dsh-toc-tail: Session Timeline and One-Click Backtrack for DeepSeek Harness Web Interface
# Introduction After scrolling through a long conversation in the web interface, two issues become apparent: locating a specific request relies solely on scrolling; and when attempting to revert to a previous request, handling the content that follows a recalled message is not intuitive. `dsh-toc-tail` addresses exactly these two problems. It compresses each user request into a horizontal line on a vertical timeline located to the right of the session column. The TOC panel supports one-click tracing back to any user request, with options to restore LLM summaries and code states during the trace. DSH's philosophy is "Everything is a Plugin," and Web interface extensions follow the same plugin mechanism. The following introduces the positioning, features, and usage of `dsh-toc-tail`. # What is it? `dsh-toc-tail` is a conversation [component] for the DeepSeek Harness Web interface.
Read Moredsh-xiaomi-tts: Add Xiaomi MiMo TTS voice reading to DSH Web
Introduction: When running agents on DSH Web, the output is plain text by default. To enable voice output, a common approach involves implementing a custom Text-to-Speech (TTS) layer: applying for an API key, setting up a proxy on the Host side, handling audio streams in the browser, and cleaning up code blocks and URLs before playback. While each individual step is not complex, integrating them can be time-consuming. dsh-xiaomi-tts streamlines this process into a directly installable DSH Web plugin. Upon installation, an "Read Aloud" button appears next to the dialogue. Clicking this button initiates playback. What is this? dsh-xiaomi-tts is maintained by ppy-web. Its purpose is succinctly defined as: adding Xiaomi MiMo TTS voice playback to DSH Web. It is built upon Xiaomi MiMo...
Read Moredsh-assistant-optimization: DSH web plugin for collapsible model output and enhanced prompts
# Introduction After starting a service using the web profile of DSH (DeepSeek Harness), several issues are encountered almost daily: - Some OpenAI-compatible APIs lack a dedicated `reasoning_content` channel; the Chain of Thought is mixed directly into the content stream, and DSH renders the entire block as plain text, burying the answer within a large chunk of thinking. - Once a turn is complete, the screen is cluttered with lines of thinking, tool calls, and intermediate explanations; what is often truly relevant is only the final reply. - If the process is manually stopped or the session encounters an error, resuming requires manually reconstructing the input. - When the model outputs Mermaid charts, the interface displays only a block of code text. dsh-assistant-optimiz
Read Moreuse-opencode-local-provider: Integrate OpenCode Zen as a local provider into dsh
use-opencode-local-provider is an MIT-licensed dsh plugin that wraps a locally running opencode serve into a dsh provider. It launches an OpenAI-compatible bridge providing chat completions and models endpoints, converting requests into opencode serve sessions. Requests are routed through the opencode client channel, requiring no OpenCode API key and not consuming the public /zen/v1 quota. To install, execute the plugin add command in the profile directory, configure the models to be exposed in `cordon.patch.yml`, and restart dsh. Configurable options include the port and provider.
Read Moredsh-llm-headers: Inject custom HTTP headers into DeepSeek Harness's LLM requests
# Introduction When deploying DeepSeek Harness (DSH) in an enterprise intranet, a common barrier arises: the egress gateway validates the request's `user-agent`, and requests lacking the specified brand text are rejected outright. However, DSH's LLM adapter forces the writing of its own attribution `user-agent`, which cannot be bypassed by simply modifying a field in the configuration file. `dsh-llm-headers` solves this problem: it is a DSH plugin that injects your configured request headers at the HTTP layer for requests sent to the LLM provider. Below is an introduction to its principles, installation methods, and typical usage. # What is dsh-llm-headers `dsh-llm-headers` is for Deep...
Read MoreLingxi (dsh-client-ui-lingxi): Consolidate scattered ideas into a DSH Web GUI plugin for the project.
Lingxi is a DSH Web GUI plugin maintained by ppjun2026 (MIT License, v0.2.0) designed to address the lack of a dedicated space for ideas to evolve from a "single sentence" to a "project proposal." It features an "Idea Pool" that displays ideas as a card wall, along with "Idea Seeds."
Read Moreopenma-ai/dsh-mcp-apps: Add MCP Apps support for DeepSeek Harness
# Foreword MCP Apps allow MCP tool results to carry `ui://` HTML resources, which are rendered by the client into an interactive interface. To get this working in DeepSeek Harness (DSH), a complete set of components is required, including the protocol Host, sandbox renderer, and display mode management. In contrast to existing approaches: DSH's Web build comes with an inline-only renderer that only covers a single surface type. Below, I introduce openma-ai/dsh-mcp-apps: it packages the MCP Apps Host and Web renderer into a standard Cordis plugin, filling in this layer of capability. # What is openma-ai/dsh-mcp-apps (npm
Read MoreSoftware Project Governance: Centralize the management of AI code delivery evidence, gates, and risks
software-project-governance is an MIT open-source project maintained by peterwangze, providing an evidence-based governance layer for AI coding delivery to address the issue where facts such as task changes, reviews, and verifications are scattered across chat logs. Core features include evidence-based delivery (recording what was changed, why, how to verify, and remaining risks), gatekeeping and centralized risk management, distinguishing between genuine independent reviews and execution in degraded/dependent environments, and getting started in five minutes by initializing the .governance/ directory via /governance. The plugin supports multiple agent loading paths such as Claude Code, Codex, and Gemini CLI, and provides a read-only local web console. The default preset is classic-phase-gate.
Read Moredsh-ask-guard: Add a timeout guard for ask_user_question to prevent turns from hanging indefinitely.
# Preface If you have used `ask_user_question` in DeepSeek Harness (DSH), you may have encountered the following scenario: the agent asks a question, and the question card is pushed to the browser via a mux WebSocket in a single push. Once the page is sent to the background, the laptop enters sleep mode, or the connection enters a half-open state, that frame is lost. Without a heartbeat, the client cannot perceive the question's existence, so the tool waits indefinitely. Practical tests have observed hangs lasting over 6 hours, and the only recovery method is manually pressing "Stop". For upstream feedback, see deepseek-ai/deepseek-harness#1554 (discussion). dsh-ask-guard is designed to address this issue.
Read Moredsh-plugin-table-zoom: One-click scrollable floating window for DSH chat tables
dsh-plugin-table-zoom is a chat table floating window plugin for the DeepSeek Harness (dsh) Web GUI, maintained by Pasumao under the MIT license, currently at version 0.3.4. It addresses the issues in DSH chat flows where long and ultra-wide tables are difficult to read completely and inconvenient to copy: since version 0.1.1-rc.2, dsh has enabled the md-table-wide branch for wide tables with 4 or more columns, but the default overflow is hidden and scrollbars only appear on hover, still restricting the reading experience. The plugin automatically injects a "⛶ Float View" button for tables with $\ge$ 9 rows or horizontal overflow exceeding 2px. Clicking it opens a floating window that supports independent scrolling, drag-and-drop movement, and resizing via the bottom-right corner (from 320×200 to the viewport limit), with the title displaying the row and column counts; supports Ct...
Read Moredsh-approval-auto-review: Automatically review DSH approval requests using a locked Guardian agent
# Introduction When agents run long tasks in DeepSeek Harness (DSH), the approval process presents a dilemma: manually confirming every tool call frequently interrupts the task, while completely opening permissions leaves no guardrails against the model's erroneous commands. The compromise is to introduce an independent, limited-capability review agent to evaluate each pending approval operation and make a decision. The plugin introduced in this article, dsh-approval-auto-review, implements this approach. # What is this dsh-approval-auto-review is an automatic approval review plugin for DeepSeek Harness, maintained by perlied03, licensed under MIT, current version 0.6.1. It handles operations that are routed to the configured...
Read Moredsh-ui-customizer: Change wallpaper and theme in the DSH Web settings dialog
# Introduction Most users of DSH spend the majority of their time on the Web GUI. The official interface style is fixed, and there is no built-in way to change the background image for the entire page or customize the color scheme to your liking. `qe2592008/dsh-ui-customizer` solves this problem by adding two sub-menus, "Background Image" and "Theme," to the official settings popup, supporting image/video wallpapers and full interface color customization. The following sections introduce its features, installation, and uninstallation methods, with content based on the plugin's README and package.json. # What is dsh-ui-customizer (GitHub repository: qe2592008/dsh-ui-customizer, npm package)
Read Moredsh-plugin-image-tools: Three tools to add image capabilities to DSH Web GUI
# Introduction When using the Web GUI of DeepSeek Harness (DSH) to create agents, there are three image-related scenarios that have always been difficult to handle: 1. The model wants you to select an image from a set of candidates (e.g., choosing a book cover), but images cannot be placed in the tabs; the model can only print the file path for you to open manually. 2. The model generates an image and wants to mix it with text in the response body, but there is no channel for this either. 3. You send an image into the chat, and a text-only model adapter directly reports `UNSUPPORTED_CONTENT`, interrupting the turn. dsh-plugin-image-tools is a DSH plugin designed for these three scenarios: image selection card, inline images in the response, and blind model receiving images. Three tools, each handling a specific task, with zero configuration.
Read Moredsh-bisect-debug: DeepSeek Harness plugin for quickly locating bugs using binary search
# Introduction When troubleshooting bugs, the most time-consuming part is often not the fix itself, but locating the issue. "The bug is definitely in the current code, but I don't know which section." "A 405 error occurs when clicking authorization, and it is unclear whether the blame lies with the frontend, gateway, or downstream API." "It was working fine last week, but I don't know which commit broke it." Facing such problems, relying on reading through code or guessing by experience is inefficient and often leads to random, ineffective adjustments. Binary search is the standard approach to handling such issues: halving the search range in each round allows for convergence to a specific function, boundary, or commit within just a few iterations. `dsh-bisect-debug` is a DeepSeek Harness (DSH) workflow plugin that encapsulates this method into an executable process. DSH's philosophy is "Everything is a plugin."
Read Moredsh-browser-plus: A DSH Plugin Making Agent Browser Truly Visible and Controllable
## Introduction A common practice in browser automation is to run the Agent in a headless process invisible to the user. The problem is that the user has no way to confirm what is happening on the page or where the Agent clicked; when encountering steps that require login or manual judgment, there is also no natural handover method. dsh-browser-plus is designed to address this issue. It keeps the browser window visible to the user while providing the Agent with reliable Chrome DevTools Protocol (CDP) capabilities. Users can interact directly on the real page, and the Agent can execute, take over, and resume tasks synchronously. ## What is it dsh-browser-plus is a visual Agent styled with EGO, specifically crafted for DeepSeek Harness (DSH).
Read Moredsh-paste-names: Enable DSH Web Input to support pasting and drag-and-drop of non-image files
# Introduction In the DSH (DeepSeek Harness) Web interface, referencing workspace files is limited to two default methods: manually typing `@` to select from a hierarchical menu, or dragging files directly into the input box. The former is time-consuming for locating deep files in large repositories, while the latter immediately triggers an error stating "Only images in PNG, JPG, WebP, or GIF formats are supported," rejecting any files or folders other than images. `dsh-paste-names` fills this gap: it parses pasted non-image files or folders into native DSH `@path` file references, and inserts absolute path text when dragging and dropping. The following sections introduce its features, implementation details, and installation instructions. # What is it? `dsh-paste-names` is a DSH Web plugin.
Read Moredsh-bayes-predict: Estimating Individual Stock Rise Probability Using Bayes' Theorem in DSH
In DSH (DeepSeek Harness), when an agent analyzes a stock, it typically yields only a qualitative description. To determine the probability of a rise within the next N days, assess the current trend, or gauge the risk of holding the position, one must either run a separate quantitative script or switch to market software to manually track indicators. dsh-bayes-predict turns this into a DSH plugin: centered on the Bayesian theorem, it fuses multiple technical indicators into a posterior probability estimate for a "rise within N days" and outputs the results directly via two entry points: the chat interface and the browser panel. The following sections introduce its positioning, core features, and installation instructions. dsh-bayes-predict is a DSH individual stock prediction plugin based on the Bayesian theorem.
Read Moredsh-nl-model-switch: Switch DSH session models with a single sentence in the conversation.
# Introduction In DSH, switching models is done natively by typing `/model` and selecting from a popup. While this workflow is acceptable on desktop, it interrupts the flow in two scenarios: first, when you chat with the model via an IM bot on your mobile device, you lack a panel to click on; second, when you are in the middle of a discussion and want to hand off the remaining task to another model without leaving the chat interface. `dsh-nl-model-switch` fills this gap: simply say "switch to deepseek-v4-flash" in the chat, and the current session's model switches immediately, preserving the context so the new model continues processing your previous request. # What is this `dsh-nl-model-switch` is a DSH host plugin created by passi.
Read Moredsh-tool-tariff: DSH Peak/Valley Time Period Query and Balance Reminder Plugin
# Preface If you want to schedule tasks to run during off-peak hours and finish before peak hours arrive, you need to be constantly aware of whether it is currently peak or off-peak and when the next switch will occur. Existing methods involve maintaining a schedule manually and cross-referencing it, which is cumbersome and prone to error; similarly, checking the remaining balance of the DeepSeek API requires interrupting your current work. `dsh-tool-tariff` is a DSH (DeepSeek Harness) plugin that integrates these functions into the session: the model can directly answer time-related queries, automatically reminds you before peak hours, and a real-time status badge remains in the top bar of the Web GUI. DSH's philosophy is "Everything is a plugin," making this capability suitable for mounting as a plugin. The following sections introduce the features, installation, configuration, usage, and precautions in that order. # What is this d
Read Moredsh-codex-bridge: A bidirectional plugin to integrate Codex CLI with DeepSeek Harness
**Introduction** When developing agents using DeepSeek Harness (dsh), you may encounter requirements such as having the agent in the conversation find an external coding agent—such as OpenAI's Codex—for a second opinion, or running a coding task in parallel. Manually handling this involves spawning a `codex` process, capturing JSONL output, polling for status, and then ingesting the results back into the conversation. This scaffold code is unrelated to the main task yet must be written from scratch every time. `dsh-codex-bridge` is designed to solve this problem. DSH's philosophy is "everything is a plugin," where capabilities are integrated into the harness as plugins. This project is a concrete implementation of that concept. Below, I introduce what it is and how to use it.
Read Moredsh-agents-plugins: A Bridge package for integrating Codex, Claude Code, and Pi plugins into DeepSeek Harness
# Introduction If you have previously installed plugins in Codex, Claude Code, or Pi, you likely have a collection of assets at your disposal: hooks, slash commands, MCP servers, and skills. After switching to DeepSeek Harness (DSH), these cannot be migrated directly—they are written according to the format of another agent runtime, and their loading methods and lifecycles do not match. Rewriting them one by one is clearly not cost-effective. DSH's philosophy is "everything is a plugin," and such bridging work can itself be completed by a plugin. The openma-ai/dsh-agents-plugins introduced below is such a bridge: allowing you to install Agents in DSH through a package.
Read Moredsh-port-guard: A DSH plugin for handling port occupancy by first identifying it
dsh-port-guard is a DeepSeek Harness plugin developed by PangYiMing (v0.1.0, MIT license) used to handle EADDRINUSE port conflicts when the dev server starts. It transforms the brute-force kill into an informed judgment: first using a script to locate the occupying process, viewing information such as start time, uptime, controlling terminal, and project ownership, and then handling three scenarios—if its own dev server is still running...
Read Moredsh-ultracode: Enable UltraCode working mode for DeepSeek Harness sessions
# Introduction In DSH (DeepSeek Harness), the quality of a development session often depends on how clearly you articulate your workflow: analyzing before acting, breaking down tasks into verifiable steps, handling independent tasks in parallel, and running tests before declaring completion. Manually writing these requirements into the prompt every time is tedious and prone to omission. `dsh-ultracode` solidifies this approach into an agent preset. It is a plugin for DeepSeek Harness that adds UltraCode workflow guidance on top of the built-in code (PTC) preset. It can be installed with a single command and takes effect immediately upon selection. The following sections introduce what it can do, how to install it, and how to use it. # What is it `dsh-ultracode` is maintained by ParticleLight, licensed under the MIT license, and the current version...
Read More@paicat1/dsh-screenshot: DSH Hotkey Screenshot and Agent Auto-Screenshot Plugin
When collaborating with agents in DSH (DeepSeek Harness), sharing screen content is frequent: error logs, UI states, or images visible only in local applications. The typical approach is to manually take a screenshot and paste it into the dialog box. While the image enters the conversation, finding the original file later is difficult. Conversely, there is usually no ready-made entry point to allow the agent to "view the screen." @paicat1/dsh-screenshot consolidates these two tasks into a single plugin: users can capture screenshots via browser hotkeys, and only the PNG path is passed to the agent; the agent can also take self-service screenshots via a tool. The entire implementation is pure PowerShell, with zero dependencies and zero binaries. The following sections will introduce its features, installation, and usage.
Read More