DSH Custom Provider Reasoning: Completing Reasoning Levels for DSH Custom Providers
dsh-custom-provider-reasoning is a DSH host plugin designed to solve the problem where custom provider models cannot select thinking intensity in the interface due to missing reasoning metadata. This plugin automatically injects `reasoningEfforts` (with default levels ranging from off to max), fills in missing `maxTokens` and `contextWindow`, and intelligently manages route-level `compat` configurations under different protocols, ensuring that reasoning level selections are actually sent to the online service. During installation, you need to add the GitHub dependency in the web profile and run `pnpm install`, and the plugin will automatically write the configuration into `settings.yaml`. Users can use `cordis
Read Morechicheng-push: Adding Multi-Channel Message Push to DSH
chicheng-push is a DSH message push plugin designed to uniformly address the notification issue after plugin tasks are completed. This plugin adds a "Push Plugin" entry in the DSH Web settings interface, supporting centralized management of over ten push channels, including Server酱, PushPlus, Telegram, Feishu, DingTalk, WeChat Work, Bark, ntfy, and Email. Users can enable, edit, delete, and test channels, with the interface automatically switching between Chinese and English. The plugin provides two types of calling interfaces for other plugins or scripts: one is the host-side `pushNotifier` service, offering methods such as list, send, and test; the other is the `/push/api/*` HTTP interface, supporting message sending via POST requests. Channel data is persistently stored in
Read Morechicheng-gate: Remote Access to DSH Web GUI, frpc Tunnel, and Panel Password Gate
chicheng-gate is a secure access plugin for the DSH Web GUI, designed to address remote access, intranet penetration, and mobile adaptation challenges. It integrates four core features: first, a remote access toggle that switches the Web GUI binding address to `0.0.0.0` to support LAN access; second, frpc intranet penetration, which automatically downloads and manages the frpc process, exposes the panel to the public internet by configuring server address, port, and other parameters, and supports real-time start/stop; third, a panel password gate that enforces password verification for non-local access (including LAN and frpc tunnel), stores passwords using scrypt salted hashing, implements login rate limiting and session management, while local `127.0.0.1` access remains password-free; fourth, mobile UI adaptation that injects CSS to optimize the mobile experience. Additionally, the plugin provides
Read Moredsh-inline-images: Rendering Local Images in DSH Conversation Body
dsh-inline-images is a community plugin for DSH designed to solve the inconvenience that local image paths in LLM replies are only displayed as text and cannot be easily viewed. This plugin scans the assistant's reply text and rewrites existing local image paths into same-origin loopback URLs, allowing them to be rendered directly within the message body. It supports 9 common formats including PNG, JPG, WebP, and GIF, and provides a click-to-zoom lightbox feature. Users can adjust the maximum image width/height (default 640x420px) through the settings interface. If a path does not exist or is a placeholder, the plugin silently skips it, maintaining plain text display to ensure compatibility. Installation requires executing a build under the Web profile and adding the plugin, applicable to DSH and agent developers. Note that this plugin runs with the current DSH process permissions, so users should review the source code and M
Read Moredsh-token-billing: A Real-Time Token Billing Plugin for DSH
dsh-token-billing is a third-party real-time token billing plugin for DSH Web, designed to address the issues of non-intuitive costs, difficult historical queries, and multi-Provider cost classification when calling DeepSeek models. This plugin is based on the MIT license, with core features including direct billing using DeepSeek's official RMB prices, automatic distinction between peak and off-peak periods, support for custom model price overrides, and multi-currency conversion. Additionally, it provides a persistent historical ledger, multi-dimensional statistical dashboard, account balance query (requires API Key), local model savings statistics, and supports exporting data in CSV or JSON formats. Users can install it via the plugin marketplace (pending confirmation), as a pnpm dependency, or from local source code, with configuration taking effect immediately. This plugin is particularly suitable for users who need real-time
Read MoreDSH Usage Dashboard Plus: DSH Web Sidebar Balance and Today's Spending Plugin
dsh-usage-dashboard-plus is a web plugin for DSH, designed to address API balance monitoring and cost statistics issues. It displays the DeepSeek API balance and today's estimated spending at the bottom of the sidebar, and provides a comprehensive usage analysis panel. The plugin supports viewing call volume, latency, throughput, and token consumption broken down by model, and can include external vision model calls in cost estimation. Core features include today's cost estimation based on price tables, peak/off-peak pricing plan support, call log filtering with a 500-row limit and CSV export, as well as historical statistics implementation inherited from Stats. Installation requires executing the `dsh plugin` command in the web configuration and restarting the service; if migrating from the old dsh-stats-dashboard, the old plugin must be removed first.
Read Moredsh-plugin-text-translation: DSH Text and Document Translation Plugin
`dsh-plugin-text-translation` is a text translation plugin for the DeepSeek Harness (DSH), designed to address the issue that direct translation in game scripts and long document localization can easily damage format structures (such as color codes, variable placeholders, and subtitle timelines). This plugin does not replace the translation model but provides an "extract-assemble" capability: it inspects files via `inspect_text_file`, extracts text via `extract_text` while automatically masking and protecting tags and placeholders from RPG Maker, Unity, Ren'Py, and others, and finally restores the translation results losslessly into the original format via `assemble_text`. It supports Mtool JSON, Ren'Py, SRT/ASS/VTT subtitles, Gett
Read MoreDPH Task Board (@dph/taskboard): Adding a Status Dashboard to DeepSeek Harness Sessions
The DPH Task Board (@dph/taskboard) is a DeepSeek Harness community plugin designed to transform multi-session management into a visualized four-column status board (Todo, In Progress, In Review, Done). This plugin treats sessions as cards, supporting drag-and-drop categorization, attaching notes, priorities, and tags, along with search, JSON export for backup, and import functionality. It is deeply integrated into the Harness sidebar, allowing configuration of the model, reasoning intensity, and grouping when creating new sessions, and it displays the real-time running status of sessions via a "processing" breathing light. Data is primarily stored in the browser's localStorage, while the host side persists the board state to a local JSON file, enabling agents to directly read and analyze current work progress. The installation process requires a Node.js 18+ environment and goes through the deployment script.
Read Moredsh-neo-skin: Overlaying Neo-brutalism Skin on DeepSeek Harness Web UI
dsh-neo-skin is a zero-dependency web UI client skin plugin within the DSH plugin ecosystem, developed under the MIT license. It overlays neo-brutalism semantic tokens on top of the built-in light, dark, and system themes, achieving hard shadows, 2px borders, sharp-cornered panels, and a high-contrast style without replacing the original themes. The plugin includes two built-in schemes, "Blue Command" and "Aged Newspaper," which can be switched in real time in settings, with the state persistently stored in localStorage. It is compatible with components such as dsh-voice-input and supports multiple theme modes. Installation requires building the bundle locally, then adding it to the Web profile via the dsh plugin command, and configuring it in the configuration file.
Read MoreDSH Telemetry Redactor: Credential Redaction for Exported Telemetry Copies
`dsh-telemetry-redactor` is a community-maintained MIT-licensed plugin designed to reduce the risk of sensitive data leakage when session telemetry is transmitted externally in DeepSeek Harness (DSH). This plugin is mounted on the official telemetry waterfall, applying recursive redaction only to the outbound copies sent to the backend, while leaving the original canonical session log unchanged and without altering existing telemetry backend configurations. Its redaction scope covers three categories of content: values under high-risk key names (such as token, password), Bearer/Basic authentication information embedded in strings, and common credential patterns (such as s
Read Moredsh-context-provenance: A Read-Only Ledger for DSH Request Evidence
dsh-context-provenance is a community-maintained, MIT-licensed DSH plugin designed for safely debugging agent behavior. It compares evidence changes between two adjacent requests (such as provider, model, system, etc.) to avoid directly exposing sensitive content like prompts and messages. The plugin adopts an observe-only mechanism, relying solely on public runtime interfaces, and marks field states as Observed, Estimated, or Unavailable. It strictly limits local memory retention to observations from only the most recent two ordinary agent-loop requests, performs no file, network, subprocess, or GPU operations, and does not modify requests. When the official Cordis inspect service is present, the plugin
Read MoreDSH-Task-Planner: An Experience-Based Task Planning Plugin for DeepSeek Harness
`dsh-task-planner` is a task planning plugin for DeepSeek Harness, designed to address the issues of lacking historical solution reuse and rule hard-coding mismatches when agents execute complex tasks. This plugin adopts a planning mode with "experiential muscle memory," and its core mechanisms include: 1. **Experience Library Management**: It consolidates historical solutions into editable Markdown files (default path `~/.dsh/planner-lessons`), supporting saving, recalling, and listing experiences, and follows a lifecycle of "draft-validate-discard/skill." 2. **Conditioned-Reflex Planning**: Before a task begins, the plugin first recalls similar historical solutions, and the LLM evaluates their suitability. If suitable, it reuses and improves them; otherwise, it re-plans from scratch, ultimately outputting a step-by-step plan that integrates a capability catalog and risk analysis.
Read Moredsh-kit: Adding Dock Panels and Free Web Search to DSH Session Workspaces
dsh-kit is an MIT-licensed plugin for DeepSeek Harness (dsh), maintained by zhouzhencheng07, requiring Node.js ≥ 22. This plugin provides a toggleable dock panel for dsh web, integrating terminal, file tree, source code management, background tasks, skill management, and mobile access features, while incorporating the free multi-source web search capability from dsh-free-search v0.2.0. Users can quickly open each panel via keyboard shortcuts: the terminal supports multi-tab background execution; the file tree supports file editing and diff, with recycle bin protection for deletions on Windows; source code management handles git operations; background tasks monitor session running status; skill management covers skill configurations at various levels. Mobile access generates links and QR codes via tokens, supporting
Read MoreDSH Forge: Standalone Profiles for DeepSeek Harness Plugin Development
DSH Forge is a reproducible, runtime-aware environment for DeepSeek Harness (DSH) plugin development. By establishing an isolated `~/.dsh-forge` Home, a `forge` Profile, and developer Agent Presets, it separates development state from formal configurations, preventing pollution of the daily-used `~/.dsh`. Its core features include locally indexing Harness documentation, source symbols, and interfaces to support rapid retrieval, managing resumable plugin experiments, providing multi-level verification workflows (`quick`, `package`, `full`), and a formal Promotion scaffold. Forge supports secure synchronization of user configurations and mirroring of out-of-tree plugins while protecting the underlying
Read MoreDSH Same-Mode Sandbox Noop: A Compatibility Plugin for Handling Redundant Sandbox Fields in DSH 0.1.0-rc.6
dsh-same-mode-sandbox-noop is an MIT-licensed compatible plugin for DeepSeek Harness (DSH) 0.1.0-rc.6, maintained by zhangzujian, requiring Node.js 22+. This plugin addresses compatibility issues caused by redundant sandbox_permissions and justification fields in tool calls such as bash, pwsh, write, and edit. It does not replace DSH or introduce new sandbox policies; instead, it intercepts requests before DSH snapshots and freezes tool parameters: if the requested sandbox mode is equal to or narrower than the current session's effective mode, the plugin removes these redundant fields; for requests that genuinely require an upgrade, unknown modes, or malformed input, it passes them through unchanged.
Read More@zhangfengshun/dsh-remote-ssh: Integrate Remote SSH Files and Terminals into DSH
The DSH plugin `@zhangfengshun/dsh-remote-ssh` provides functionality similar to VSCode Remote-SSH, allowing users to directly operate remote supercomputers or servers within the DSH built-in interface without needing to synchronize files to the local machine first. The plugin supports SSH connections (including key/password authentication and ProxyJump) and enables direct reading and writing of remote files in the "Files" tab, while automatically entering a remote interactive terminal in the "Terminal" tab (key authentication only). It also supports creating remote native workspaces and provides 12 session-aware `remote_ssh_*` model tools to simplify parameter configuration. Installation requires Node engine >=20, with installation performed via the `dsh plugin add` command followed by restarting DSH, and it must run under bun.
Read Moredsh-pluginmanager: Hierarchical, Descriptive, and Uninstall Management for DSH Web Plugins
dsh-pluginmanager is a community-developed plugin for the DSH Web Profile settings page, designed to address the issues of lengthy plugin lists, unclear categorization, and difficult management. The plugin divides plugins into three layered views: native extensions (read-only), user extensions (manageable), and running plugins (temporary), providing intuitive search filtering, count badges, and bilingual display in Chinese and English. Its core features include disabling/enabling user extensions, complete uninstallation (cleaning up configurations and dependencies), and re-registration (incorporating into dependency management), as well as support for editing plugin descriptions, with built-in Chinese introductions for over 90 core plugins. The installation method adopts a bundle approach, adding and configuring package.json via pnpm to avoid conflicts with manual patches. During operations, changes are first persisted and then the loader is hot-swapped to ensure safety, with backups provided.
Read MoreDSH File Explorer: A File Browser Plugin for the DSH Web UI
dsh-file-explorer is an MIT-licensed plugin for the DSH Web UI, designed to address the pain point of switching to a terminal or editor when viewing source code, configurations, and build artifacts in the chat interface, enabling lightweight file operations within the session context. The plugin supports lazy-loaded directory trees, inline editing of text files (with version detection), and an IDE-style right-click context menu covering functions such as create, rename, copy-paste, move, and delete, as well as drag-and-drop movement, real-time refresh, and cross-platform trash bin support. Its panel can be docked or floated and automatically follows the current workspace. Installation requires executing the specified command with pnpm, with DSH version requirements ranging from 0.1.0-rc.7 to 0.1.1-rc.2, and access is available via the title bar entry after restart. In terms of security, write operations are strictly limited to the root directory of the current workspace, with read-only browsing for other areas.
Read MoreDSH-BrowserUse: A browser-use Style Browser Automation Plugin for DeepSeek Harness
`dsh-browseruse` is a browser automation plugin for DeepSeek Harness (DSH), implemented purely in JavaScript with no build step required. It drives a standalone Google Chrome instance via `playwright-core`, enabling operations such as opening web pages, reading content, clicking, inputting text, scrolling, and switching between tabs, while supporting autonomous decision-making based on page state during multi-step tasks. Core features include: 11 fine-grained `browser_*` tools; `browser_task` for natural language autonomous tasks (default maximum of 30 steps); `browser_schedule` for scheduled tasks within the current session; a user confirmation mechanism before sensitive operations (e.g., payments, deletions); manual takeover during CAPTCHA detection; and an independent Pr
Read Moredsh-plugin-opencode-bridge: Integrating OpenCode Skills and Configuration into DeepSeek Harness
`dsh-plugin-opencode-bridge` is a DeepSeek Harness (DSH) plugin developed by YYTbit under the MIT license, designed to achieve "zero-migration" bridging between OpenCode and the DSH environment. This plugin primarily addresses the need to integrate existing OpenCode toolchain capabilities into DSH, with core functionalities including automatically loading OpenCode skills (from the path `~/.config/opencode/skills/<name>/SKILL.md`) and injecting configuration context (from the path `~/.config/opencode/opencode.jsonc`). Users can use the `dsh plugin --prof` command to further manage this plugin.
Read Moredsh-plugin-meta-memory: Structured Long-term Memory for DeepSeek Harness
The dsh-plugin-meta-memory is a plugin in the DSH ecosystem designed to address issues of dispersed long-term memory and context reuse. It organizes memory into a "unit + record" structure, where each record distinguishes between `brief` (summary) and `full` (complete content), supports automatically injecting relevant `brief` into the Agent context, and maintains a global `index` to organize clues. The plugin provides a CLI tool `mm` that supports operations such as initialization, list viewing, reading, searching, and index management, allowing users to name units in a `verb-modifier-noun` format, enabling hierarchical storage and on-demand loading of memory. Its main features include memory classification, quick summary scanning, full record retrieval, and automatic context injection. Configuration options include `m
Read Moredsh-multi-selection-comments: DSH Multi-Selection Incremental Comment Plugin
`dsh-multi-selection-comments` is a DSH client plugin (v0.4.3, MIT licensed) developed by yuanc1204, designed to optimize the user experience of annotating multiple text segments in assistant replies or tool outputs. The plugin addresses the issue where traditional methods stack multiple quoted passages, causing the main input box to become excessively long, and offers a Codex-inspired interaction design. Its core functionality enables users to continuously select multiple text segments and independently specify follow-up requirements for each, with only compact status pills displayed above the main input box to maintain a clean interface. Users can manage selected content via a popup, quickly add entries with `Enter`, insert line breaks with `Ctrl+Enter`, and the requirement field is optional. The plugin only injects the annotations as structured context into the message upon sending, and displays them as pills in the chat history to avoid exposing underlying details.
Read MoreDeepSeek Harness Plugin dsh-file-panel: Right-Side Session File Panel
`dsh-file-panel` is a right-side file panel plugin for DeepSeek Harness (DSH), maintained by `yu2025-luo` and licensed under the MIT License. This plugin addresses the cumbersome issue of manually locating file paths generated by agents in the `dsh` web GUI. It automatically tracks files created or downloaded during a session, displays them centrally in the right-side panel, and supports inline image previews, text content viewing, and metadata display, as well as features such as locating files in the file manager, opening them with default applications, and copying absolute paths. Installation requires adding the plugin via a command and restarting the `dsh web` server. Its security boundaries are strict: the RPC channel is restricted to local loopback address calls, file reading is limited to observed paths only, and the panel state is cleared upon process restart without scanning historical files. This
Read Moredsh-prime-agent: A Persistent TypeScript REPL Plugin for DeepSeek Harness
dsh-prime-agent is an RLM-first control plane plugin for the DeepSeek Harness (DSH), providing a persistent TypeScript REPL environment for Prime preset sessions. Unlike the official one-shot semantics, this plugin converges model-visible tools into a single `repl` interface, allowing variables, functions, and object states declared in previous REPL cells to be reused in subsequent cells. Other DSH capabilities such as `tools`, `agents`, and `jobs` are not exposed in the model schema but are preloaded into the cell for direct invocation; attempting to call other tools directly is rejected by the guard. Core capabilities include local composition operations such as `apply_
Read MoreMathModelingAgent: An Evidence-Driven Verification Plugin for Mathematical Modeling and Competition Problems
MathModelingAgent is a DSH plugin designed for open-ended mathematical modeling, prediction, optimization, and other tasks, addressing the issue where code runs correctly but conclusions may still be erroneous. Its core mechanism registers key conclusions as Claims, which are then verified through Obligations and Evidence, explicitly distinguishing among three states: PASS, FAIL, and INCONCLUSIVE, with particular emphasis on retaining the INCONCLUSIVE state when evidence is insufficient to avoid forcing a definitive conclusion. The plugin provides two Skills: math-modeling-agent for advancing the model, and math-modeling-audit for independent auditing without modifying the original text. It supports optional tools such as Python, Lean, and Wolfram for numerical computation.
Read More