dsh-cache-stabilizer: Stabilize DeepSeek Harness cache prefix
`dsh-cache-stabilizer` is an MIT-licensed plugin for DeepSeek Harness (DSH) designed to stabilize system prompt prefixes to improve the reuse rate of Provider-side Prompt Cache. Its core functions include: first, moving the working directory (cwd) from the system prompt to the runtime-context snapshot in DSH standard or headless coding personas, ensuring that different projects share the same prefix while retaining the correct cwd; and second, normalizing the object-key order of tool schemas to facilitate prefix stabilization. This plugin does not create a new cache layer or alter the Provider's caching strategy. Additionally, it provides `
Read MoreOLEDCare: OLED Burn-in Prevention Plugin for DeepSeek Harness
OLEDCare is an open-source plugin designed for the DeepSeek Harness (DSH) Web GUI, aimed at mitigating OLED screen burn-in caused by static dark backgrounds and bright text during long intelligent agent conversations. By dimming interface pixel brightness, this plugin safeguards the screen without disrupting the user's existing workflow. Its core features include the Nap mode, which offers a true-black full-screen screensaver; setting the dark theme background to pure black; reducing the brightness of static borders; dimming text and light-colored elements based on Gamma perception; and applying stepped dimming according to idle or focus states. Additionally, it supports hue rotation drift over time and allows users to customize various parameters. Settings are persisted via browser `localStorage`, falling back to in-memory storage if unavailable. Installation requires Node.js.
Read Moreseekmaid-pet: Auto-start Windows Desktop Pet for DSH
seekmaid-pet is a Windows 11 desktop pet plugin within the DSH ecosystem, designed to visualize the status of long-running DSH sessions. It supports bubble notifications for task initiation, completion, new messages, and tool calls, and allows users to bidirectionally communicate with the current DSH session. As an independent site project, it functions as either a native desktop pet resident in the system tray or a DSH plugin that automatically launches or closes alongside DSH. Key features include operation without a console window, single-instance restriction, auto-start support, and zero-configuration automatic detection and initialization. Installation requires a Windows environment with Python 3.10+, PySide6, and Node.js, and involves installing the plugin via a local directory or GitHub link. Initialization requires running setup_windows.bat to create...
Read Moredsh-theme-cyberpunk: DeepSeek Harness Web Cyberpunk Theme and Local Music HUD
dsh-theme-cyberpunk is a cyberpunk-themed plugin for the DeepSeek Harness (dsh) Web interface, designed to optimize the local user experience. It creates a neon atmosphere through visual effects such as a rotating RGB color wheel, neon glows, and CRT scanlines. Additionally, it supports a built-in or custom dark background image library, with user selections being persisted. The core highlights of the plugin include a draggable and collapsible HUD interface in the top-right corner, integrated with a one-click toggle for switching between "Cyber" and "Default" appearances, as well as a fully functional local music player. The player supports adding audio files such as mp3, flac, and wav, featuring play/pause, random order, various loop modes, progress seeking, and volume adjustment. Both the playlist and playback state are persisted, though playback needs to be manually resumed after refreshing the page. Installation requires Node.js 22+.
Read Moreinterest-memory: Add cross-session long-term memory to DeepSeek Harness
`interest-memory` is a local long-term memory service maintained by djasdh, designed to address context discontinuity between DeepSeek Harness (DSH) sessions. Built with Go, it utilizes SQLite-vec vector indexing for local storage and retrieval, supporting fully offline operation. The service integrates with DSH via the Cordis plugin `@djasdh/interest-memory-dsh-bridge`, automatically recalling relevant memories at the start of a session and pushing conversation logs at the end. Core features include evidence-based memory storage, a three-stage verification process (check/claim/conflict), automatic merging of semantically similar entries, replacement marking for old entries, and graph-based query visualization. The system provides a REST API to facilitate memory sharing among multi-agent frameworks.
Read Moredsh-usage-cost: Calculate DeepSeek API cost per session per round
`dsh-usage-cost` is a DeepSeek Harness plugin designed to calculate API costs in real-time on the Web, round by round, per session. By projecting `usageTimeline` on the server side, the plugin records the token usage (including misses, outputs, and cache read/writes) and timestamps for each assistant turn. It provides cost readings across four dimensions in the chat interface: Conversation, Workspace, Today, and Last 5 Hours. Its core advantage is the automatic adaptation to the peak and valley pricing logic implemented since August 17th, which switches rates based on timestamps; if historical sessions lack timeline data, it falls back to flat-rate calculation. The plugin is distributed as an official bundle, requiring no additional building when installed from the Git repository—users simply need to execute the install command, restart the Web app, and refresh the page. Users can customize settings by editing `lib/client.js`.
Read Moredsh-habits: Maintain user-global AGENTS.md in Web Settings
dsh-habits is a DeepSeek Harness (DSH) plugin designed to simplify the maintenance of the user's global instruction document `$DSH_HOME/AGENTS.md` through a web settings interface. The plugin provides a "My Habits" editor in the General area of DSH Web Settings, allowing users to directly edit and save preference or constraint text in their browser. Once saved, the text is automatically written to the AGENTS.md file and injected into subsequent sessions by DSH's built-in instructions loader. The plugin implements data read/write via REST interfaces `/api2/habits/describe` and `/api2/habits/update`, and adopts S...
Read Moredsh-airdrop: Drag files into DSH, AI reads them
dsh-airdrop is a web-based plugin for DSH designed to resolve file upload challenges. Users simply need to drag files or folders into the DSH window; attachments will be sent along with messages, and the AI will automatically read the content. The plugin supports attachments in any format without type restrictions and provides in-browser preview functionality. Files are stored only in a workspace isolated to the current session, ensuring they do not interfere with each other. For large files, the plugin supports chunked upload, resumable uploads, and automatic retries for gateway errors, with a single file limit of 20 GiB. Installation requires Node.js >= 22. You can enable it by adding the `dsh-universal-attachments` package via the `dsh plugin` command and restarting the service. For security reasons, it is recommended to access remotely via an SSH tunnel to avoid exposing the service directly to the public internet. Note that empty directories may not be preserved during use.
Read MoreDSH Personal Workbench Plugin: @dely0/dsh-personal-workbench
`@dely0/dsh-personal-workbench` is a personal workbench plugin based on DSH Web, designed to integrate a calendar, a tree-structured task list, and an AI assistant, making it suitable for local single-user scenarios. Data is stored entirely locally in the `~/.dsh/workbench` directory, and the API is restricted to loopback access to ensure security and eliminate cloud sync dependencies. Core features include natural language task entry, AI clarification and breakdown, post-execution user verification, AI review, smart sorting, daily and weekly report generation, as well as the accumulation of a personal knowledge base and error collection. The plugin supports multiple installation methods, including via the `dsh` command, npm, or directly mounting the source code. After installation, the service must be restarted and the page hard refreshed. Users must pre-configure DSH model credentials, as AI operations will actually consume tokens. This plugin is designed for single-user local usage only.
Read MoreMaestro Skills: A Unified Plugin Packaged for DSH and Multi-class AI Coding Agents
`ddtcorex/maestro-skills` is a unified skills repository that packs 31 skills into a general-purpose plugin compatible with multiple platforms such as DeepSeek Harness (DSH), Claude Code, and Codex CLI. The plugin includes 17 domain-specific skills (covering Magento 2, Laravel, Symfony, WordPress, and general PHP, based on the Govard environment) and 14 process skills (derived from `obra/superpowers` v6.3.0, covering brainstorming, TDD, debugging, etc., and licensed under the MIT license). In DSH, users can add it via the command `dsh plugin --profile web add github:d...
Read Moredsh-vision-skill: Enabling DSH Pure Text Models with Image Recognition, OCR, and Object Localization
dsh-vision-skill is a standard vision plugin within the DSH ecosystem, developed by DDDFXYqiming and licensed under the MIT license. It aims to address the limitation that pure text models cannot directly process images. By registering a native `vision` runtime skill, the plugin enables text models to invoke local tools to perform functions such as image recognition, OCR, object localization, and dominant color analysis. Its core tools include: `vision_analyze` (supports 6 modes such as general, code, and evidence), `vision_ocr` (text extraction preserving layout), `vision_ground` (object localization returning coordinates), `vision_detect` (enumerating elements like buttons), and `vision_dominant_colors` (local computation).
Read Moredsh-layered-memory: DSH Cross-Session Long-Term Memory Plugin
dsh-layered-memory is a cross-session long-term memory plugin for DeepSeek Harness (DSH), licensed under the MIT License, designed to address the issue of context loss following a session. The plugin persistently stores verified facts, task experiences, and environmental information on disk, supporting cross-session retrieval. Core features include L1-L3 layered memory, namespace isolation, BM25 full-text search, content deduplication, and cross-namespace experience enhancement. The plugin provides 14 memory tools, such as `memory_write` (mandatory evidence verification), `memory_search`, and `memory_index`, and supports progressive tool exposure. Consistency maintenance relies on deduplication, manual updates, and candidate merging; the system does not automatically detect contradictions, and it is strictly prohibited to physically discard verified facts, allowing only compression.
Read Moredsh-annotation-patched: Cite assistant messages and respond point-by-point
dsh-annotation-patched is a browser-side enhancement plugin for DSH Web, maintained by DDDFXYqiming, and licensed under the MIT License. The plugin aims to address issues of context loss and unclear positioning that occur when manually copying and pasting assistant replies for follow-up questions. Its core functionality allows users to select text from assistant messages, generate quote blocks by clicking the "Quote" button, and send them together with their message. Upon receiving the message, the model responds item-by-item with numbering and includes expandable quote markers. The plugin utilizes an explicit single-button mechanism, supports the automatic cleanup of old quote blocks, and synchronizes the position of quote tags during multi-line input. Furthermore, it offers a local patch replay tool to facilitate replaying patches based on a manifest when upgrading the base, thereby generating new client files. Installation can be performed using the `dsh plugin` command from GitHub or a local directory.
Read Moredsh-custom-logo: Replace favicon and sidebar logo for DSH Web
`dsh-custom-logo` is a lightweight plugin designed specifically for the DSH web platform, aiming to simplify the deployment of custom branding. Adhering to the "everything is a plugin" philosophy, this plugin allows users to upload just a single square image and a wide horizontal image to automatically replace the browser favicon and sidebar logo, without the need to manually prepare multiple resized assets or modify frontend resources. Upon installation, a "Brand Settings" section is added to the DSH settings, supporting both Chinese and English. Image processing is performed in the browser; changes take effect immediately after upload with an automatic page refresh, without requiring a service restart. Generated image files are written atomically to the `branding` folder within the user directory. Automatic cache clearing is implemented via dynamic CSS injection and URL versioning parameters to ensure custom configurations are preserved during upgrades. The installation command is `dsh plugin...
Read Moredsh-image-pathify: Enabling DSH models without image capability to process chat images
dsh-image-pathify is an MIT-licensed plugin for DeepSeek Harness (DSH) Web, designed to address the need for models that cannot directly process images (such as deepseek-v4) to understand images in chats. By replacing images with local file paths before sending a message, the plugin prompts the model to invoke the built-in `analyze_image` tool to obtain textual descriptions using the configured OpenAI-compatible vision API. It supports passing multiple images at once and maintains original behavior for models that already support image recognition without exposing the relevant tools. Installation is simple: run `dsh plugin add dsh-image-pathify`, start the Web interface, and then configure the API key, vision model, and interface address in the settings page.
Read Moredsh-my-go: DSH-based Multi-agent Orchestration Plugin for Sisyphus
dsh-my-go is an agent orchestration system built on top of DeepSeek Harness (DSH) and adopts the MIT license. The system unifies the management of 7 professional sub-agents (such as Hermes and Oracle) through the Sisyphus scheduler, adopting a star topology structure to ensure that only one sub-agent runs at any given time. Its core capabilities include flexibly configuring the provider, model, and reasoningEffort based on job types, supporting off-peak routing based on Beijing Time to optimize resource usage, and providing communication tools such as go_work as well as the Sisyphus quality control mechanism. Model configurations are verified before dispatch, and if they do not exist, they fall back to parent session routing. The system provides a WebUI settings page and a tree diagram panel for convenient visual management. Installation requires...
Read Moredsh-web-search-searxng: Connect DSH's web search to self-hosted SearXNG
dsh-web-search-searxng is an independent plugin package for the DeepSeek Harness (DSH) web profile, designed to allow users to switch the backend of the `web_search` function to a self-hosted SearXNG instance, thereby eliminating dependency on external APIs. Built under the MIT license and developed using pure ESM modules, this plugin requires no build steps and can be installed and automatically activated via the simple `dsh plugin` command. In terms of configuration, it supports settings via environment variables `SEARXNG_BASE_URL` (default localhost:8888), `SEARXNG_NUM_RESULTS`, and the optional `SEARXNG_API_KEY`, and can also be configured directly...
Read Moredsh-plugin-md-outline: DSH Markdown Outline and Structure Check Plugin
dsh-plugin-md-outline is a DeepSeek Harness (DSH) plugin maintained by d-ouyang. Written in pure ESM JavaScript and released under the MIT license, it requires no build steps. It aims to automate the resolution of structural issues in long Markdown documents by adding the `md_outline` tool, enabling the automatic output of the document title tree and structural checks. Core features include outputting an H1-H6 nested title tree with line numbers, as well as detecting heading level jumps, duplicate headings, H1 uniqueness issues, and unclosed code fences. The plugin supports three modes: `outline`, `lint`, and `both`. It allows limiting the outline depth via `maxDepth` and supports recursive scanning of directories.
Read Morecwbcheng/dsh-knowledge-graph: Split document body, images, or conversation logs into a knowledge graph with bidirectional positioning.
`dsh-knowledge-graph` is a DSH Cordis plugin designed to transform section documents, images, or AI conversation history into structured knowledge graphs. It supports asynchronous splitting of body text (up to approximately 1 million characters) and images (1-4), utilizing multimodal models to extract visual semantics. The plugin generates 8 types of nodes and 12 types of relationships, providing bidirectional linking between the graph and the original text, as well as evidence verification and remediation functions. Users can question and correct content through quick health checks, AI auditing, and external fact-checking, ensuring every conclusion is bound to verifiable evidence citations. Furthermore, it supports incremental appending, chapter filtering, structured retrieval, and evidence-based Q&A, and allows export to PNG, JSON, or CSV formats. Installation requires cloning the repository and instructing the Agent to run the code within a DSH Web session. This plugin is pure JavaScript.
Read MoreIntroduction to the DSH Command Optimization Plugin dsh-command-opt
`dsh-command-opt` is a slash command optimization plugin for DeepSeek Harness (DSH), designed to enhance the command usage experience on the Web. Its core features include intelligent command name completion and candidate rotation via Tab/Enter; providing parameter format guidance, ghost text hints, and a persistent popup; fixing the issue of an empty command directory in tool-opened sessions (subagents), supporting direct command execution and operations like archiving and copying via the header menu; and fixing command output anomalies and missing header buttons in empty conversations. The plugin uses the MIT license and requires Node.js 22 and the pnpm environment. It is installed to a local DSH source code checkout via the `link:` method and takes effect after building. It does not alter the underlying command definitions of DSH, but targets only...
Read MoreDSH ExpMem: Adds a Checkable Long-Term Memory Plugin for DeepSeek Harness
DSH ExpMem is an MIT-licensed community plugin for DeepSeek Harness, designed to address the issues of local storage and retrieval of long-term personal memory. It writes user habits, task experience, and engineering insights to local JSON files, preserves DSH Session as the raw word-for-word history layer, and does not duplicate Session events. The plugin adopts a layered memory architecture, supports prompting the Agent to retain persistent knowledge when context pressure is high, and performs retrieval sorting based on relevance, recency, and importance. Its core advantages include supporting auditable reflection (observations supported by citations, interruptible and resumable), maintaining a complete trust lifecycle from candidate to deprecated, and importing Markdown memories from Claude Code and Codex as candidate records with source evidence. The plugin runs lightweight, not...
Read Moredsh-ssh-remote: OpenSSH remote workspace plugin for DSH
dsh-ssh-remote is a DeepSeek Harness (DSH) plugin developed by CrazyShout, designed to address the needs for file, shell, and terminal operations on remote machines. Its core advantage lies in reusing the local OpenSSH configuration to avoid repetitive maintenance of authentication and proxy strategies, while providing a stable remote execution boundary for DSH. The plugin supports discovering specific hosts from the local `~/.ssh/config` and adding remote workspaces via a standard interface. It utilizes a versioned helper to handle remote file read/write (featuring security mechanisms such as atomic publishing and version review), shell execution (supporting bounded output and reconnection), and terminal operations, while achieving process and PTY isolation on Linux via bubblewrap.
Read Moredsh-deepseek-balance-widget: View AI balance in the dsh Web sidebar
dsh-deepseek-balance-widget is a dsh Web sidebar plugin maintained by crazy-L118, designed to centrally display the balances, daily consumption, and token usage of DeepSeek and MiMo, with an automatic refresh every 30 seconds. This plugin is based on the MIT license and requires Node.js >= 18. It is installed via the dsh plugin command (version 2.4.2 is recommended), and the dsh web service requires a complete restart after installation. Core features include a sidebar real-time entry and a details popup. The popup supports managing platforms, switching views, and adding or removing new platforms. DeepSeek users need to configure an API Key, and optionally configure a Platform Token to view detailed usage statistics and status during busy times.
Read Moredsh-plugin-xquant: Integrating xquant quantitative research MCP into DeepSeek Harness
dsh-plugin-xquant is a DSH plugin maintained by codingDuan (MIT licensed), designed to seamlessly integrate the xquant quantitative research MCP into DeepSeek Harness, resolving the pain point of cumbersome configuration for external tools in dsh Web. After installation, users can directly query factor definitions, calculate factor time series, browse strategy logic, and view historical performance within dsh Web, organizing the results into readable research conclusions. This plugin is positioned as a research assistance tool and does not provide buy/sell recommendations or market forecasts. The prerequisite for use is configuring `XQUANT_API_KEY` in the dsh process environment; the plugin reads exclusively from environment variables and does not write to local files or perform uploads. It connects to the xquant.shop official MCP service by default and supports via...
Read Moredsh-interview: AI Interview Practice Assistant in DeepSeek Harness Web
dsh-interview is a local-first AI interview practice assistant running within the DeepSeek Harness Web, adopting the MIT open-source license. It integrates interview practice into the dsh web workflow, supporting five modes: "Memorizing Common Interview Questions," "Mock Interview," "Resume Topic Prediction," "Scenario-based Questions," and "LeetCode Hot 100." The core workflow encompasses questioning, answering, AI scoring and feedback, knowledge explanation, and providing ready-to-express answers, while automatically saving data such as questions, answers, and summaries without requiring the deployment of an additional database. The plugin provides a visual workspace that supports searching, filtering, editing, and exporting practice records. Users require a Node.js 22.5+ environment and can install and enable the plugin via the dsh plugin command. This tool is suitable for job seekers in technical development roles to review theory, prepare for interviews, and practice algorithms.
Read More