evo-subagent: DeepSeek Harness Plugin: Role-Based Subagent Routing and Evolution
The `evo-subagent` plugin for DeepSeek Harness (DSH) is designed to address pain points in routing and experience accumulation in multi-agent systems. It supports creating and managing role-based subagents, using an evolution mechanism to maintain `prefercmd.md` and `memory.md`, thereby accumulating reusable commands and experience while saving tokens. The plugin supports binding and routing, strict validation, workspace isolation, and `spawn`/`fork` execution modes, and provides security isolation mechanisms. Installation requires DSH `0.1.5-rc.3` and Node.js `>=24.2.0`. Users can configure Markdown binding files to define roles, utilize built-in roles (such as the code reviewer) and `/evo-route-doctor`, among others.
Read Moredsh-plugin-rag: DeepSeek Harness Local Knowledge Base Retrieval Plugin
The DSH Harness framework provides the `dsh-plugin-rag` plugin, designed to address local codebase retrieval needs for agents while avoiding the latency, cost, and privacy risks associated with external APIs. The plugin supports indexing Markdown, code, and configuration files, using a purely local TF-IDF algorithm for keyword matching without additional dependencies. Once installed and enabled, it can search files locally or display index statistics through the `rag` command, making it suitable for agent scenarios that require decision-making based on local data.
Read MoreDeepSeek Harness Context Compression Plugin: dsh-plugin-context-compressor
`dsh-plugin-context-compressor` is a context management plugin for DeepSeek Harness, designed to address high token consumption and truncation issues in long conversations. It intelligently filters and retains key decisions, errors, and preferences, removes redundant or failed records, and summarizes the remaining content in a structured format. Users can trigger compression with specific commands, reducing costs while improving the agent's reasoning efficiency.
Read MoreDeepSeek Harness Code Review Plugin
`dsh-plugin-code-review` is a structured code review plugin designed for DeepSeek Harness to enhance an agent's review capabilities. It teaches systematic review steps covering correctness, security, performance, style, and testing, helping agents identify logic errors, security vulnerabilities, and coding issues. Users can integrate the plugin into DSH with an installation command and then simply issue a review request in the chat. The agent will then automatically generate results with severity labels, helping developers quickly pinpoint issues.
Read Moredsh-plugin-auto-docs: Automatic Documentation Generation Skill for DeepSeek Harness
`dsh-plugin-auto-docs` is an automated documentation generation skill for DeepSeek Harness, maintained by developer YYTbit. It aims to reduce the burden of manual project documentation maintenance by using intelligent agents to automatically extract information from the project structure, code signatures, and Git history, and generate standardized READMEs, API documentation, change logs, and JSDoc comments. The plugin significantly lowers the manual cost of documentation maintenance through automated workflows.
Read Moredsh-plugin-agent-dashboard: DeepSeek Harness Multi-Agent Status Dashboard
dsh-plugin-agent-dashboard is a multi-agent dashboard plugin for DeepSeek Harness, designed to address the challenge of monitoring multi-agent parallel execution. It allows agents to output structured status reports through instructions, including runtime status, task dependencies, progress, and time information, without requiring a direct web interface. Maintained by YYTbit and released under the MIT license, the plugin can be installed via plugin management commands and is suitable for scenarios that require real-time monitoring and reporting of agent progress.
Read MoreDeepSeek Harness SQLite Data Tools Plugin
`dsh-tool-sqlite` is a zero-dependency plugin for DeepSeek Harness (DSH) designed to handle local `.db` files. Leveraging built-in Node.js modules, it provides core features such as listing files, inspecting table structures, and executing read-only SQL queries. To ensure security, the plugin enforces strict restrictions: only read-only operations are allowed, SQL statements are whitelisted, and path boundaries and output row counts are limited. Developers can easily enable it in DSH configuration to explore database data directly within the workspace.
Read Moredsh-token-diet: A Pre-Use Token-Saving Toolkit for DeepSeek Harness
`dsh-token-diet` is a toolkit in DeepSeek Harness (DSH) for handling large-volume inputs, reducing token consumption at the source through structure-preserving summarization. It provides pure-function tools for text, JSON, and CSV, and supports configuring three presets or a manual compression ratio via the Web UI or settings. The plugin provides real token-saving feedback statistics, features a pure-function safety model, requires Node.js 22+, and can significantly improve context window utilization efficiency in long conversations.
Read MoreDeepSeek Harness Plugin dsh-skillopt: Nightly Self-Check and Skill Consolidation for Intelligent Agents
dsh-skillopt is a plugin for DeepSeek Harness, designed to enable self-evolution in agents. It integrates the SkillOpt-Sleep engine, allowing agents to review past sessions offline, replay tasks within their own API budget, and optimize skills. The plugin provides a full closed loop covering data collection, dry runs, execution, and consolidation, with tools such as `skillopt_status` and `skillopt_run`, and supports scheduled nighttime tasks. Through read-only data collection and isolated verification mechanisms, it helps developers improve agent performance without incurring additional costs.
Read Moredsh-lan-share: A LAN File Sharing Plugin for DeepSeek Harness
`dsh-lan-share` is a local file-sharing plugin for DeepSeek Harness (DSH). It uses native Node.js modules to provide file browsing, download, and upload services over HTTP within a local area network. The plugin has zero dependencies, supports token or fixed-password authentication, and includes security features such as read-only mode, path boundary protection, rate limiting, and automatic hiding of sensitive files (e.g., `.env` files and keys). Users can enable it via a configuration file and control the service using commands such as `lan_share_start`, `lan_share_status`, and `lan_share_stop`.
Read Moredsh-plugin-login-gate: Add Password Login and HTTPS Reverse Proxy to DeepSeek Harness
The DeepSeek Harness (dsh) plugin adds a password login gate and an HTTPS reverse proxy to the Web UI, addressing the security and compatibility issues caused by the default loopback-only access and cross-device LAN access. It supports single-password login, self-signed certificate encryption, automatic protocol redirection, and full forwarding of streaming output. For LAN access, the plugin fixes issues such as 403 errors, an unavailable settings page, and WebSocket handshake failures by rewriting the Host header and applying patches to support non-loopback IP addresses. To use it, configure the password file, generate the certificate, and restart the service. Note that the password is stored in plaintext, and `lanPatch` modifies the dsh source code; after updates, the patch script must be re-run.
Read MoreDeepSeek Harness Auto-Approval Plugin for Privilege Escalation Requests
`dsh-approve-for-me` is a plugin for DeepSeek Harness used to automatically handle privilege elevation requests. It allows users to configure a Reviewer model to automatically approve low-risk requests in the background, or handle high-risk requests according to the selected mode (Accept Only or Accept & Refuse). Installation requires running a script or manually modifying the configuration, and the service must be restarted for the plugin to take effect. When installing, care must be taken with the Bundle order, and if no model is configured, the plugin falls back to manual approval.
Read MoreSliding Rheostat Voting and Trend Viewing Plugin
`water2004/dsh-liang-transformer-vote` is a DeepSeek Harness (DSH) web plugin that integrates with the “Sliding Transformer” voting system. It synchronizes the status of an external voting website to provide voting controls, historical trend charts, and data-unlocking features in the DSH interface. The plugin supports automatic refreshing every 30 seconds and includes an API rate-limiting mechanism. Installation requires source-site deployment, Cookie configuration, and environment prerequisites such as Node.js.
Read Moredsh-tool-vision: DashScope Image Description Tool for DeepSeek Harness
`dsh-tool-vision` is an open-source plugin in the DeepSeek Harness (DSH) ecosystem, designed to introduce visual capabilities to text models. It leverages DashScope's OpenAI-compatible interface and uses the `image_describe` model-side tool to directly read local images and generate descriptions. The plugin offers two core features: it enables the model to invoke the tool with a path parameter to analyze images, and it provides a "paste bridging" capability that converts pasted images into path text, thereby bypassing the host's image access restrictions. Users must configure their own DashScope API key, and after installation, the tool will appear in the model's tool catalog. Licensed under the MIT License, it is suitable for scenarios that require handling local visual context in conversational flows.
Read Moredsh-reminder: Task Completion Alert Sound Plugin for DeepSeek Harness
The dsh-reminder plugin adds a real-time auditory feedback mechanism to DeepSeek Harness (DSH), resolving the lack of feedback for agent tasks. It reports task lifecycle events via sounds and desktop notifications (supporting Windows, macOS, Linux, and WSL), with a three-level audio fallback mechanism. After installation, users can configure sound packs, volume, and mute windows in the Web interface, taking effect in real time. By default, sounds play only when a task completes or terminates abnormally, and permission and dependency checks are supported.
Read Moredsh-thread: Session Memory Plugin for DeepSeek Harness
The DeepSeek Harness (DSH) plugin `dsh-thread` provides a deep conversational memory solution, using an embedded SQLite database to store event streams and ensuring that coding agents maintain contextual coherence over long-duration interactions. It delivers core capabilities such as lossless capture, structured state injection, deterministic BM25-based retrieval, behavior-contract skills, and explicit decision/preference logging. The plugin supports session isolation, resource cleanup, and wrap-up deposition, helping agents overcome memory fragmentation and achieve accurate, coherent long-running operation.
Read Moredsh-plugin-wecom-bot: Integrate DSH Agent via WeCom Long-Lived Connection
The DeepSeek Harness plugin `dsh-plugin-wecom-bot` provides an interaction method based on WeCom long-lived connections, addressing the pain points of traditional setups that require public URLs, encryption and decryption, and allowlists. It supports streaming replies, zero public-network dependencies, session management, and forwarding of permission approvals. Users must install dependencies and configure the BotID and Secret, then establish a connection via WebSocket. The plugin supports both `per-user` and `per-message` modes, allowing users to switch working directories, manage sessions, and approve permissions by replying to commands directly in WeCom. Known limitations include connection count limits and unsupported message types.
Read Moredsh-maestro-runtime: DSH Runtime Plugin Compatible with Maestro-Flow
`dsh-maestro-runtime` is a DSH plugin designed to adapt the Maestro-Flow runtime to DSH sessions. Core features include Guard security protection, Context injection, Onboarding guidance, background Knowledge Graph (KG) synchronization, Delegate notification handling, Team collaboration heartbeat logging, and Coordinator bridging. The plugin is installed and mounted via pnpm, covering P0+P1 scenarios to enable seamless integration and deep interaction.
Read MoreDeepSeek Harness Plugin: zhangzifei/dsh-model-aliases
The DeepSeek Harness plugin dsh-model-aliases simplifies model switching by providing persistent model aliases. It allows a provider, model, and reasoning level to be bound to easy-to-remember names, enabling real-time two-way synchronization between the alias selector and native controls. The configuration includes fault tolerance and validation mechanisms, and is compatible with DeepSeek Harness Web 0.1.2-rc.1. Users can enable it with an installation command, configure it in Settings, and then quickly switch models in the chat interface, making it ideal for users who frequently adjust model parameters.
Read MoreDeepSeek Visual Plugin
The DeepSeek Harness plugin intercepts messages containing images and converts images that pure text models cannot process into textual descriptions generated by a vision model, enabling it to interpret images. It allows multimodal models to pass through directly, while pure text models automatically invoke a vision model to translate them. Users must deploy the code to `$DSH_HOME/profiles/node_modules` and configure the model and API key in `cordis.patch.yml`. The plugin covers session logs and tool execution, supports deployment on Windows, Linux, and macOS, and includes a self-check script.
Read Moreqmd-autosearch: Automatically Augmenting QMD Semantic Search When Models Run grep/glob
qmd-autosearch is a plugin for DeepSeek Harness (DSH) that aims to address the lack of semantic understanding in traditional `grep`/`glob` searches. When a model runs searches in a knowledge base directory, the plugin automatically triggers QMD semantic retrieval and injects the results into the model context. Its core features include: LLM-reranked lex+vec queries; prioritization of the latest user message as the semantic query term; asynchronous result injection that does not block the workflow; and duplicate-prevention and fault-tolerance mechanisms. Users must configure the QMD service address and collection list. The plugin does not require manual invocation; once installed, it automatically supplements knowledge base searches with relevant document results, helping the model retrieve semantically related file content.
Read MoreDeepSeek Harness Skill Management Plugin: dsh-skills-manager
dsh-skills-manager is a client plugin for DeepSeek Harness designed to address the difficulty of locating fragmented skills. By reading the host registry, it provides a unified skill management view on the settings page. The plugin supports viewing the skill list, real-time search and filtering, reading details through a detail panel, and one-click copying, with automatic refresh and theme adaptation capabilities. Currently, it is used only as a read-only management interface, helping developers centrally browse and manage large numbers of Agent skills.
Read Moredsh-local-file-upload: Uploading Local Files in Web Composer
`dsh-local-file-upload` is a client plugin for DeepSeek Harness designed to address the issue of uploading local files to the Session workspace in Web Composer scenarios. The plugin supports uploading multiple files via a toolbar button or paste (10 MiB per file), automatically saves uploaded files to the `.dsh-uploads/` directory, and generates reference markers. It enforces security restrictions by allowing only local transfer and rejecting symbolic links. Uploaded files must be cleaned up manually, and it is recommended to add them to `.gitignore`.
Read MoreDeepSeek Harness Vision Bridge Plugin: Routing Image Messages to OpenAI-Compatible Endpoints
The `zh-u-hb/dsh-vision-bridge` plugin for DeepSeek Harness (DSH) listens to the DSH message pipeline to address the issue that text-only models cannot process images directly. When it detects that the active model does not support images, the plugin forwards the image to a user-configured OpenAI-compatible vision model and substitutes the returned text back into the conversation flow. It allows users to configure the API endpoint, API key, and model ID via a web page, and automatically assembles prompt and image data. Note that it supports only the OpenAI protocol and handles image messages only at specific pipeline stages.
Read Moredsh-gesture-mouse: A Camera Gesture Mouse Plugin for macOS
`dsh-gesture-mouse` is a macOS gesture-mouse plugin for DeepSeek Harness (DSH). It performs visual recognition via a local Swift Helper and leverages Vision models to support actions such as index-finger movement, thumb-pinch clicks, finger scrolling, and an open-palm safety stop. The plugin includes a strict safety cooldown mechanism and supports installation and configuration via the DSH CLI. It requires macOS 13 or later, and users must grant camera and accessibility permissions. Currently in V0, it depends on real-person calibration and is best suited as an auxiliary input component for DSH. It does not support multiple hands or screen recording permission.
Read More