dsh-global-task-list: DeepSeek Harness Global Task Library Plugin

`dsh-global-task-list` is a plugin designed for DeepSeek Harness to address issues such as fragmented tasks, invisible progress, and state desynchronization. Core features include: 1. **Task CRUD and Persistence**: Provides `task_add/list/update/delete` tools, with tasks stored in a cross-session `storageDomain` to support long-term maintenance. 2. **Persistent Floating Panel**: A browser-side panel located in the bottom-right corner that subscribes to SSE in real-time to refresh task status, supporting modification, deletion, and viewing of structured content. 3. **Background Job Synchronization**: Associates tasks via `jobId` to automatically sync terminal states of background jobs (such as completion or failure), adapting to master-slave Agent collaboration workflows. 4. **Generative UI**

Read More
dsh-plugin-8r: DSH's 8R local determination plugin

dsh-plugin-8r is a local judgment plugin for DeepSeek Harness (DSH), maintained by SensenMeng. Its core function is to automatically inject a bold "诶!…" prefix at the beginning of the AI response when a user message triggers a specific rule (8R) via local JavaScript logic. This process is completed entirely locally; the model does not need to understand the rules and does not consume context or tokens, thereby avoiding reliance on model judgment. The plugin uses `agent/pre-step` for judgment, injects the prefix into the stream output, supports failure retry, and prevents duplicate injection within the same turn. It only takes effect for the main session, excluding sub-agent messages, and provides a local diagnostic interface. Installation requires executing an add command in the DSH profile directory and restarting, and depends on the dependency `@de...

Read More
DSH Session Manager: Add a set of common operations for DeepSeek Harness Web sessions.

Semidia/dsh-session-manager is a community plugin (v0.1.1, MIT License) for the DeepSeek Harness (DSH) Web client, designed to enhance session management efficiency. The plugin enhances the interface through two main entry points: a right-click context menu in sidebar session rows supporting pinning/unpinning, marking as unread, renaming, archiving, forking, opening in the file manager, copying the working directory/title/ID/deep link, and opening in a new window; and a Session Manager entry at the bottom of the sidebar for the current session, supporting export to ZIP, archiving, forking, and copying information. The plugin uses only publicly available client services, but some deep link features require the `dsh-deeplink` plugin.

Read More
dsh-fleet: DSH Multi-Machine Computing Power Networking Plugin

dsh-fleet is an MIT-licensed plugin for DSH designed to connect idle computing power from local networks (LANs), cloud ECSs, or friends into a shared team, enabling cross-machine task dispatch and result return. It supports Windows, Linux, and macOS and relies solely on built-in Node.js modules. Core features include generating unique machine identities (machineId and SK), multi-team management, HMAC-SHA256-based trust handshakes, heartbeat monitoring, and online status viewing. The Master side dispatches tasks using the `fleet_dispatch` tool, while the Worker side automatically executes tasks and maintains sessions for traceability. Upon startup, the plugin registers 12 model tools and a sidebar panel, providing operations such as machine card generation, team joining, and member status checking. It is installed using the dsh plugin command.

Read More
dsh-adb: ADB Device and Bench Operation Plugin for DeepSeek Harness

dsh-adb is an MIT-licensed plugin developed by SamXiaBing, designed to provide Android device and vehicle test bench debugging tools for DeepSeek Harness (DSH) agents. It encapsulates core functions such as device discovery, structured Logcat reading, APK installation and file transfer, performance snapshot collection, device health reporting, crash monitoring, and conditional waiting. The plugin supports running Logcat tasks in the background and uses an operation ledger for persistent recording to support application rollback. Additionally, it provides one-click inspection and a Web device panel to facilitate real-time viewing of logs and performance data. Installation requires adding the plugin via the `dsh plugin add` command and restarting the GUI, with configuration options including `adbPath` and `defaultSerial`. This plugin is applicable for...

Read More
dsh-plugin-muapi: Integrating MuApi generative models into DeepSeek Harness

`dsh-plugin-mui` is a DeepSeek Harness (DSH) plugin maintained by SamurAIGPT and released under the MIT license. By registering the `muapi_generate` tool, it encapsulates over 100 image, video, and audio generation models from the MuApi platform into DSH agents, simplifying the development of glue code for authentication, task submission, result polling, and output URL retrieval. The plugin workflow involves submitting a generation task to the MuApi API, polling for results at configured intervals, and finally returning the output URL. Key configuration items include the enabled status, API Key environment variable reference, base URL, polling interval, and timeout (default 600 seconds). Users need to use `dsh plugin ad...`

Read More
dsh-tool-describe-image: Add image understanding capability to DSH

`sala003/dsh-tool-describe-image` is an MIT-licensed plugin within the DSH ecosystem designed to equip text models with image processing capabilities. By connecting to any OpenAI-compatible vision endpoint (such as Bailian, Zhipu AI, etc.), it converts images into text descriptions or structured HTML for use with DeepSeek. Core features include a Web interface for "paste and recognize," tool invocation via path to describe images, querying DeepSeek account balance, and providing a customizable entry point for a "Whale Girl" desk pet. The plugin supports zero-configuration startup, allowing API keys to be configured later via environment variables or the settings panel. Installation requires DSH 0.1.0-rc.6+, installed globally via npm and enabled. It supports PNG, JPG, WebP, and GIF formats, with a single-image limit of 8Mi.

Read More
dsh-git-plugins: DeepSeek Harness Git and Source Code Control Plugin Suite

`sakthiveltofficial/dsh-git-plugins` is an MIT-licensed plugin suite for DeepSeek Harness (DSH) designed to convert Git operations and platform interactions into agent-callable tools. It requires Node.js >= 20 and git CLI support. Core features include: 1. Local repository operations covering common commands such as clone, commit, push, merge, and rebase; 2. Hosted platform operations supporting PR/MR, issues, releases, CI/CD, and security alert management for GitHub, GitLab, Bitbucket Cloud, Azure DevOps, and Gitea; unsupported operations will return...

Read More
Pack Agent: Pack skills, rules, and MCP into .pack.zip and install to harness

pack-agent is an npm package designed to simplify DeepSeek Harness (DSH) and agent configuration management, adhering to the philosophy that "everything is a plugin." It addresses the issues of error-prone and difficult-to-uninstall manual copying of skills, rules, and MCP configurations across different harnesses such as Claude Code and Codex. The plugin packages these configurations into .pack.zip files via commands and installs them into the target harness, utilizing an installation ledger to achieve clean uninstallation. It supports multiple harnesses including claude-code and codex, requires Bun >= 1.1, and is licensed under the MIT License. Core functions include initialization, export, installation, and configuration ejection (uninstall), and it supports Model Context Protocol (MCP).

Read More
DeepSeek Harness Plugin dsh-skill-importer: Import, organize, and call Agent Skills in the Web UI

`dsh-skill-importer` is a Web UI plugin for DeepSeek Harness (DSH) designed to address issues such as the dispersed management of Agent Skills, duplicate naming, and missing resources. This plugin allows users to import, verify, deduplicate, and migrate Skills from local Markdown files, URLs, or other AI coding agents like Claude Code and Codex via a web interface, automatically incorporating them into Harness's discovery mechanism. Core features include support for batch migration of full Skills directories, preservation of scripts, references, and resource files, and the provision of intelligent deduplication and replacement protection mechanisms. Users can invoke Skills via the Composer Picker, the `/skills` command, or by directly typing their names.

Read More
LoongPort DSH Plugin: Integrating an OpenAI-compatible Provider into DSH

`loongport` is a plugin and CLI tool for DeepSeek Harness (DSH) maintained by SailingLoong, designed to simplify the integration of OpenAI-compatible providers. This plugin (v0.2.0, MIT licensed, requires Node.js >= 20) primarily offers two functions: first, a visual configuration entry in DSH under "Settings → LoongPort," where users can select pre-verified providers, paste API keys, and choose models; and second, advanced CLI commands to write custom OpenAI-compatible endpoints into the DSH configuration. The CLI supports options such as `--base-url`, `--model`, and `--route`.

Read More
DSH Cost Meter: Round-based Cost Tracking for the DSH Web GUI

DSH Cost Meter is a cost tracking plugin for the DeepSeek Harness Web GUI, designed to address the issue of historical session cost recalculation caused by price adjustments. Its core mechanism employs price snapshot anchoring to ensure that the cost of each completed response round is fixed at the price applicable at the time of usage, unaffected by subsequent changes. Key features include snapshot-based cost display per round and per message, real-time account balance retrieval and caching, real-time cost estimation during streaming (replaced with precise values upon settlement), peak-valley pricing support, sub-agent cost aggregation, and a bilingual interface. The plugin offers interface components such as Composer dock readings, a cost view tab, and real-time estimation in the header, and supports manual refresh interfaces, but access is restricted to local or trusted hosts only. Users can quickly install it to the Web profile via the DSH CLI, and it is compatible with DSH.

Read More
dsh-coding-preset: Preset for using persistent pwsh in DSH coding mode on Windows

dsh-coding-preset is a deprecated (DEPRECATED) Windows-adapted DSH coding mode preset designed to replace the default bash with a persistent PowerShell 7 (pwsh) session. This preset retains the core boundaries of the official minimal mode, exposes only the pwsh and str_replace_editor tools, and implements state preservation via node-pty/ConPTY. In the danger-full-access mode, variables, functions, and directories persist between invocations; whereas in read-only or workspace-write modes, it executes single-shot commands without state preservation. It also supports single-shot sandbox elevation, allowing the carrying of reasonable...

Read More
dsh-BigfishPet: DeepSeek Harness Universal Desktop Pet Plugin

dsh-BigfishPet is a universal desktop pet plugin designed specifically for DeepSeek Harness (DSH), licensed under the MIT License. Configured via the DSH settings page, the plugin uses PySide6 to drive a transparent desktop helper displaying a "Whale Girl" image, mapping agent task states in real-time. Its core functionality includes a real-time state driver based on `session/event` that reduces tasks to five states: thinking, working, waiting, completed, and error, displaying corresponding animations based on priority (Waiting > Error > Working > Thinking). The plugin supports rich interactive experiences, such as triggering random dialogue with the left mouse button, automatically muttering or falling asleep when idle, and controlling the host process or opening the WebUI with the right mouse button. State data is persisted to `~/.dsh/pet.json`.

Read More
rxa3c/chat2skill: Extract project memory and reusable skills from assistant sessions

`rxa3c/chat2skill` is a plugin designed to address the issue of corrections, preferences, and project facts in assistant sessions failing to be stably accumulated. Its core functions are "post-session extraction" and "pre-session injection": by analyzing session signals, it distills information into local SQLite memory and `SKILL.md` skill files, automatically retrieving and injecting it in subsequent sessions. The plugin supports LLM mode (Proposer/Generator/Judge flow) and non-LLM mode (keyword detection). Data is stored locally, while the cloud is used solely for stateless execution and claims not to store data. Typical extraction requires approximately 4 LLM calls and is triggered only upon detecting correction or constraint signals. Documentation suggests it is best suited for Claude Code, Codex, and Cursor, while other agents require lifecycle support.

Read More
dsh-login-gateway: Adds login gatekeeper and reverse proxy for DSH Web UI

dsh-login-gateway is a DeepSeek Harness plugin developed by runfali, designed to provide controlled external access to the dsh Web UI, which listens exclusively on the local loopback interface. The plugin provides a username/password login interface at 0.0.0.0:3081; upon successful authentication, it performs full reverse proxying of HTTP and WebSocket requests to the local 3080 port while automatically rewriting request headers to pass dsh's trust verification. Built on Node.js 22+, the project features zero runtime dependencies and is released under the MIT license. It employs a zero-intrusion design, ensuring the installation process does not modify any existing dsh files or configurations. Core security features include login rate limiting based on IP and username, User-Agent session binding, and automatic credential revocation upon password changes.

Read More
dsh-qwen-mm: Adding Multimodal Reading and Tool Integration to DeepSeek Harness

dsh-qwen-mm is a DeepSeek Harness (DSH) plugin maintained by RRRosmontis, designed to enable text-only models to process multimodal content such as images, videos, and documents via MCP tools. This plugin integrates the capabilities of Qwen-MM-Plugins, offering features like multimodal reading, cloud-based vision/OCR/ASR, web search, video processing, and 3D/CAD toolchains. Its core functions include basic reading without a key, an advanced API requiring a DashScope key, a retrieval function requiring keys from search service providers, and support for long-video Q&A as well as tools like Blender and FreeCAD. A standout feature, "Drag-and-Drop Image Upload," allows users to drag images into the input box for the model to read; however, this feature relies on core changes to DSH that have not yet been merged into the official version and requires the use of a specific fork.

Read More
dsh-searxng: Integrate SearXNG search for DeepSeek Harness

dsh-searxng is an MIT-licensed plugin (v0.2.1) for DSH designed to provide Agent-based Web search capabilities via SearXNG using `ctx.web`. The plugin supports two primary modes: automatically deploying a managed, loopback-only SearXNG Docker instance, or connecting to existing local or remote SearXNG instances without starting Docker. It requires Node.js 20+ and Docker Compose v2 (Docker path only) and does not support Podman. Core features include quickly initializing configuration and activating a Profile via the `setup` command, supporting port specification, and multi-profile environment isolation; it provides `

Read More
DSH Plugin dsh-longtask-orchestrator: Long Task Orchestration

The `dsh-longtask-orchestrator` within the DSH plugin ecosystem is an orchestrator plugin designed specifically for long tasks and follows the MIT license. It adopts a multi-model collaboration mode, emphasizing the philosophy of "everything is a plugin": Codex is responsible for planning, scoring, and final review, DSH handles specific execution, and Kimi manages summaries, cross-validation, and visual acceptance. The plugin decomposes long tasks into standard steps, including initialization, plan generation, subtask execution, result scoring, supplementary validation, and final review. It registers a total of 9 tools and supports the automatic triggering of loop templates via skills. Installation requires using a local link command and restarting the DSH web interface. Typical usage includes switching to a specific model group, initializing the task, executing subtasks according to a checklist, iterative scoring, and final review. In terms of configuration, Codex defaults to invoking local binary files and performs auto-detection.

Read More
dsh-wallpaper: Configure wallpaper, blur, and transparent surfaces for DSH Web

dsh-wallpaper is a community-maintained DSH Web wallpaper plugin (v0.1.0, MIT License) designed to customize backgrounds and adjust display effects. It supports independent wallpaper configuration for the main interface and the trajectory page, offering options such as no wallpaper, built-in gradients, image upload, folder import, and image cropping. Users can fine-tune blur, surface opacity, text color schemes, and image fill position, while the trajectory page provides multiple color schemes. Data is stored in browser IndexedDB (for images) and localStorage (for parameters), with a note that clearing the cache may result in image loss. It supports installation via GitHub, Release, or source code and is cross-platform compatible, though the folder import feature relies on browser webkitdirectory support (Firefox does not support it). After installation, it needs to be configured in DSH Web...

Read More
dsh-shortcuts:为 DeepSeek Harness WebUI 提供可自定义键盘快捷键

dsh-shortcuts is an open-source keyboard shortcut plugin developed for DeepSeek Harness WebUI, aiming to address issues where common operations are scattered across menus and mouse reliance interrupts the rhythm of input. Based on the MIT license, this plugin runs client-side without network requests or access to business data, supporting two modes: dynamic loading at the WebUI session level and static installation for the Desktop host. Its core functions include providing 34 predefined shortcuts categorized into six main groups: Session, View, Clipboard, Model, Permissions, and System. It allows users to freely record, clear, or disable any key combinations and features automatic conflict detection. Users can view the shortcut quick reference guide and diagnostic panel via ⌘/ and seamlessly toggle between three permission levels—Read-only, Workspace Write, and Full Access—using ⇧Tab.

Read More
rice-awa/dsh-lan-gateway: A plugin to securely expose the DSH Web GUI to the LAN / Public Network

`rice-awa/dsh-lan-gateway` is a DSH plugin designed to securely expose the DSH Web GUI to a local area network (LAN) or the public internet. It resolves the security trade-off where DSH binds exclusively to 127.0.0.1 (preventing access from other devices) versus directly listening on 0.0.0.0 (which poses security risks) by employing a reverse proxy gateway. The plugin maintains DSH's local binding and establishes a separate gateway to forward requests, implementing a source-based access control strategy: LAN and local sources have password-free access, while non-LAN sources require verification via a login page and HMAC session cookies. The plugin supports optional TLS (with self-signed or custom certificates), rate limiting, CSRF protection, and hot configuration management. Users can enable, configure passwords, and monitor status via the `lan_gateway` command or DSH chat.

Read More
AgentAstra: A Spatialized Conversation and Trajectory Visualization Plugin for DeepSeek Harness

AgentAstra is an experimental client plugin for DeepSeek Harness designed to project the Agent execution process into a traversable spatial view, replacing the traditional linear message flow display. By leveraging three dimensions—semantic time, causal channels, and information hierarchy—the plugin intuitively visualizes key nodes and their causal triggers, such as user prompts, Agent replies, tool calls, sub-Agent delegation, file read/write operations, and external resource access. Core features include displaying conversation content as readable cards with support for keyboard navigation, timeline dragging, and node focus interactions; allowing precise file snapshots to be dragged directly into chat box drafts; binding the view to the current session to automatically release WebGL resources upon unloading; and providing a text fallback when WebGL is unavailable. Installation is performed by adding the plugin using the `dsh plugin` command.

Read More
dsh-git-credentials: Keep Git platform API token values out of model context

`dsh-git-credentials` is a DeepSeek Harness plugin designed to address the security challenge of preventing tokens from entering the model context when calling APIs of Git platforms like GitLab and GitHub within DSH agents. The plugin decouples credential management from model logic, allowing the model to process only business parameters, while token values are decrypted by the plugin from local AES-256-GCM encrypted storage and directly used in the outbound Authorization headers. Its core features include limiting tool invocation scope by Provider (e.g., `gitlab_*`, `github_*`), and providing a Web settings panel that supports CRUD operations for sites and hot token rotation, with changes taking effect immediately without requiring a restart.

Read More
dsh-recall: Adding session history original text retrieval for DeepSeek Harness

dsh-recall is a DeepSeek Harness (DSH) plugin maintained by Relistencode, designed to address the difficulty of locating original information after context accumulation and compression in long sessions. This tool transforms "searching chat history" into a recall mechanism callable by an agent. Version 0.2.2, based on the MIT license, requires a Node >= 22 environment and runs completely offline. Its core functions include a three-layer hybrid retrieval that merges literal, fuzzy, and semantic results, and automatically degrades when semantic coverage is insufficient. It adopts a progressive disclosure strategy, returning lightweight coarse recall first and expanding the original text window as needed. It aggregates multiple mentions of the same topic into complete events. It supports the agent to actively recall defaults when the context is compressed or missing.

Read More