dsh-plugin-bilibili: Provides Bilibili search, subtitles, and frame-level viewing capabilities for DeepSeek Harness
# Introduction When integrating information sources for agents, Bilibili is an unavoidable category: finding videos on specific topics, summarizing episodes, and verifying the actual content. The challenge lies in the fact that Bilibili pages are highly dynamic, making direct web scraping unreliable. Providing only the subtitle text to the model is insufficient; when subtitles state "as shown in the figure," the model lacks context on the visual content. The philosophy behind DeepSeek Harness (DSH) is "everything is a plugin," where external capabilities are integrated into the session as tools. dsh-plugin-bilibili implements this approach: upon installation, the agent gains five `bilibili_*` tools covering keyword search, video metadata, subtitle transcripts, direct playback links, and video frames. Crucially, frames are returned as image chunks, enabling models with image capabilities to view the actual visual content directly, rather than relying solely on subtitles.
Read Moredsh-visualizer: Directly render charts and widgets in the conversational flow without modifying DSH source code.
dsh-visualizer is an external plugin (v0.3.0, MIT) for DSH (deepseek-harness) maintained by Moses14159. It allows models to render visual content directly within the conversational flow without modifying DSH source code by reusing existing sessions within DSH.
Read Moredsh-liangwengu: DeepSeek API Peak/Idle Time Reminder Plugin (DSH)
# Introduction DeepSeek API pricing is tiered by time zones: 9:00–12:00 and 14:00–18:00 (Beijing Time) are peak hours, while the rest are off-peak hours (half price). It is easy to forget to check the time when busy with tasks. I ran tasks during peak hours all afternoon before realizing it; this kind of thing is not uncommon for API users. Existing solutions involve setting alarms or running cron scripts in the terminal, but they do not follow your working interface. dsh-liangwengu has developed a DSH (DeepSeek Harness) plugin: it pops up a cartoon modal, sends system notifications, and plays sounds to directly tell you "Peak is here, run; Peak is gone, start saving." The following sections introduce its notification mechanism, UI capabilities, and installation method. # What is this
Read Moredsh-plugin-rankings: Automatically organize DSH plugin ecosystem Top 10 / Top 100 rankings daily
The philosophy of DSH is "everything is a plugin," and the number of plugins in the dsh-plugin ecosystem is continuously growing. Selecting a suitable plugin usually requires manually searching GitHub, browsing community aggregation directories, and comparing npm download counts. Since this information is scattered across multiple channels, there is no direct ranking list available to show which plugins are trending or have high download volumes. The `dsh-plugin-rankings` introduced below is designed to address this issue: it curates Top 10 and Top 100 rankings for the dsh-plugin ecosystem on a daily basis. It supports sorting by popularity (GitHub Stars) or download volume (npm weekly downloads) in descending order, with popularity being the default. The interface is integrated directly into the DSH Web settings page.
Read Moredsh-plugin-github: Equip DSH Agents with GitHub Retrieval Capability
# Introduction When building agents with DeepSeek Harness (DSH), you often encounter requirements such as querying repository metadata, searching for issues in a repo, or reading source code files. The philosophy behind DSH is "everything is a plugin," meaning such network retrieval capabilities do not need to be integrated from scratch, as ready-made implementations already exist in the community. The `dsh-plugin-github` introduced below is one such example: after installation, the agent gains two tools that can directly search GitHub repositories and issues, and fully read a single resource. # What is it `dsh-plugin-github` is a DSH plugin maintained by moxingovo under the MIT license. It solves the problem of enabling agents to interact via tool calls to...
Read Moredsh-web-search-free: Free Web Search Plugin for dsh with Multi-Engine Fallback
dsh-web-search-free is a free web search and crawling plugin for dsh maintained by MochiNek0 (v1.3.0) designed to replace the official `deepseek-official` search channel. The official channel triggers full Messages model calls for every search, consuming double tokens and mandating `DEEPSEEK_API_KEY`. This plugin enables the host process to directly call each engine's dedicated retrieval endpoints, costing 0 model tokens for retrieval and using each engine's own quota for billing. It is installed as a bundle layer and automatically falls back to the default channel after uninstallation and restart. It supports 8 engines (TinyFish, AnySearch, Exa, Tavily, Firecrawl, Brave, SerpApi, Jina).
Read MoreDeepSeek Harness Hub: Wrapping dsh Web UI into a Native Desktop Window
# Introduction When using `dsh`, the Web UI typically runs in browser tabs: windows are mixed among a cluster of tabs, lacking a distinct taskbar identity; switching tabs risks missing the timing of task completion; parallel multi-sessions are merely makeshift via browser tabs; and title bars and icons cannot follow `dsh`'s theme changes. For developers using `dsh` as their primary interface, these are daily frictions. `dsh`'s philosophy is "everything is a plugin," and the desktop-ification of the tool is also addressed by the plugin layer—this is the DeepSeek Harness Hub (dsh-hub) introduced below. It uses Tauri 2.x native windows to host the `dsh` Web UI, filling in desktop capabilities such as the system tray, notifications, session tabs, interactive terminal, and window memory. The project itself also...
Read Moremmx-quota-tool: Directly read MiniMax token quota in DSH web
mmx-quota-tool is a DSH web plugin maintained by mtty-ai (v0.1.0, MIT License) that displays the usage percentage of MiniMax token packages within the session input area. Clicking expands a detail panel to view per-model 5-hour and weekly usage, reset countdown, and absolute values. Icons change color based on usage levels (<50% green, <80% yellow, >=80% red) and are automatically hidden for non-MiniMax models. The plugin supports Shift+click to force refresh and a zh/en bilingual interface. It consists of a Host component (polls upstream every 60 seconds) and a Client component (polls the host every 5 seconds and mounts the UI). The installation command is `dsh plugin --profile web add gi...
Read Moredsh-plugin-model-switcher: Switch models and reasoning levels in DSH using keyboard shortcuts
# Introduction During agent development, frequent comparison between different models is often required: running a task with the current model and then switching models to compare results, or adjusting reasoning levels based on task complexity. Manually opening the model selector, selecting a model, and adjusting levels repeatedly is quite tedious. DSH's philosophy is "everything is a plugin," making high-frequency, small operations ideal for plugins. The `dsh-plugin-model-switcher` introduced below is exactly that: it adds two sets of keyboard shortcuts to the DSH web client—one for switching models and one for switching reasoning levels—accompanied by toast notifications to confirm each action. # What is this `dsh-plugin-model-switcher` by Mrlilil
Read Moredsh-model-memory: Supplementing DSH with persistent memory for custom model reasoning levels and channel preferences
dsh-model-memory is a DSH plugin developed by Mutx163, designed to address the issue of missing thinking level configurations for custom API models and the loss of model and thinking strength preferences when switching channels. It mounts a thinking level configuration bar for custom models within the official "Settings -> Models" section, supports low/medium/high/max levels, and allows for hot-reloading of configurations. Additionally, it remembers the last selected model and thinking intensity across sessions per channel, ensuring preferences are preserved when switching channels.
Read Moredsh-inherit: One-click import MCP servers and Skills from Claude Code, Cursor, Codex, and cc-switch to DeepSeek Harn
dsh-inherit is a plugin (version 0.1.0, Developer preview, MIT license, unofficial) developed by community developer MayBeTheWorld for DeepSeek Harness (DSH). It enables one-click importing of MCP servers and Skills scattered across Claude Code, Cursor, Codex, and cc-switch into DSH.
Read MoreDSH Skill Mover: One-click transfer of skills from other Agents to DeepSeek Harness
**Introduction:** It is becoming increasingly common to use multiple agents on a single machine: Claude Code, Codex, Cursor, and Kimi each maintain their own skill directories, located at `~/.claude/skills`, `~/.codex/skills`, and `~/.cursor/skills` respectively, which are not interoperable. When you wish to switch to or use DeepSeek Harness (DSH) alongside them, your existing skills must be either reinstalled or manually copied—requiring you to handle directory structures, naming conflicts, and duplicates yourself. The DSH Skill Mover introduced below is designed for this specific purpose: it automatically scans the skills installed by various agents on the machine, allows you to select them, and moves them into DSH.
Read Moredsh-subagent-codex: Delegating tasks to local Codex CLI in DeepSeek Harness
# Introduction The philosophy of DSH is "everything is a plugin," allowing the main agent to delegate tasks to sub-agents via a subagent mechanism. However, standard subagents run within DSH and share the same environment as the main conversation. If you want to delegate a specific task to OpenAI Codex—using Codex's own model and tool chain within an isolated process—you must manually switch terminals, execute commands, and paste the results back into the session, which is a disjointed workflow. dsh-subagent-codex fills this gap. The following sections introduce its positioning, implementation, and usage. # What is this dsh-subagent-codex (current version 0.1.1) is maintained by mjylfz and uses MI...
Read Moredsh-llm-finish-reason-tolerance: Properly terminating streaming responses missing the finish_reason
# Introduction When connecting OpenAI-compatible models to agents, there is a subtle yet critical compatibility issue: the streaming response from some gateways only pushes content deltas and end markers, without ever sending the `finish_reason` field. DSH's model client, pi-ai, classifies this stream as truncated ("Stream ended without finish_reason"), which the host then maps to a `TRANSPORT` error finish. The result is that while the content of every request is delivered intact, the turn ends in failure. A typical example is the Snowflake Cortex REST gateway (`*.snowflakecomputing.com/api/`)
Read Moredsh-token-stats: Statistics on Token consumption in DeepSeek Harness Web UI
# Introduction When running sessions with DeepSeek Harness (DSH), token usage is scattered across individual session logs. To determine how much each model consumed this month or which days were used most frequently, one must manually review the logs or write scripts to perform the statistics. This is time-consuming and prone to error. `dsh-token-stats` is a token usage statistics plugin built for the DSH Web UI. It utilizes session logs as its data source, aggregates usage into views by day and by model, and displays the results directly in the settings panel. Below is an introduction to its features, installation method, and working principle. # What is this? `dsh-token-stats` is maintained by MoonlitDropOfBlood and follows the MIT License. The README states:
Read More@mojiexuan/dsh-git: Add a VS Code-style Git panel to the DeepSeek Harness web UI
@mojiexuan/dsh-git is a VS Code-like Git source code management plugin designed for the DeepSeek Harness web UI (MIT License, version 0.2.1), mounting a panel in the sidebar to enable staging, committing, pushing, publishing branches, pulling, and branching.
Read Moredsh-archive-manager: Retrieve archived sessions from DSH
## Introduction In DeepSeek Harness (DSH), once a session is archived, it disappears from the sidebar. While log files remain local, the Web UI does not provide an entry point to browse archives; retrieving them relies on remembering titles or navigating through directories. dsh-archive-manager addresses this issue by rendering archived sessions visible and manageable—grouping them by project, supporting search, and enabling one-click restoration or permanent deletion for each archive. ## What is it? dsh-archive-manager is a community plugin maintained by MoonlitDropOfBlood, built on DeepSeek Harness. It is not an official DeepSeek product and is licensed under MIT. In a nutshell:
Read Moredsh-patchouli: Local Memory and Knowledge Hub for DeepSeek Harness
# Introduction When developing agents with DeepSeek Harness (hereinafter referred to as DSH), the "everything is a plugin" philosophy implies many choices for memory capabilities. Hindsight, MemOS, and Graph Memory, listed in the compatibility table, all belong to this category. The problem is that each memory plugin comes with its own storage and retrieval algorithms, and their integration methods are different. To use them together in the same environment, one must adapt interfaces individually and maintain data separately; switching implementations requires modifying the integration layer. `dsh-patchouli` addresses precisely this issue: it decouples "data integration" from "algorithm implementation" by placing a unified memory and knowledge service in the middle. Below is an introduction to this plugin. # What is this? The positioning of memorax-ai/dsh-patchouli is Deep...
Read Moredsh-playwright-native: Registering local native Playwright CLI as a DeepSeek Harness pass-through tool
# Introduction Integrating Playwright with an Agent typically involves a semantic wrapper: splitting `test`, `install`, and `show-report` into multiple tools and designing a parameter schema for each subcommand. The problem with this path is that the Playwright CLI already possesses a complete feature set. No matter how comprehensive the mapping in the wrapper layer, options are inevitably missed, and the range of capabilities actually available to the Agent depends on the wrapper author. In DSH (DeepSeek Harness), "everything is a plugin"; all tools visible to the model originate from plugin registration. Thus, an alternative approach emerged: register a pass-through tool that passes the parameters provided by the Agent directly to the local CLI. `mitao-su/`
Read Moredsh-blue-archive-shiroko: Blue Archive Theme and Shiroko Desktop Pet for DSH Web Client
`dsh-blue-archive-shiroko` is an unofficial fan-made theme plugin for the DSH Web Client developed by mldhao (v0.6.0, MIT License), providing a Blue Archive style visual and interactive experience for DeepSeek Harness. Core features include: a complete UI theme centered around academy blue, cyan highlights, and white frosted glass panels; a Shiroko desk pet that supports dragging, resizing, and autonomous interaction; petting interaction and sound effects triggered by holding the head for approximately 420ms; and Codex-style short bubbles with streaming character-by-character text display. In terms of utility, a short press on the desk pet allows querying the API balance via a Host-side RPC call to the DeepSeek official interface. The API key does not enter the browser, and the RPC is restricted to loopback address calls, ensuring security. Installation requires the DSH plugin system.
Read Moredsh-msg-edit: Adding Message Editing and Regeneration to DeepSeek Harness Web
dsh-msg-edit is a DSH Web plugin maintained by @moton16 (MIT license, version 0.1.0) that adds "Edit this response" and "Regenerate response" buttons to assistant messages in DeepSeek Harness Web. The plugin consists of two parts: the host side provides four HTTP RPCs (rewrite-message to rewrite a single response, edit-message to edit user messages and regenerate, regenerate to regenerate responses, get-message to read text for prefill), while the client side declaratively injects buttons and inline editors into the assistant message action bar. The entire package is a pure declarative Cordis bundle with no build steps. There are three installation methods: dsh plug
Read Moredsh-vision-adapter: Adding drag-and-drop image recognition capability to DeepSeek Harness
dsh-vision-adapter is a DSH-exclusive plugin maintained by motongv (MIT License) that solves the issue where DSH text models cannot recognize images. The chat interface remains unchanged; images are first converted into text descriptions by the vision model, which is then submitted to DeepSeek for reasoning and response. The plugin acts as an in-process HTTP bridge within DSH to register vision.
Read Moredsh-sound-lab: Configure event sound effects and character desktop pets for DSH Web GUI
dsh-sound-lab is a DSH plugin maintained by miiaowuwu (MIT License) that solves the issue of losing focus on the window or missing events when running DSH sessions. It provides three functions via the Web GUI: independent sound effect configuration for four trigger conditions—session end, option popups, permission requests, and stopping—where popups and permission requests are considered "attention events" for priority alerts. It supports uploading and managing a local sound library (supporting multiple audio formats), including three built-in sound effects that cannot be deleted and are only soft-hidden. Additionally, it integrates with Alibaba Cloud Bailian to generate voice using cloned voices with one-click import (at the user's own expense). On the desktop side, it defaults to an animated desktop pet of Angelina from Arknights, reacting to events with celebrations, confusion, sleepiness, etc., and can also be switched to a simple circular ball.
Read MoreDeepSeek-Harness-Hanako-Memory: Manage dsh assistant presets via a web page, featuring a built-in Hana-style memory system
moononnn/DeepSeek-Harness-Hanako-Memory is a plugin for the DeepSeek Harness (dsh) ecosystem (npm package name: dsh-assistant-manager, version 0.4.1, MIT License). It integrates the memory system of Hana (openhanako) with assistant configuration into dsh, providing a web interface to manage assistant presets and eliminating the hassle of manual YAML editing. Core features include selecting four archetypes (Hanako/Butter/Ming/Kong) when creating a new assistant, which automatically generates template identities and personalities and saves them as dsh presets. It supports full-field editing, avatar cropping and uploading, deletion protection, setting as the main assistant, and sorting. The memory section provides a master switch, add/delete pinned memories, and management for today/week/lifetime.
Read Moredsh-github-connect: One-click GitHub connection at the bottom left of the DSH dialog
dsh-github-connect is a DSH plugin (MIT licensed) developed by Moon-shiyue that enables the AI to directly interact with the user's GitHub. The connection method uses the GitHub OAuth device flow or pasting a PAT; it does not require a self-hosted callback server, and tokens are not relayed through third parties, only stored locally in `.github-auth.json`. Once connected, the AI can use the `github_api` tool to access...
Read More