dsh-unsloth-hands: Offloading repetitive tasks of DeepSeek Harness online models to local Unsloth Desktop
dsh-unsloth-hands is a pure client-side plugin maintained by MicroHEROX (MIT licensed, v0.1.0), allowing DeepSeek Harness online models to perform repetitive short tasks such as batch rewriting, translation, string cleaning, and screenshot OCR.
Read More@MarvekG/dsh-plugins: Fix DSH redundant sandbox upgrade errors and WSL path open failures
This article introduces the DSH out-of-tree plugin suite `@MarvekG/dsh-plugins` maintained by MarvekG (MIT-licensed, renamed from the old package `@MarvekG/dsh-bug-fix`), providing separate entry points for two common issues. The first one is `sandbox-same-mode`: when the session is already `danger-full-access`...
Read Morecatalog-capabilities-zh: Maintain a source-based Chinese capabilities catalog for DSH
catalog-capabilities-zh is an MIT-licensed community project maintained by MeowTnT3r. Acting as a DSH bundle plugin and runtime skill, it maintains a source-based Chinese capabilities catalog for skills, plugins, and marketplace capabilities. It follows the "everything is a plugin" philosophy, addressing the pain points of scattered DSH user plugins, predominantly English READMEs, and difficulty in finding them. Its core feature is a default curated mode where only user-confirmed entries are written to the documentation; scanning serves only as a discovery aid, and explicit full inclusion requires the `--all` or `full` mode. It does not rewrite installation logic; instead, it orchestrates existing trusted installers in the environment during installation. If the target is already installed, it only verifies update instructions. When only providing a link, it reads public repository metadata for checks and introduction without executing the repository code. Entries support sorting by Chinese...
Read Moreskill-bartender: Minimal Skill Allocation — Task-Skill Matching for DeepSeek Harness (Meta-skill)
# Introduction DSH's philosophy is that everything is a plugin. After loading many skills, the agent faces a constantly growing directory. While it can see the name and description of each skill, the common behavior is "over-dumping": loading too many skills, loading the wrong skills, or missing the workflow skill that was originally intended to string the whole task together. This presents an asymmetric cost: once a wrong skill is loaded, the skill content remains in the conversation history; whereas missing a load only results in the loss of a single tool call. Since missing is cheaper than being wrong, a reasonable strategy is to err on the side of loading less. skill-bartender is implemented according to this logic—it does not provide specific capabilities, but is only responsible for deciding "which skill this task should load, or whether to load it at all."
Read Moredsh-theme-pack: 16 Third-Party Themes for DSH Web GUI
dsh-theme-pack is an MIT-licensed DSH plugin maintained by math-lrz. It provides 16 sets of popular community color themes for the DSH Web GUI, including Catppuccin, Gruvbox, Nord, Dracula, Solarized, and others, featuring both light and dark variants. It integrates non-intrusively via D...
Read Moredsh-vulnsec-bridge: A security plugin for integrating VulnClaw and DeepSec into DeepSeek Harness
# Introduction When performing security-related tasks in DeepSeek Harness (dsh), the common practice is to open a separate terminal to run penetration testing or code auditing CLIs, then copy the output back to the session for further analysis. As tool capabilities are not registered as dsh tools, the model cannot invoke them directly, and their results cannot be integrated into subsequent automated workflows. dsh-vulnsec-bridge is a bridging plugin designed for this scenario: it wraps VulnClaw (AI penetration testing CLI) and DeepSec (Shield code auditing + Spear authorized penetration) into dsh plugins, providing 18 security tools and a graphical "Security Workbench" panel. Below, we introduce its features, installation methods, and typical usage. First, let's discuss the prerequisites: the plugin's penetration...
Read More@max-null/dsh-memory: DeepSeek Harness's Cross-Session Plaintext Memory Plugin
# Introduction Using a harness like DSH for agent development, a common pain point is that nothing is retained between sessions: when you tell the model in the previous round that "This project uniformly uses Vue3 `<script setup>`", it forgets it in the new session. The common remedy is to stuff memory into a vector database and rely on similarity search—the penalty is that the memory itself becomes a string of unreadable numbers, and expired information is hidden in the database, making it unobservable and unfixable. `@max-null/dsh-memory` takes a different path: all memory is plaintext, retrieval uses deterministic BM25, and writing requires passing a manual confirmation gate. Below is an introduction to its design and usage. # What is this `@max-null/dsh-memory` is a cross-platform solution for DeepSeek Harness.
Read Moredsh-file-changes: DSH Web Plugin to Display File Changes Panel at the End of Each Response
# Introduction When developing agents using DeepSeek Harness (DSH), a common issue arises: determining which files a model creates or modifies during a single turn often requires sifting through lengthy transcripts. The built-in "Produced" section only provides a general list of artifacts; to understand exactly what changes were made to a specific file, one must still inspect it manually in the editor. `dsh-file-changes` is a DSH Web plugin maintained by mixin-ai that addresses this specific problem: it lists the files created or modified during the current turn directly beneath the message at the end of each turn. It supports viewing diffs directly, opening files with the system default application, or locating them in the OS file manager. This article introduces the plugin in the order of features, installation, underlying principles, and considerations. # What is it?
Read MoreMingxi2077/dsh-plugin-review: Equip DSH with a code review discipline
dsh-plugin-review is a DSH (DeepSeek Harness) plugin developed by Mingxi2077, designed to embed a code quality review discipline into agent coding workspaces. Its core features include procedural health scoring across 8 dimensions (structure, maintainability, consistency, robustness, testing, documentation, performance, security) that is independent of subjective model descriptions; real-time visualization of results using pure SVG radar charts; and support for AI to actively trigger review, repair, and verification after completing a module via the `review_audit` tool. Review history supports categorization, pagination, and a trash bin (retained for 7 days by default).
Read MoreDSH Plugin Introduction: mattismegevand/dsh-dock
This article introduces the DSH plugin `mattismegevand/dsh-dock`, but due to the failure to fetch the repository README, verifiable information is extremely limited. Only the repository path, GitHub address, and its inclusion in the community directory (skillhub.cn) have been confirmed; however, inclusion in a directory does not constitute an official recommendation. The description, star count, and category listed in the directory are unverified. Reliable sources for the feature list, installation commands, usage examples, and license are unavailable, so this article does not elaborate on them. The author advises evaluators to open the repository directly to review the README and LICENSE, and to browse the source code to confirm its behavior before deciding whether to install it. Additionally, it is reminded that DSH plugins run with `dsh` process permissions; in the absence of documentation, one should interpret...
Read Moredsh-approval-comment: Add "Reject with comment" to DSH Web Approval
# Introduction If you are running agents using DSH (DeepSeek Harness) in web mode, you have likely encountered these two issues: 1. When a tool call triggers an approval window and you click "Reject," the model does not stop; instead, it proceeds to the next round of reasoning, guessing the reason for your rejection and continuing with an alternative plan. 2. When you manually stop a task, the rejection is not added to the model's context, and subsequent requests proceed as if nothing happened. In real-world collaboration, refusals typically come with a reason: "Do not touch this file; move it to the working directory." The default approval process neither captures this reason nor allows the model to rethink based on it. `dsh-approval-comment` is a plugin designed to fill this gap. # What is this? `dsh-approval-comment` is a...
Read Moredsh-plugin-manager: One-click Enable/Disable Installed Plugins in dsh Web GUI
# Preface Most users of DSH embrace the philosophy that "everything is a plugin": functionality is mounted on demand and can be detached when unnecessary. However, in practice, enabling or disabling a plugin often involves opening the `cordis.patch.yml` file in the profile, manually writing `disabled: true`, and then restarting the web service to verify the effect. With a large number of plugins, this process becomes tedious. `dsh-plugin-manager` brings this functionality into the web GUI: it lists the downloaded plugins in the current profile, allows for one-click enable/disable via buttons, and persists changes to the patch file, taking effect immediately. Below is an introduction to its features, installation methods, and usage notes. # What is it? dsh-plugin-manager by Mg-
Read Moredsh-terminal: A DSH terminal plugin awaiting verification
# Introduction DSH (DeepSeek Harness) operates on the philosophy of "everything is a plugin": capabilities such as terminal access, browsing, and file operations can be augmented into agents via third-party plugins. A member of the community maintains a plugin directory that catalogues numerous such plugins, and dsh-terminal is one of them. However, this introduction will be brief. The reason is straightforward: as of the writing of this article (2026-09-07), the scraping of this plugin's GitHub README failed, and I lack first-hand materials to verify its functionality, usage, and license. Rather than fabricating a detailed feature list, I have placed the confirmed information and links here, leaving the rest for you to verify in the repository yourself. # Confirmed Information The following points are taken from the entry in the plugin directory and are confirmed to be valid:
Read Moredsh-smart-profile: Selecting Plugins for DeepSeek Harness from the Repository
dsh-smart-profile is a DSH plugin maintained by Makima667 (MIT, v1.0.0, Node ≥ 20, zero dependencies). Designed to address the pain point of uncertainty regarding which plugins to install in an "everything is a plugin" ecosystem, it provides a plugin selection process based on the repository's technology stack. It first heuristically detects the project's technology stack (covering front-end and back-end frameworks, databases, infrastructure, CI, testing tools, monorepo, etc.).
Read Moredsh-session-stars: Cross-Workspace Star and Reopen DeepSeek Harness Sessions
dsh-session-stars is a Web-only plugin for DeepSeek Harness Web, maintained by malevrigns and licensed under the MIT license. It addresses the pain point where useful Sessions are scattered across different Workspaces, making them difficult to find again. The plugin provides a global cross-Workspace favorites center. Users can favorite a Session with one click in the Session title bar, access the favorites center from the bottom of the sidebar or the collapsed track, browse all favorites with a priority on the newest, reopen still-active Sessions, and explicitly remove expired entries. The title, Workspace, running, and archived status of favorite entries are rendered from real-time DSH snapshots, and synchronization across multiple tabs is automatic via the `storage` event. Installation requires placing the plugin...
Read Moredynamic-schulte-dsh: A Schulte Grid plugin for the dsh Web UI to play while waiting for the model
`dynamic-schulte-dsh` is a waiting-time mini-game plugin for DSH, developed by marvin9551. It allows users to play the Schulte Grid within the Web UI while waiting for the model to respond during a DeepSeek Harness intelligent conversation, making use of fragmented time ranging from a few seconds to dozens of seconds. Adhering to the DSH philosophy of "Everything is a plugin," it is installed via `dsh plugin add`, registers a "Schulte" entry at the bottom of the Web UI sidebar, and features a breathing green dot indicator when the model is thinking. Core features include a classic n×n static grid (3×3 to 7×7), a rotating dynamic roulette wheel, waiting prompts, localStorage-based local persistence for history and settings, and optional number highlighting and color distraction enhancements. The architecture is split into two halves: the host part is a purely placeholder plugin.
Read MoreDSH Claude Style Reasoning Slider: Give DSH a New Animated Reasoning Level Slider and Model Selector
# Introduction In DSH (DeepSeek Harness), switching models and reasoning intensity relies on the native model selector within the session input area. In practice, there are two issues: first, different vendors have inconsistent naming conventions for levels, requiring users to memorize mappings for terms like `off`, `med`, `extreme`, and `ultracode`; second, custom routers (e.g., OpenRouter) often lack `reasoningEfforts` configurations, leaving the interface without an entry point for adjusting levels. DSH's philosophy is "everything is a plugin," meaning that the model slot in the session input can also be taken over by a plugin. The DSH Claude Style Reasoning Slider introduced in this article does exactly that:
Read More@max-null/dsh-chat-rail: Canvas-style message navigation bar for DSH sessions
`@max-null/dsh-chat-rail` (v0.6.1, MIT) is a DSH plugin maintained by Max-Null, positioned as a "scrolling-style message navigation bar" to address the issue of repeatedly scrolling through long sessions to find historical messages. It places a vertical navigation bar on the right side of the conversation panel, featuring an indicator dot for each user message. It smoothly expands the full list on hover, and clicking allows jumping to the specific message. Key features include: `scroll-spy` to highlight the reading position and automatically center it; synchronization with `dsh-better-sidebar` via CSS variables to avoid overlap; deep history navigation using the official precise loader on kernels 0.1.2-alpha.3+, with an automatic fallback to page-by-page loading on older kernels; a tooltip preview panel that pops up when hovering over a message item; and support for starring/favoriting (stored in localStorage).
Read Morerejection-check: A DeepSeek Harness-based plugin for checking bid rejection items
# Preface Anyone who has drafted a bid document knows this: no matter how excellent the content is, triggering even a single disqualification clause in the bidding document renders all previous efforts futile. A common self-audit method involves compiling "invalid bids" and "disqualification clauses" from the bidding document into a checklist, then manually searching through hundreds of pages of the bid document to locate corresponding content, while also checking for typos and inconsistencies. This process is time-consuming and highly susceptible to human error. The philosophy behind DeepSeek Harness (DSH) is "Everything is a plugin": transforming a fixed workflow for a specific task into a bundle that, once installed, registers as a skill, enabling the model to execute the workflow within Harness. The rejection-check plugin introduced in this article exemplifies this concept: by uploading both the bidding document and the bid document, it is responsible for extracting disqualification clauses...
Read Moredsh-fish-tts: Connecting Fish Audio TTS to DSH Web GUI
dsh-fish-tts is an open-source community plugin maintained by MaRi23333 that integrates Fish Audio Text-to-Speech into the DeepSeek Harness Web GUI. Users can listen to assistant replies using their own voice (including cloned voices) directly, without needing to copy text or write separate scripts. Features include a read-aloud button next to each message (supporting interruption, switching playback, and canceling), an auto-read toggle for the input field, and a settings page where users can configure the model, voice reference_id, and API Key (stored with AES-256-GCM encryption).
Read More@max-null/dsh-draft-polish: One-click draft polishing next to the send button
# Preface When collaborating with agents in DSH (DeepSeek Harness), the instructions or messages written for the model often become incoherent during the drafting process. The common approach is to open a separate session, paste the draft for the model to rewrite, and then copy it back—a process that involves many steps of switching back and forth. `@max-null/dsh-draft-polish` integrates this into the input box: there is a ✨ button to the left of the Send button. Clicking it invokes an LLM to polish the current draft, directly filling the result back into the input box. Polishing does not create a new session or save to history, and remains fully functional even while the session is generating output. Below, the plugin is introduced in the order of features, installation, usage, and settings. # What is it? One-sentence positioning: Tailored for DeepSeek Harness.
Read Moredsh-opg-usage: View OpenCode Go plan quota in real-time within the DSH Interface
## Introduction When running tasks using the OpenCode Go plan, checking remaining quota and reset times typically requires leaving the current interface to look elsewhere. Checking once is insufficient; you also need to remember the time of your last check to estimate consumption, or you risk accidentally hitting the limit. dsh-opg-usage solves this problem by integrating the quota panel directly into the DSH interface and maintaining a persistent floating capsule for instant checks. Below, we introduce its features, installation, and usage. ## What is it? dsh-opg-usage is a DSH Web plugin maintained by M4sh1r0-444 and licensed under the MIT License. It is a draggable, collapsible floating capsule that displays the usage percentage of the OpenCode Go plan across three quota windows: rolling window, this week, and this month.
Read Morejune-agent: June Dual-Personality Agent and Supporting Skills, installable as a DSH plugin
june-agent is an npm package maintained by lemonorangeapple that packages two fixed-personality Agents and two skills into a cross-hosting solution. June-M features a "catgirl" personality (temperature 0.1, obedient and concise), while June-S features a "dominant" personality (temperature 0.3, disciplinary pushback). The skills include Git backup strategies (git-backup) and design coding standards (june-style). The package provides both Claude Code format (agents/ and skills/ directories) and DeepSeek Harness format (presets/ directory), DSH.
Read Moredsh-themes: Apply a complete theme to DSH runtime
# Introduction When using DSH (DeepSeek Harness) for daily development, the web interface is the part you stare at the most. It defaults to only one look. If you want to switch to dark mode, follow the system theme, or migrate your VS Code theme that you've used for years, you usually have to manually edit styles, which are lost after a restart. The philosophy of DSH is "everything is a plugin," so appearance customization can also be delegated to plugins. The dsh-themes plugin introduced below is just such a solution: it features a built-in color palette, light/dark/system-following modes, Open VSX search and installation, VS Code theme import, a persistent theme library, and restoration after restart. # What is dsh-themes? dsh-themes is an appearance and theme plugin for the DSH runtime, created by Mang...
Read MoreHarness Insights: A Local-First Usage Insights Panel for DeepSeek Harness
# Introduction If you use DSH (DeepSeek Harness) to run agent tasks on a daily basis, you have likely encountered the following problem: token consumption is scattered across various sessions. To understand which model is used most frequently over a period of time, the cache hit rate, or which tool calls are most frequent, you have to manually go through session logs to perform statistics. Harness Insights solves exactly this problem. It adds a local-first "Usage Insights" page under Harness settings, aggregates and displays structured metadata from local sessions without modifying the upstream Harness source code. The following sections introduce its positioning, features, installation method, and data scope. # What is Harness Insights
Read More