dsh-project-kanban: Project Board Plugin for DeepSeek Harness
dsh-project-kanban is a kanban plugin within DeepSeek Harness (DSH), designed to address the issue of scattered planning, task decomposition, and defect tracking information for intelligent agent developers during conversations. This plugin enables agents to directly create and update planning cards through `kanban_*` tools, with data isolated by workspace (project) and persistently stored in disk JSON files. Core features include: card tags and custom colors, high/medium/low priority marking, column sorting and drag-and-drop insertion, keyword and multi-dimensional filtering search, due date setting and overdue highlighting, recycle bin management, batch operations, statistical data viewing, 50-step undo, cross-project card movement, and source session traceability. Installation requires using the `dsh` CLI, with recommendations to add it from GitHub or npm, and some versions...
Read Moredsh-status-bar: DSH Configurable Status Bar Management Plugin
dsh-status-bar is an MIT-licensed DSH client plugin maintained by Starlight-bananice, designed to address issues with the native bottom status bar, such as information truncation and display limitations in narrow windows. The plugin replaces the built-in stats line, offering a highly configurable status bar that supports toggling and rearranging 17 information segments, including model, context pressure, tokens, task queue, and more, with multi-line wrapping to prevent truncation. Its core features include real-time TPS updates based on assistant/chunk events, fee estimation supporting peak/valley pricing and CNY/USD selection, zero-configuration default activation of 13 information segments, and a Usage & cost dialog box that includes cost trend graphs and usage history. The plugin is compatible with DSH 0.
Read Moredsh-plugin-terminal: Adding a Bottom Terminal Panel to the DSH Web GUI
dsh-plugin-terminal is a bottom terminal panel plugin developed specifically for the DeepSeek Harness (DSH) Web GUI, designed to allow users to use multi-tab Shell at the bottom of the page without leaving the browser. Based on node-pty, the plugin supports ConPTY on Windows and openpty on Linux/macOS, addressing the needs of executing commands in parallel within the chat interface, viewing logs, and running full-screen terminal programs like Codex and Claude Code. Core features include: supporting the creation, closure, and restart of multi-tab sessions, with processes continuing to run when switching tabs; persistent storage of session status, working directories, and scrollback history, which can be restored after page refresh or restart; compatibility with xterm.js 6 features, and support for t
Read Moreoss-prompt-optimizer: Optimizing Raw Instructions in DSH into Structured Prompts
`oss-prompt-optimizer` is a prompt optimization tool within the DSH plugin ecosystem, designed to rewrite brief instructions into professional, structured prompts to address output generalization issues. The plugin operates through harness's LLM service without direct API connections or credential exposure, and it is licensed under MIT. Core functionalities include: providing the `prompt_optimize` tool for agents to support single-time optimization and iterative rewriting; offering a `ctx.promptOptimizer` programming interface for other plugins; integrating a one-click optimization feature in the composer input box; enabling automatic optimization triggered by the `/optimize` prefix (enabled by default, effective only for specific prefixes); and possessing contextual and situational awareness capabilities to automatically inject roles, tasks, and other elements.
Read Moredsh-tdai-memory: TencentDB Agent Memory Plugin for DeepSeek Harness
dsh-tdai-memory is an MIT-licensed plugin (v0.2.14, requiring Node >=22.18) for DeepSeek Harness (DSH), designed to provide a long-term memory layer for the Web profile. It enables memory retention and structuring through L0 conversation capture, L1 structured memory extraction (facts/preferences/events), L2 scenario generation, and L3 user profile construction. The core functionalities of the plugin include automatically recalling and injecting relevant memories as dynamic context during prompt assembly, as well as providing two search tools: `tdai_memory_search` and `tdai_conversation_search`. The installation method is via `dsh plugin`.
Read MoreDeepSeek Harness Plugin better-model-provider: Declarative Inference Effort, Input Modality, and Capacity by Model
better-model-provider is a DeepSeek Harness (DSH) web plugin designed to address functional limitations caused by missing model capability declarations in OpenAI-compatible providers. This plugin offers a standalone settings interface, allowing users to declare and edit individual model capabilities without directly editing YAML configuration files. Its core features include customizing reasoning effort levels, configuring input modalities (such as enabling image input to avoid rejection), and editing context window sizes and maximum token capacities. The plugin supports sparse overrides for official catalog routes, saving only differences from default values and enabling one-click restoration. It also allows quick creation of declarations for unconfigured routes, though API keys still need to be managed through the official interface. Compatible with DSH 0.
Read Moredsh-wallpaper-engine: Integrate Downloaded Wallpapers from Wallpaper Engine into DeepSeek Harness Web Background
`dsh-wallpaper-engine` is a MIT-licensed DeepSeek Harness (DSH) web plugin designed to integrate locally downloaded image or video wallpapers from Wallpaper Engine into the DSH interface as backgrounds. Maintained by sakuraaa667, the plugin supports automatic discovery of wallpapers from Steam Workshop and local projects, offering thumbnail browsing, search, one-click application, and real-time refresh on the settings page. It enables video looping with muted playback and high-resolution image adaptation, provides panel and background brightness adjustments to optimize readability, and persists wallpaper selection states, automatically restoring them after restart. Installation requires a Node.js >=22 environment, is added via the `dsh plugin` command, and can be configured through a configuration file.
Read Moredsh-plugin-call-me: Enabling DeepSeek Harness to Relay Commands via Your Phone
dsh-plugin-call-me is a DSH community plugin designed to address interaction challenges when the operator is offline. It utilizes the `/call-me` service to dial an iPhone, reading out the issue and transcribing the user's verbal response back to the current session, enabling remote approval or supplementary instructions. Core functionalities include the `call_me` (bidirectional voice call, blocking for up to 5 minutes) and `text_me` (one-way text message) tools. Users can configure automatic phone outreach upon session completion or approve operations by verbally answering "Yes" during tool authorization. The plugin exclusively supports iPhones, requires Node.js >=20 and DSH >=0.0.1, and is released under the MIT license. Installation involves pairing an iPhone to obtain a 10-digit number for configuration, with no support for Android or self-hosting.
Read Moredsh-session-deeplink: Accessing and Sharing DSH Sessions via URL
dsh-session-deeplink is a web client plugin for DeepSeek Harness (DSH) designed to enable direct access and sharing of sessions via URLs. This plugin allows users to open a specific session through a link like `/?session=<id>` and automatically synchronizes the session ID in the browser's address bar when switching sessions, while preserving other query parameters. It operates entirely on the client side, adding no load to the host server. For installation, it is recommended to install it via npm packages to the web profile, or it can be pulled from GitHub. When using pnpm 10+, authorization for building must be granted in the configuration file. The plugin is currently developed based on DSH 0.1.0-rc.6, and as DSH is in a developer preview stage, future versions may require updates. Before use,
Read Moredsh-token-viewer: CC Switch-style token consumption statistics for the DeepSeek Harness Web GUI
`qwert702/dsh-token-viewer` is an MIT-licensed read-only statistics plugin for DeepSeek Harness Web GUI. Adhering to the "everything is a plugin" philosophy, it does not modify prompts or initiate new requests. Instead, it centrally displays token consumption, cache hit rates, cost trends, and account balance by reading the host's `usageLog` and proxy balance interface. Core features include the real-time TokenDock above the Composer, balance cards in the sidebar, usage statistics panels, and trend charts based on per-request data. The plugin supports filtering data by time range (Today/7d/14d/30d/All) and tabs (Request Logs, Projects, Models), and provides views for request and model statistics along with average cost. Price calculations are based on a built-in peak-valley list or P
Read Moredsh-reasoning-slider: Switching reasoning gears within the DSH model selector
`qjcnmd/dsh-reasoning-slider` is an MIT-licensed DSH plugin designed to address the cumbersome process of configuring reasoning levels when frequently switching models in DeepSeek Harness. This plugin embeds the reasoning level slider directly into the model selector overlay, allowing users to intuitively view supported levels upon selecting a model and switch them in real-time by dragging the slider, with changes taking effect upon release. Additionally, the plugin supports keyboard arrow keys and mouse wheel operations, automatically retaining the current level when switching models. If the target model does not support the current level, it automatically reverts to the default level, significantly reducing operational overhead. Before installation, ensure that `pnpm` and `dsh` are available in the environment, then add the plugin via the `dsh plugin` command from npm or GitHub, and restart the `dsh web`
Read Moredsh-meow-smooth: A DSH Front-end Experience Plugin for Mobile Devices
dsh-meow-smooth is a DSH frontend enhancement plugin maintained by Phant0Meow under the MIT license. It aims to optimize the mobile experience without any modifications to the DSH core. This plugin primarily addresses issues such as unsuitable interfaces, inconvenient interactions, and lack of notifications when accessing DSH on mobile devices. Its core functionalities are divided into three main sections: First, it optimizes the mobile UI interaction, including features like collapsing input fields on blur, line breaks with the enter key, automatic adjustment of the top bar, auto-collapsing sidebars, edge-gesture drawers, and preventing accidental zooming, all of which enhance mobile usability. Second, it offers a multi-level reminder and notification system, supporting in-page card prompts and system notifications (requiring HTTPS and VAPID keys), as well as compatibility with Web Push and Webhooks (such as Bark). This is particularly useful in multi-session scenarios, ensuring effective reception of task completions, permissions, and other alerts.
Read Moredsh-tool-stat: DSH Statistics Tool Plugin: Descriptive Statistics, Percentiles, Frequency Distributions, and Correlation Calculations
`dsh-tool-stat` is a statistical tool plugin for the DSH workflow, designed to transform numerical computations from model-based mental calculations into reproducible, structured operations. This plugin, built on pure functions and deterministic execution with no dependency on third-party numerical libraries, offers four core functions: `describe` computes descriptive statistics such as mean, variance, and quartiles; `percentile` calculates percentiles using linear interpolation; `frequency` generates frequency distributions and computes proportions; and `correlation` calculates Pearson or Spearman correlation coefficients, returning a specific status instead of an error value in cases of zero variance. The plugin enforces strict security constraints, rejecting NaN/Infinity values, normalizing `-0`, and limiting the number of observations (1 to 100,000) and percentile requests (up to 100).
Read Moredsh-file-claim: File Claim and Protection for Concurrent DSH Sessions
dsh-file-claim is a DeepSeek Harness (DSH) plugin designed to resolve file conflicts and overwriting issues when multiple concurrent sessions edit the same workspace. Maintained by Nwflower and licensed under MIT, the plugin is zero-dependency and build-free, relying solely on Node.js built-in modules and Git. Its core features include: 1. **File Ownership Management**: Provides `claim_files` and `release_files` tools for exclusive ownership declaration and release of files/directories, supporting idempotent merging and directory-level overrides. 2. **Heartbeat and Takeover Mechanism**: Automatically refreshes heartbeats and cleans up stale ownership of crashed sessions via DSH lifecycle events, with support for `stale`
Read Moredsh-agent-conductor: Dispatching Tasks to External Agent CLIs within DeepSeek Harness Sessions
dsh-agent-conductor is an MIT-licensed plugin maintained by MJorgin, designed to extend the capabilities of DeepSeek Harness (DSH) sessions. It allows users to dispatch self-contained tasks to 11 external Coding Agent CLIs (including Codex, Claude Code, Gemini, Cursor, and others) within DSH, directly bringing back the standard output results from the target CLI into the current conversation. The tool offers two integration methods: a zero-dependency Skill, which enables natural language triggering by copying the `skills/conductor/` directory; and an optional Host-only Bundle, which can be installed as a Profile tool via the `dsh plugin` command.
Read Moredsh-windows-ocr: Providing Local Windows OCR Image Input for DSH Text Models
This article introduces the DSH plugin `maxwell-feng/dsh-windows-ocr`, designed to address the issue of text models being unable to directly process images. The plugin utilizes the Windows built-in `Windows.Media.Ocr` engine to locally recognize text within images and sends the recognized text rather than the original image bytes to the model, enabling pure text models to "understand" image content. Core features include default-only sending of OCR text for privacy protection, support for an optional `passthrough` mode for vision models to receive raw images, and a fail-closed mechanism that rejects image attachments when the plugin is not loaded. Before installation, ensure a Windows 10/11 environment and the corresponding language pack are in place, which can be quickly installed via npm commands or manually configured.
Read Moredsh-file-review-tab: Incorporating DSH File Change Reviews into the Sidebar
dsh-file-review-tab is an MIT-licensed plugin for DSH web, designed to centralize the management of file changes generated by agents. It integrates the review capabilities of left0ver/dsh-file-review into the dsh-better-sidebar, while preserving the quick review entry at the end of the conversation. Core features include: displaying file changes grouped by turns in the sidebar, supporting line-level red-green diff viewing, single-file or per-turn undo and reapply; recognizing delete operations such as the 'rm' command and marking them as "Deleted"; automatically archiving earlier turns to optimize long session experiences; supporting file editing review in PTC/Code Mode; as well as session isolation and adaptive interface layout. This plugin is suitable for developers who frequently review file changes and requires dependency on Deep.
Read Moredsh-selection-ask: One-click quote of selected dialogue text to composer
dsh-selection-ask is a DeepSeek Harness (DSH) web plugin maintained by lzbaclz, designed to optimize the context reuse experience. Users simply need to select specific text in the conversation interface, and a "Ask DeepSeek" button will appear next to the selection. Clicking this button automatically fills the selected content into the composer in Markdown blockquote format, eliminating the need for manual copy-pasting. The plugin only processes text in conversation records, using the `inputActions.setDraft` API to write drafts, ensuring security. Version 0.1.0 is available under the MIT license, with pre-built packages provided for installation without local building. Installation requires adding the plugin via the `dsh plugin` command from a GitHub or local link, and enabling it may prompt a restart.
Read Moredsh-funpack: Integrating Procrastination, Atmosphere, and Desktop Pets into the DSH Web Interface
dsh-funpack is a lightweight MIT-licensed plugin designed for the DeepSeek Harness Web interface, aimed at providing instant feedback and interactive entertainment during development breaks. This plugin runs as a single file with zero dependencies, offering a one-click command button above the input box. It integrates features such as random compliments, developer fortune, procrastination reports, Pomodoro timer, break suggestions, ambient sound synthesis, and virtual pet interaction. Key features include support for customizable Live2D virtual pets with appearance and voice options (supporting Cubism 2/4 models), Web Audio-based fileless ambient sound generation, an accumulable achievement season system, and a "Boss Stealth" mode (which hides entertainment elements and simulates a work interface). Additionally, the "Code Garden" feature combines focus time with virtual planting to create a simple tracking record. The plugin can be installed via the `dsh plugin` command.
Read Moredsh-balance: Permanently Monitor DeepSeek Balance and OpenCodeGo Quota in dsh Web
`dsh-balance` is a persistent plugin for `dsh web`, designed to simultaneously display the balance of the DeepSeek API account and the remaining quota of the OpenCodeGo plan. This plugin is open-source under the MIT license and maintained by linshule. Its core features include a draggable and position-memory badge in the lower-left corner, a click-to-expand details layer (containing a progress bar and reset time), and a standalone settings page. Users can manage the DeepSeek API Key (persisted locally with 0600 permissions) and the OpenCodeGo Key (defaulting to reading from credentials or manually specifying). The plugin supports automatic data refresh at set intervals (default 1 minute) and automatic retries during network jitter. Installation requires adding the plugin first and then restarting `dsh web`, with support for installing from GitHub or locally.
Read Moredsh-whale-animation: Providing Whale Animation Status for DSH Web Turn Status
dsh-whale-animation is an MIT-licensed plugin for DSH Web, designed to enhance visual cues for interface states. The plugin offers six sets of original looping whale animations corresponding to Deep Dive, Sonar, Tool Run, Stream, Calm, and Retry states, and includes a lightweight animation director mechanism. Its core features include automatic animation rotation every 9 seconds and a smart override mechanism based on English and Chinese keywords, which automatically switches to the corresponding animation when detecting text such as search, tool run, writing, waiting, or error. The plugin strictly limits visual scope, involving no accounts, data storage, network requests, or user content processing, ensuring zero runtime external resource dependencies. It supports dark themes and adapts to various viewport sizes, and automatically reduces dynamic effects when the system enables "reduce motion."
Read Moredsh-task-dag: Adding Real-Time Tasks and Workflow Topology to DeepSeek Harness
`dsh-task-dag` is a Web plugin for DeepSeek Harness (DSH), designed to address the issue of scattered multi-agent task states. Based on DSH's existing Client projections, this plugin integrates Sessions, sub-agents, Agent Teams communication, and Workflow execution into three interactive topology views—Overview, Agent Teams, and Workflow—without requiring additional databases or polling. Core features include: the Overview serving as an entry point to display the overall topology; the Agent Teams view presenting team structures, task dependencies (blockedBy DAG), and communication overlays, with the note that its data relies on Host composition events.
Read Moredsh-layered-memory: DeepSeek Harness L0-L3 Layered Distillation Memory Plugin
`dsh-layered-memory` is a memory plugin designed for DeepSeek Harness, aimed at addressing issues such as context loss and repetitive questioning in long conversations and cross-session interactions. It automatically executes background processes including L0 raw conversation capture, L1 atomic memory extraction, L2 scene block integration, and L3 core profile distillation to transform conversational information into reusable memory assets. Before model execution, the plugin automatically recalls relevant memories based on length and time budgets, injecting them into the context to avoid slowing down the conversation. The plugin offers three proactive query tools, such as `memory_search`, supports four session-level memory tiers—daily, work, smart, and closed—and features a "write-only, no-read" toggle, same-session deduplication, automatic context compression resets, and time-based weighted recall functionality. The LLM token costs incurred during the distillation process are recorded.
Read Moredsh-reasoning-settings: Add a Reasoning Intensity Settings Page for DSH Custom llm-pi-ai Providers
`dsh-reasoning-settings` is a MIT-licensed plugin for DSH, primarily used to configure the reasoning strength of custom `llm-pi-ai` providers. This plugin addresses two core issues: firstly, it allows users to declare reasoning strength levels from `off` to `max` for custom models, defining the actual wire value sent and the provider's default values; secondly, it optimizes subagent invocations to correctly inherit the parent's provider, model, and reasoning strength, avoiding the use of incorrect official default settings. The plugin provides a settings page to automatically read added providers and models, enabling secure parameter configuration via the official API without touching API keys. It extends the `subagent` and `subagent_for` functionality.
Read MoreDSH Plugin dsh-workspace-explorer: Browse Workspace Files and Insert References in Web UI
dsh-workspace-explorer is a native plugin for the DSH Web UI, designed to address the inconvenience of manually inputting file paths in a workspace. It adds a "Workspace Files" button to the session header, which opens a directory tree browser for the current workspace. Core features include on-demand lazy loading of directories, automatic hiding of noisy directories, display of colorful file icons, and support for appending file references to the input box via click or drag-and-drop, along with multi-select batch insertion. The plugin includes a built-in settings panel for adjusting parameters such as hidden directories and preview line counts, and supports text pagination preview as well as direct editing and saving to disk. It adapts to light and dark themes based on DSH design tokens and supports both Chinese and English. Installation can be completed via the `dsh plugin` command or through dynamic plugin methods. This plugin is ideal for developers who frequently need to reference local source code, configuration files, and other files in models, reducing
Read More