DSH plugin for macOS that provides fully local image understanding and OCR services based on the macOS Vision Framework, ensuring images do not leave the local machine. This plugin is licensed under the MIT license, and its core functions include image description, text extraction, and table structure recognition. It offers two main tools: `view_image` and `ocr_image`. The former is used to describe image scenes, characters, and QR codes, etc.; the latter supports text extraction and can obtain table structure and coordinates via the `layout=true` parameter. Supported input formats include http(s) URLs, base64, and local paths. Installation requires using the `dsh plugin add` command and adding it to the configuration file...
Read MoreDshViewModes: Add Verbose, Normal, and Summary Session Display Modes for DSH Web
DshViewModes is a display plugin for the DSH Web client maintained by NigelYao, designed to optimize user experience through views with different levels of granularity. This plugin is based on the MIT license and only modifies browser-side display without interfering with core files, providing three modes: `Verbose`, `Normal`, and `Summary`. `Verbose` retains complete debugging information such as thinking and tool calls; `Normal` (default) groups process steps by semantic stages, making it suitable for daily reading and supporting the expansion of details; `Summary` hides process noise and displays only the result summary. Users can switch modes via the session header or keyboard shortcuts `Alt+1/2/3`, the selection state is saved in `localStorage`, and the interface language follows Harness Settings.
Read Moredsh-billing: DSH Web GUI Real-time Billing Monitoring Plugin
dsh-billing is a real-time billing monitoring plugin for the DSH Web GUI, centrally managing local model call usage, balance, and cost estimation. Its core features include a sidebar displaying real-time balance (refreshes every 2 seconds) and status alerts, as well as an expandable panel for viewing consumption details, token usage, and recent transactions. It supports DeepSeek v4 peak-valley pricing and multi-vendor unit price matching, and allows syncing of official balances and unit prices via API Key. The plugin offers a settings page for configuring thresholds, unit price tables, and data management, and integrates the `billing_balance` Agent tool for querying. Ledger data is stored persistently and automatically restored upon restart. Installation requires no build and supports direct installation from GitHub or locally. Please note that it only tracks local calls, API keys are stored in plain text locally, and billing is estimated. It is suitable for real-time monitoring scenarios.
Read Moredsh-model-balance: Display account balance and quota before the DSH Web GUI model selector
dsh-model-balance is a plugin designed specifically for the DSH Web GUI, aiming to address the pain point of frequently switching consoles to check balances during the development of multiple model providers. This plugin does not alter the model invocation chain; instead, it displays the real-time balance or quota of supported API vendors—such as DeepSeek, StepFun, Kimi, OpenRouter, MiniMax, and xAI—directly before the model selector. For vendors without an API interface, such as Tongyi and Xiaomi, it provides a link to their console. The plugin features an intelligent refresh mechanism, including instant display when switching models, forced refresh after a conversation ends, timed polling every two minutes, and manual refresh on click. It also utilizes distinct caching strategies for success and failure results to ensure state accuracy. Users can easily install it using the npx command from GitHub or npm.
Read Moredsh-subagent-profile: Manage DSH Sub-agent's Model, Reasoning Strength, and Tool Scope Per Task
`muzyLink/dsh-subagent-profile` is a sub-agent dispatch plugin for DeepSeek Harness, aiming to unify the management and optimization of sub-task configurations. This plugin enables users to flexibly select models, reasoning intensity, and tool scopes based on task requirements, and save common combinations as reusable named Profiles. Its core capabilities include built-in multi-layer security check mechanisms, cost estimation and savings analysis, and a comprehensive decision ledger that records requests, effective configurations, execution processes, and settlement details, achieving transparency and traceability in dispatch. After installation, users can manage Profiles via the settings page, use the `dispatch` tool for foreground or background dispatch, and review decision records in session tabs. Regarding security, the plugin removes the high-risk tool `run_code` by default and requires a host environment.
Read MoreDeepSeek Harness Desktop: Embedding the dsh web interface into a native desktop window
DeepSeek Harness Desktop is a desktop window plugin within the DeepSeek Harness ecosystem, designed to address the user need for quick local access to the `dsh web` interface without relying on a browser. Built on Electron, this plugin directly loads the real Web UI of the current `dsh web` profile instead of an independent frontend, thereby preserving all original capabilities. It supports multiple launch entry points, including desktop shortcuts, the Start menu, the installation directory, and the `/desktop` command within the Web UI. By default, it uses a light theme and can be switched to the Dark Codex skin, with external links opened via the system browser. Installation requires Node.js 22.19+, and it is available as a Git clone, Zip portable version, and Setup.exe.
Read Moredsh-nuphus-mcp: Mount nuphus-mcp's computer-use tools in DSH
dsh-nuphus-mcp is an MIT-licensed DSH plugin maintained by mrpulor-gh, designed to mount nuphus-mcp as a native DSH tool, enabling Agents in the web profile to possess desktop or computer usage capabilities. The plugin registers 38 tools prefixed with `mcp__nuphus-mcp__*`, automatically installing missing dependencies upon first startup. It also enables the `--confirm-write` security mechanism by default, requiring the `confirm: true` parameter for write operations. It supports configuring parameters such as `serverName`, `command`, and timeout via `NUPHUS_MCP_*` environment variables and `cordis.yml`. The installation command is `npx -p @deepseek-ai/ds`.
Read Moredsh-find-skill: Integrating skills search, installation, and lifecycle for DSH
`dsh-find-skill` is an MIT-licensed plugin for DeepSeek Harness (DSH) (v0.2.1) designed to bridge the `vercel-labs/skills` ecosystem, providing LLM-driven skill search, installation, and lifecycle management. The plugin supports three scopes: temporary, project, and global. Temporary skills are visible only within the current session and are automatically cleaned up upon session destruction; project or global skills are persisted to the hosted root directory. Core features include `skill_find` (searching based on the `skills.sh` API), `skill_install` (isolated installation via CLI), and `skill_remove`, along with a `/skill` subcommand for manual maintenance. Configuration options include the search base URL, default scope, and Provider ranking.
Read Moredsh-vision-tools: Adding image understanding capabilities to DSH sessions
`dsh-vision-tools` is a visual plugin within the DSH ecosystem, maintained by moon09300731 (MIT License). It provides a global `vision_understand` tool that uses OpenAI-compatible APIs (supporting Zhipu, Alibaba, SiliconFlow, and OpenAI) to understand local images, enabling image descriptions, text recognition (OCR), and image-based Q&A. Images can be sent via paste, drag-and-drop, or button selection; they are automatically saved to a specified directory and filled into the prompt. The plugin supports rate limiting auto-degradation, real-time application of global and project-level configurations, and provides an "Image Recognition" entry in the Web interface. The installation command is `dsh plugin --profile web add dsh-vision-tools`, and `dsh web` needs to be restarted. Configuration requires...
Read More