dsh-skill-router: Rule-based skill routing before step execution
# Introduction When building agents with DeepSeek Harness (DSH), the timing of skill loading is usually left to the model to decide. Skills are exposed as a catalog to the model, visible at every step, forcing the model to re-evaluate whether to load them constantly. The result is unstable decision-making: skills may load too late, load the wrong one, or fail to load entirely. dsh-skill-router shifts this to a rule-driven approach: before each step, pure rules match the latest user message. If a match is found, the corresponding skill content is injected into the current step; otherwise, no intervention occurs. The entire process involves zero LLM calls and zero token costs. Below, we introduce its usage. # What is it? dsh-skill-router is a
Read Moredsh-lens: Get diagnostics while coding in DeepSeek Harness
# Introduction When using DSH (DeepSeek Harness) as a coding agent, a common pain point arises: after the model performs `write` / `edit` / `bash`, determining whether the generated code contains type errors or violates lint rules requires you to run the toolchain yourself. The later the problem is discovered, the higher the repair cost. In the Pi host, pi-lens has already solved this: the engine runs LSP, linters, formatters, type checkers, and structural analysis while the agent is writing code. dsh-lens simply connects the same engine to DSH. Below is an introduction to its positioning, capabilities, installation, and configuration. # What is dsh-lens dsh-lens is pi-lens for DSH.
Read Moredsh-vision-plugin: Add an "Image Input" declaration checkbox to the DSH model directory
# Introduction When integrating custom models into DSH, you may encounter a scenario where the endpoint itself supports multimodal capabilities, but sending an image directly in the chat dialog results in an error stating "Image input not supported." The reason is that the DSH host reads the `input: ["text","image"]` declaration on the model entry to decide whether to pass the original image directly to the model, while the interface in "Settings → Models → Model Catalog" has never exposed an entry point for this declaration. dsh-vision-plugin aims to address this step: by adding an "Image Input" checkbox on the model page in the Model Catalog. Once checked, the declaration is written into DSH settings, allowing the custom model to directly receive images in the chat dialog. # What is it dsh-vision-p
Read Moredsh-mcp-sync: Scan local MCP configurations and register tools for direct model invocation.
<tool_call>Bash<arg_key>command</arg_key><arg_value>ls -la "D:\D-Users\D-Administrator\.zcode\workspace\default" | grep -E "obj_|arxiv"</arg_value><arg_key>description</arg_key><arg_value>List obj* and arxiv related directories</arg_value></tool_call>
Read Moredsh-dbhub-live: A DBHub bridge plugin enabling DSH to directly and safely operate on databases.
Preface When developing with DSH, allowing the model to query the database directly is a common requirement. There are two typical approaches: pasting the DSN into the conversation for the model to connect itself, which causes the password to enter the model's context; or maintaining a persistent database MCP service yourself, which requires managing ports and process lifecycles. dsh-dbhub-live takes a different path: the password remains only on the host side, while the model is given only a `source` handle. It spawns a one-time process for each query and recycles it immediately after execution. Below, I introduce its design, installation, and usage. ## What is it dsh-dbhub-live is a DSH plugin maintained by mr-mihu, currently version 4.0.0, licensed under the MIT license. It serves as a bridge based on DBHub (the database MCP server):
Read Moredsh-git-component: Add a Git panel to DeepSeek Harness WebUI
# Introduction When working with the DeepSeek Harness (DSH) WebUI, checking what has changed in the current repository, committing, and pushing usually requires switching back to the terminal to run `git status` and `git commit`. This back-and-forth often interrupts your current tasks. The dsh-git-component plugin solves this problem: it mounts a floating panel on the right side of the WebUI to display workspace changes in real-time. Commits and pushes can be completed directly within the panel, and if the commit message is left blank, it will be automatically generated. Below, we introduce the positioning, features, installation, and usage of this plugin. # What is this? dsh-git-component is a Git plugin for the DeepSeek Harness Web GUI, created by nieyun...
Read Moredsh-llm-ollama: Integrating Ollama Cloud Chat and Web Search Crawling for DeepSeek Harness
dsh-llm-ollama is an MIT-licensed DeepSeek Harness (DSH) integration plugin for Ollama Cloud, maintained by NOirBRight, designed to address the lack of an entry point for editing third-party models on the DSH Models page. Its core features include: enabling Ollama Cloud chat via Ollama's OpenAI-compatible Chat Completions endpoint (via the shared pi-ai adapter); implementing model discovery through native APIs (/api/tags, /api/show) to provide metadata such as context window, vision, thinking, and tools; and registering ollama-cloud's Web Search and Web...
Read Moredsh-remote-server: Mention servers with @ and execute SSH commands in a DeepSeek Harness session.
dsh-remote-server is a DeepSeek Harness plugin (MIT License) developed by MRZHUH, enabling models to mention the server using '@' within a session.
Read Moredsh-plugin-api-usage-monitor: A DSH sidebar plugin to turn the DeepSeek API balance into a health bar
# Introduction When calling the DeepSeek API, there are several items that require frequent manual verification: how much balance remains, how much has been spent today, and whether the current pricing is peak hours. While the official platform can display this data, you need to switch to the development environment to view it separately. dsh-plugin-api-usage-monitor brings these three items into the Web sidebar of DeepSeek Harness. Below, we introduce its features, installation, and daily usage. # What is this dsh-plugin-api-usage-monitor (version 1.9.0-rc.4) is an API usage, balance, and cost monitoring plugin designed for the DeepSeek Harness Web sidebar, created by ninedaywo
Read Moredsh-approval-mode: Add a toggleable approval mode button to DSH
# Introduction When running agent tasks with DSH, permissions are generally selected from three tiers: Read Only, Workspace Write, and Full Access. Choosing Workspace Write restricts file operations within the sandbox to the workspace, which is safer; however, the downside is that every tool invocation requires a manual approval click. With numerous task steps, these confirmations significantly disrupt the workflow. Switching to Full Access eliminates the approval popups but also removes sandbox restrictions on file operations, keeping the two features coupled. `dsh-approval-mode` decouples these two aspects: it maintains the Workspace Write permission while adding an "Approval Mode" button nearby, transforming "step-by-step confirmation" into an independent toggle. The following introduces this plugin.
Read Moredsh-chime: Make DeepSeek Harness Web GUI play a ding-dong when a task is completed
## Preface When running agent tasks via the dsh Web interface, a common workflow is to "send the command, switch to do other things, and later switch back to check on progress." Switching too early results in idle waiting, while switching too late means the task has already stopped and is waiting for your input. dsh-chime is designed for this exact scenario: as soon as the current session's task ends, a chime plays in the browser, eliminating the need to stare at the page. The philosophy of DSH is "everything is a plugin," making such small interface enhancements ideal for plugin-based completion. Below is an introduction to the positioning, features, and usage of dsh-chime. ## What is this dsh-chime (GitHub repository `mystery-god/dsh-chime`) is for the DeepSeek Harness Web GUI
Read Moredsh-agent-pill: DSH's agent activity pill and floating summary panel
dsh-agent-pill is a DSH web plugin maintained by mozhuanzuojing (MIT, v0.14.8) that addresses visibility issues during long-running agent tasks: it condenses key session activities into a status capsule in the top-right corner, allowing a floating panel to be expanded via Ctrl+Alt+P or by clicking the capsule. The plugin is divided into a host side (aggregating Goal, Subagents, Agent status, workflow execution, and background tasks, exposing a JSON API via /pill/api) and a client side (the capsule and anchored panel). Main features include a draggable capsule, position persistence, theme following DSH's light/dark mode switching, section folding, and empty state hiding; Goal cards display stages, round progress bars, and workflow history (latest...
Read Moredsh-webfile: Adding Approval-Gated S3/FDS/FTP File Tools to DeepSeek Harness
# Introduction The philosophy of DeepSeek Harness (DSH) is "everything is a plugin." Once you need an agent to help you move logs or sync files, remote storage becomes unavoidable: S3, Xiaomi Galaxy FDS, or even still-operational FTP servers. The common approach is to write keys directly into the configuration and let the model call command-line tools within a shell—this exposes keys in the configuration and session logs, and operations like deletion and overwriting are not intercepted. dsh-webfile addresses this issue: it encapsulates three types of storage into eight agent tools. Read operations do not require approval, while change operations request approval one by one. Large file transfers are handled as background tasks that can be canceled at any time. Below is an introduction to this plugin. # What is this? dsh-webfile is DeepSeek H...
Read Moredsh-web-attention-badge-enhanced: Add triple session reminders to DSH Web UI
This article introduces the attention badge plugin `nbfys/dsh-web-attention-badge-enhanced` for the DeepSeek Harness (DSH) Web UI, which is an enhanced fork of `Luaphes/dsh-web-attention-badge`. When a session is waiting for user input (e.g., `ask_user` prompts, approval hints) or has been completed but not opened, the plugin provides reminders via three channels: a clickable number badge in the top-left corner, tab title prefixes `[Waiting]`/`[Done]`, and whale favicon recoloring; amber indicates waiting for input, while green indicates completion. Compared to the upstream v0.3.2, the changes are concentrated in `lib/client.js` in three locations: replacing the count with tab title prefixes,
Read Moredsh-mcp-xcode: Integrate Xcode headless capabilities into DSH agent
# Introduction When developing macOS/iOS applications in DSH (DeepSeek Harness), there is a practical challenge: tasks such as building projects, running unit tests, viewing SwiftUI previews, and operating simulators are typically performed within the Xcode UI, preventing agents from intervening. DSH's philosophy is "everything is a plugin," and such capability gaps are usually filled by plugins. Starting with Xcode 27, the system comes with built-in headless MCP services (`xcrun mcp-server` / `xcrun mcpbridge`) that expose the aforementioned capabilities in the form of MCP tools. However, to use them within DSH, a bridging layer is required to convert these MCP tools into...
Read Moredsh-computer-use-windows: Windows Computer Use Plugin with Security Access Control
The difficulty in enabling agents to operate on Windows desktops lies not in a simple "click," but in three specific challenges: the screenshot captured by the model might be of the wrong window; custom-rendered interfaces like WeChat and Electron lack accessibility trees, rendering the `element_index` path unusable; and operations such as deletion or payment are irreversible without a confirmation step. Common approaches either rely entirely on guessing coordinates from screenshots or require the target application to provide a complete UI Automation tree; neither is stable in real-world desktop environments. Below, I introduce `dsh-computer-use-windows`, a plugin for DeepSeek Harness (DSH). DSH's philosophy is "everything is a plugin," and this plugin fills the gap for Windows.
Read Moredsh-tui: An Interactive Terminal Interface for DeepSeek Harness
`dsh-tui` is a full-screen interactive terminal UI plugin developed by the community for DeepSeek Harness (DSH), serving as a plugin-based supplement...
Read Moredsh-codex-relay: Provider Plugin for Selecting Codex Models in DeepSeek Harness
# Introduction The design philosophy of DeepSeek Harness (hereinafter referred to as DSH) is "everything is a plugin," and model inference is also handled by provider plugins. If you are already logged into Codex CLI locally and wish to directly utilize the models exposed by your native Codex account and their inference strength within a DSH session—rather than manually transferring context between two separate tool sets—you need a bridging layer. The plugin introduced below, `@necokeine/dsh-codex-relay`, accomplishes exactly that. # What is this `dsh-codex-relay` is a DSH provider plugin maintained by necokeine, belonging to the model inference category. The current version is 0.1.2, and the license is MIT. It works by...
Read Moredsh-tryboard-plugin: A built-in Trello-style Kanban board in the DSH Web GUI
dsh-tryboard-plugin is a DSH Cordis plugin maintained by nexsjournal (MIT License) featuring a built-in Trello-style Kanban board in the DSH Web GUI. It introduces a new entry point above the "Settings" section in the sidebar, with the board page occupying the main content area. It supports multiple boards and defaults to four columns (To Do/In Progress...
Read Moredsh-plugin-ima-sync: Automatically Upload DSH Conversation Progress to Tencent IMA
dsh-plugin-ima-sync is an MIT-licensed DSH plugin developed by nan1010082085 that automatically uploads DSH Web conversation progress to Tencent IMA's Daily Notes and Work Knowledge Base. It mirrors the local Claude Code `ima-upload` workflow and is designed exclusively for the web platform. Install with `dsh plugin --profile web add`.
Read Moremuretai-dsh-skill: Connect your DSH agent to the Muretai agent network
muretai-dsh-skill is a dsh bundle plugin (MIT licensed, v0.2.53) designed to connect agents on DeepSeek Harness (dsh) to the Muretai network—a network where AI agents from different users discover and communicate directly through an introduction mechanism, featuring signed and end-to-end encrypted messages without a directory service. The plugin provides four core capabilities: on-device minting of Ed25519 self-identities (private keys stored only in key files); invitation-based introduction and encrypted communication (can join via public community rooms without an invitation); in-bound email wake-up to replace polling; and MCP tools such as whoami, read_inbox, and send_message. There are three installation methods: one-step installation via install.sh (
Read MoreMu-scorpio/token-usage-counter: Add persistent token usage statistics to DeepSeek Harness
# Preface When using DSH (DeepSeek Harness) to run agents, token usage is the most direct cost indicator: tracking how many were used over a period, cache read/write operations, and which providers and models consumed the most requires a persistent and queryable record. DSH includes a built-in usage accumulator and usage page that suffice for basic monitoring; however, if you wish to separate the four categories of usage reported by the service provider into individual tracking, calculate daily activities locally, or access this data within your own plugins, you need an independent statistical component. The Mu-scorpio/token-usage-counter introduced below is such a plugin. # What is it Its positioning in one sentence: For D
Read Moredsh-balance: Add a floating model balance widget to the DeepSeek Harness Web GUI
dsh-balance is a DSH Web GUI client plugin designed to persistently display model account balances in the bottom-right corner of the page. It supports four service providers: DeepSeek, SiliconFlow, Moonshot, and OpenRouter, automatically identifying and invoking the corresponding balance interface based on the baseURL domain. Core features include a draggable floating card (displaying total balance, top-up/reward details), automatic refresh every 60 seconds (with a 30-second host-side cache), and a global toggle in the settings page (taking effect immediately). API keys can be provided in two ways: via the settings page or the credentials service. All balance queries are performed on the host side, ensuring keys are never sent to the browser. The plugin has zero runtime dependencies and requires no build process. API routes are protected by loopback and same-origin policies.
Read Moredsh-jina: Integrating jina-cli's 12 API tools into DeepSeek Harness
# Introduction DSH's built-in `web_search` covers general web retrieval; however, to obtain arXiv official direct links, SSRN paper originals, or perform web screenshots, text vectorization, and document reordering, one typically needs to write glue code at the host level and register the tools manually. dsh-jina eliminates this step: it encapsulates the Jina AI API capabilities (behind `jina-cli`) into 12 `jina_*` model tools within DSH and provides an API key configuration card with real-time balance checking on the Web settings page. Below, we introduce its features, installation methods, and selection considerations. # What is dsh-jina dsh-jina is maintained by minatoAI, licensed under MIT, and is currently at version 0.5.2.
Read Moredsh-peak-cost-mode: Automatic traffic saving plugin for DeepSeek peak-valley pricing periods
dsh-peak-cost-mode is an open-source DSH web plugin (MIT, author moon09300731) designed for DeepSeek V4 peak-valley pricing: peak output prices on weekdays from 9 AM to 12 PM and 2 PM to 6 PM are double those of off-peak hours. During peak hours, the plugin automatically injects a "Peak Cost Mode" prompt segment, instructing the model to respond in a "caveman" ultra-concise style—discarding pleasantries and hedge words while preserving code, commands, paths, and other technical content word-for-word, without compressing safety warnings. In off-peak hours, users can manually enable active cost-saving mode via the savings report, and the switch persists across peak/off-peak cycles and after restarts. The plugin intercepts every streaming call to estimate savings based on actual output tokens and a compression ratio (default 0.6), converting the savings into CNY and categorizing them into peak and active types. Data persists per wave.
Read More