dsh-model-router：DeepSeek Harness 任务感知型模型路由
DeepSeek Harness 插件 `dsh-model-router` 旨在实现“一切皆插件”，通过任务感知智能路由模型。它根据任务类型（如研究或闲聊）和模型能力，自动调度请求至最优模型，优先保证研究质量或使用免费模型处理闲聊。该插件支持 Grok 限制、速率限制处理、子代理及模式切换，提供 dryRun 验证和详细日志，有效简化了智能体模型维护。
Read Moredsh-ov-memory: Integrating OpenViking Persistent Memory into DeepSeek Harness
dsh-ov-memory is a plugin that integrates OpenViking persistent memory into DeepSeek Harness. With features such as auto-recall, session mirroring, threshold-based commit, and offline replay, it bridges the gap between conversational flow and memory. The plugin supports MCP tool bridging and the `/memlearn` command, helping developers build agents with long-term memory and context reuse capabilities. Installation requires DSH 0.1.x and a specific Node version, and namespace conflicts should be considered when using it.
Read MoreIntroducing the dsh-seed-society Plugin
`dsh-seed-society` is a workflow plugin for DeepSeek Harness (DSH) built around the core philosophy of “everything is a plugin.” It encapsulates the Agent Society based on the Yogācāra eight-consciousness architecture, memory-consolidation tuning, and the MCP tool bridge as independent components. The plugin includes four core parts: fixing the deep-thinking model configuration, solidifying memory-consolidation parameters, establishing the MCP toolchain, and integrating six seed skills. Installation requires Python 3.10+ and the relevant dependencies. The official recommendation is to install it with a single `dsh plugin` command. After installation, restart DSH to activate the plugin, then verify the MCP tools and memory status to ensure the system supports the eight-consciousness architecture and dual-loop runtime capability.
Read MoreMemoBranch: Git-Native, Auditable Long-Term Memory for AI Agents
MemoBranch is a local-first long-term memory layer that transforms agent memory into a Git-native, version-controlled Markdown wiki. Through an evidence-driven mechanism, it organizes conversations and candidate knowledge into auditable, traceable canonical memory, supporting cross-source compilation, hybrid retrieval, and atomic commits. Core features include source-backed answers, structural and semantic maintenance, envelope encryption, and server-side access control. It is designed for production-grade agent scenarios that require high reliability and rigorous knowledge management, ensuring clear memory provenance and traceable rollback of logical changes.
Read MoreDeepSeek Harness WeChat Integration Plugin (dsh-wechat-suite)
The `dsh-wechat-suite` plugin for DeepSeek Harness (DSH) uses the Tencent iLink protocol to connect a DSH Profile to a WeChat personal account, enabling bidirectional communication. It includes `wechat-gateway` and `wechat-conversation-node`, supporting multimodal message transmission for text, images, and voice, as well as OCR, STT, and TTS capabilities. It features automatic reconnection, message deduplication, input indicators, and conversation control, and provides a management console. Installation requires Node.js 22, pnpm, and a dedicated WeChat account, but has a single-instance limitation and a risk of account ban; specific credentials must be configured.
Read MoreDeepSeek Harness Sentry Plugin
`dsh-sentry` is an open-source DeepSeek Harness plugin designed to resolve LLM context overflow caused by overly large Sentry error logs. The plugin connects to the Sentry API in read-only mode and applies strict structured pruning to the returned data, ensuring that agents can access debugging information without exceeding token limits. It supports retrieving project lists, searching issues, and reading pruned stack traces and event details. To use it, a user authentication token and specific permission scopes must be configured, and a strict data pruning strategy must be followed (removing sensitive information and limiting the number of frames and source context) to fit agent workflows.
Read Moredsh-action-outbox: Batch Review of DeepSeek Harness Tool Side Effects
`dsh-action-outbox` is a batch review inbox plugin for DeepSeek Harness, designed to address the irreversibility of external write operations performed by agents. It allows users to locally stage tool calls before executing tool side effects, enabling full inspection, editing, and review via a Web interface. Staged calls are submitted for execution only after explicit approval. Core features include a visualized inbox, parameter editing and substitution, fault-tolerant handling, and persistent recovery across restarts. By configuring `cordis.patch.yml`, it implements zero-target dispatch, TOCTOU protection, and restart re-authorization security semantics, ensuring the rigor and revocability of agent operations.
Read Moredsh-pdf-reader: A Content-Aware PDF Reading Plugin for Vision Models
The DeepSeek Harness plugin `dsh-pdf-reader` is designed to address the loss of detail caused by token limits when visual models process PDFs. Built on PyMuPDF, it automatically identifies page content types: it extracts plain text from text-dense pages and renders high-DPI images for pages with dense charts, tables, or formulas. It provides three core tools: `pdf_scan` (overview), `pdf_read_page --mode mixed` (mixed mode), and `pdf_render_region` (region refinement). The plugin supports a "preview-content-refine" workflow, intelligently caches images to save tokens, and is suitable for processing documents with complex layouts and formulas.
Read Moredsh-business-days: Calculating Business Days
The dsh-business-days plugin for DSH is designed to provide agents with native business-day calculation capabilities, supporting the exclusion of weekends and holidays based on dates to help build smarter business workflows. Maintained by istone, it is implemented in pure Node.js, does not rely on external network interfaces, and supports offline use. Installation is simple, requiring only a single command to enable. Users can directly invoke its tools within a conversation. Since the plugin runs with process-level privileges, it is recommended to review the source code and license before use.
Read MoreIntroduction to the DSH Plugin: satan9394/dsh-git-guardrails
satan9394/dsh-git-guardrails is a security plugin for DeepSeek Harness (DSH) agents, designed to provide safety guardrails for their Git operations. Drawing on the design philosophy of well-known skill plugins, it intercepts dangerous commands to prevent code repositories from accidental misuse or malicious damage. It is suitable for scenarios where agents need to be granted Git permissions while their behavior is strictly constrained. Users should consult the official repository for installation instructions and, since the license information is not specified, carefully audit the source code for security before installing it.
Read Moredsh-memory: Persistent Memory Storage for DeepSeek Harness
`dsh-memory` is a community plugin for DeepSeek Harness designed to address the lack of persistent cross-session memory for agents. The plugin enables agents to store, retrieve, delete, and search facts using stable key-value pairs, with tag support. It provides four core tools: `memory_set`, `memory_get`, `memory_delete`, and `memory_search`. Installation is simple and requires no build step, relying on a storage backend provided by the environment, such as SQLite or JSON. This tool is well suited for scenarios where specific state or a knowledge base must be maintained across multi-turn conversations.
Read Moredsh-statecore: Native Memory Plugin for DeepSeek Harness
`dsh-statecore` is a native memory plugin for DeepSeek Harness, designed to address the fragmentation and limitations of traditional MCP servers in session memory management. By mounting the StateCore memory engine into the DSH plugin context, it enables automatic memory ingestion, injection, and persistence. The plugin supports automatic identification of real conversations, summary compression protection, a web-based management interface, and native tool calling, while sharing storage with Claude Code. Built on local SQLite storage and released under the MIT License, it provides agents with an auditable, tightly integrated long-term memory solution.
Read Moredsh-water-heater: Water Heater Capacity Selection Tool
dsh-water-heater is a DSH plugin maintained by uckkk for calculating water heater capacities in liters. It is implemented in pure Node.js, requires no network dependencies, supports precise calculations based on input parameters, and returns detailed results. Users can install and enable it using the terminal command `dsh plugin add github:uckkk/dsh-water-heater`, with no additional configuration needed. The plugin provides agents with foundational capabilities for common-sense everyday calculations, but it is recommended to review the source code and license (MIT) before installation.
Read Moredsh-study-plan: Exam Prep Time Allocation Tool
The core philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” `dsh-study-plan` is an exam-preparation time allocation tool maintained by `uckkk`, implemented in pure Node.js for precise calculations without network dependency. The plugin registers the `study_plan` tool, enabling parameter-based calculations with detailed results, and is well suited for rapidly integrating exam-planning capabilities into the DSH ecosystem.
Read Moredsh-highway-drive: High-Speed Driving
The DeepSeek Harness plugin `dsh-highway-drive` provides query capabilities for highway driving tips and safety considerations. The plugin is released under the MIT license, implemented in pure Node.js, and requires no network dependencies. Developers can install it via a terminal command and directly invoke its `highway_drive` tool in a session. Before installation, ensure that the environment meets the relevant dependency requirements and pay attention to permissions and security.
Read Moredsh-dehumidify: Dehumidifier Selection
The DeepSeek Harness (DSH) plugin `dsh-dehumidify` is designed to assist with home appliance selection, particularly dehumidifier model selection. By accurately calculating parameters such as dehumidification capacity, it provides clear selection recommendations. The installation command is `dsh plugin add github:uckkk/dsh-dehumidify`, and it supports directly invoking the `dehumidify` tool in conversations. Built with Node.js and without network dependencies, it is suitable for complex requirements involving rapid model comparison and calculations.
Read Moredsh-curling: A Plugin for Providing Hairstyling Solutions to Agents
The DeepSeek Harness plugin dsh-curling is a “fun restyling” plugin focused on hair styling solutions. It does not rely on external network access, is implemented in pure Node.js, and returns specific restyling references based on parameters. It requires the Cordis and DSH tool dependencies before use and can be invoked in sessions after installation. This plugin is suitable for scenarios where agents need to be configured with specific hairstyle logic; the associated permissions and licensing should be noted.
Read MoreDeepSeek Harness Plugin: dsh-baijiu-pick
The DeepSeek Harness (DSH) plugin `dsh-baijiu-pick`, maintained by developer uckkk, is dedicated to providing parameter-based baijiu selection recommendations to agents. It is implemented using a pure Node.js tech stack and does not rely on external network requests, addressing the need to obtain clear recommendations in specific scenarios. It can be installed with `dsh plugin add github:uckkk/dsh-baijiu-pick`, after which it can be invoked directly in a DSH session. Note that it runs with the permissions of the current process. We recommend reviewing the source code and the MIT license before use.
Read Moredsh-standard-toolkit: DeepSeek Harness Tool Manager
`dsh-standard-toolkit` is a plugin for DeepSeek Harness (DSH) designed to address the chaos in agent tool management, centered on the philosophy that "everything is a plugin." The plugin manages tools through a "tool staging area," loading them on demand only when triggered and automatically destroying them afterward, thereby saving tokens and preventing context pollution. It supports session isolation, two-level trigger-word matching, and security review mechanisms. Additionally, the plugin allows the model to create new tools on the fly via `register_new_tool` and provides an installation script to simplify the deployment process.
Read Moredsh-project-progress: Project Progress Documentation Shared in the Workspace
dsh-project-progress is a DeepSeek Harness plugin that creates a shared `PROGRESS.md` document in the DSH workspace to manage project status and share context across sessions. The document is divided into two sections—project profile and progress timeline—and supports read/write operations through three tools: `project_update`, `project_facts`, and `project_status`. The plugin runs locally with no network calls, automatically appending progress only in top-level sessions, making it suitable for developers who need to maintain context across multiple sessions.
Read MoreIntroduction to the DeepSeek Harness (DSH) Persistent Memory Plugin
The `disc0nct/dsh-memory-plugin` plugin for DeepSeek Harness (DSH) addresses the challenge of context loss caused by session isolation in agent development. It provides persistent storage, hybrid retrieval, and categorized management, supporting the invocation of seven tools (such as storage, search, and statistics) via `ctx.tools` to manage cross-session memory. The plugin includes atomic writes, concurrency safety, and data recovery mechanisms, supports up to 5,000 memories, and effectively improves the coherence of long conversations.
Read Moredsh-research-first: Let DSH agents research first, then act.
**Introduction** A common pitfall when using coding agents to modify code is acting without investigation: failing to check current API signatures in documentation, reviewing existing implementations in the repository, or checking if the community has encountered similar issues, and resorting to trial and error after writing the code. One remedy is to hard-intercept before tool calls, forcing an investigation first. However, the investigation itself might be impossible—due to offline environments, network restrictions, or the inability to fetch documentation—causing the task to stall. *dsh-research-first* takes a different approach: it does not block by default. Instead, when an agent attempts to write, modify, or execute code without having investigated, it injects a model-visible reminder into the session and logs it. If investigation tools fail, it alerts the user, allowing them to decide whether to proceed. The following sections introduce the functionality, installation, and configuration of this plugin. **What is dsh-research-first**
Read Moredsh-cursor-acp: Enables DSH chat agents to delegate independent tasks to the local Cursor CLI
dsh-cursor-acp is a DSH community plugin maintained by loeanxi (MIT license, v0.2.0). It registers a `cursor_agent` tool, enabling the agent in the DeepSeek Harness chat to delegate well-defined independent tasks (such as implementing features or reviewing files) to a locally logged-in Cursor CLI (via ACP) for execution. The prerequisites are to install and log in to the Cursor CLI (or use `CURSOR_API_KEY`), and to set proxy environment variables in the same terminal window that will be used (browser proxies do not work for terminals). After installing via npm or a one-line command from GitHub, restart DSH, and go to Settings → Cursor subagent.
Read MoreDHS-multi-agent-plugin: Integrating multi-agent orchestration into DeepSeek Harness
# Preface When writing agents using DSH (DeepSeek Harness), a single agent loop can cover most linear tasks. However, once tasks become more complex—such as researching first, then reviewing, and finally summarizing—you must handle task dependencies, parallel scheduling, and failure recovery yourself. Scattered logic is difficult to test and reuse. Below, I introduce Li3NGa's DHS-multi-agent-plugin, which transforms this orchestration infrastructure into a standalone plugin. # What is it? DHS-multi-agent-plugin is a DeepSeek Harness plugin. Its README positioning is succinctly defined as: "Turning DeepSeek Harness into a powerful multi-agent orchestration engine." The repository is maintained by Li3NGa.
Read Moredsh-experts: Represent the expert team as a data catalog and automatically register as a routable dsh skill
# Introduction A common challenge in agent orchestration is that expert personas, division of labor, and reporting rules are scattered across prompts and code; altering the review panel necessitates modifying the engine. The philosophy behind dsh (DeepSeek Harness) is "everything is a plugin." Following this logic, libinyam/dsh-experts transforms the "expert team" into a user-editable data directory: each team is automatically registered as a model-routable skill (`experts-<team-name>`), decoupling the engine from the lineup. The following sections introduce its design, installation, and usage. # What is it dsh-experts is a multi-expert plugin for dsh maintained by libinyam, released under the MIT license. One-sentence positioning: Translating the expert persona list into native dsh skills—
Read Moredsh-lcx-codex: Completing Native Compact, Hosted Search, and Prompt Cache for DSH's GPT Responses path
When developing agents using the GPT/OpenAI Responses route in DeepSeek Harness (DSH), if the compression, Replay, and search stages do not align with the provider's native capabilities, issues such as mid-way implementation changes and repeated cache invalidations are easily encountered. DSH's philosophy is "everything is a plugin," and these gaps are perfectly suited to be filled by plugins. The `dsh-lcx-codex` introduced below performs exactly this: it takes over the DSH GPT Responses path to provide Native V2 Compact/Replay, Hosted Search, Prompt Cache, and continuous web operations.
Read MoreBigLead: B2B Lead Mining in DSH
BigLead (kobenfang/biglead) is a DSH skill dedicated to B2B lead generation. It packages the standardized workflow of "searching for target companies—multi-channel cross-verification—extracting contact information—managing lead databases" into a plugin, suitable for B2B sales, market research, and competitive analysis. It adheres to the Agent Skills open standard and is compatible with frameworks such as OpenClaw and Claude Code. It requires Python 3 and internet search capabilities to operate.
Read Moredsh-rich-artifacts: A DSH plugin to publish workspace files as chat Artifacts
dsh-rich-artifacts is a DSH plugin developed by Inkotake (MIT licensed) designed to solve the inconvenience of users having to manually view files on the host machine after an agent delivers them. It publishes workspace files as chat Artifacts: the model can render PNG/JPEG/WebP/GIFs inline in the conversation by calling the `publish_artifact` tool (with parameters including path, title, and display mode), while other files are generated as cards with a download button. Files are stored in an ArtifactStore based on SHA-256 content addressing, with metadata stored in `~/.dsh/artifacts/v1/`. Persistent session events only record metadata without storing the file body, ensuring files remain accessible correctly after reloading or forking historical sessions.
Read Moredsh-pain-point-check: forcibly inject the dsh guard plugin for Three Questions when two experiments do not converge.
dsh-pain-point-check is a mandatory guard plugin for the DeepSeek Harness (dsh) maintained by ICCuse. It addresses the pain point where agents lose meta-cognition after entering the "problem-solving state" and repeatedly attack the same problem from the same direction (confirmation bias, sunk cost, narrative loops). Unlike the official `repeat-tool-reminder`, which is advisory only, this plugin adopts a veto-style gating mechanism: after two experiments on the same question fail to converge, it utilizes `tools/result` (to track failures and repeated calls), `agent/pre-step` (to reset counters when the user asks a new question), and `tools/pre-execute` (to reject all non-investigative tool calls, while allowing `read`, `grep`, `web_search`, `ask_user_question`, etc.).
Read Moredsh-playwright-cli: Integrating Playwright CLI into DSH's Agent Loop
dsh-playwright-cli is a Host-only plugin for DSH agents designed to encapsulate common Playwright CLI operations as tools callable by the agent, thereby implementing automated testing workflows. The plugin provides four core functionalities: `playwright_version` probes the local CLI status; `playwright_install` automatically installs browsers and missing test dependencies; `playwright_test` runs test suites, supporting various parameter configurations; and `playwright_show_report` generates and returns the route or file path to the HTML report. The plugin runs based on DSH's shell executor, inheriting its sandbox strategy, timeout control, and environment variable configuration.
Read Moredsh-layered-memory: DSH Cross-Session Long-Term Memory Plugin
dsh-layered-memory is a cross-session long-term memory plugin for DeepSeek Harness (DSH), licensed under the MIT License, designed to address the issue of context loss following a session. The plugin persistently stores verified facts, task experiences, and environmental information on disk, supporting cross-session retrieval. Core features include L1-L3 layered memory, namespace isolation, BM25 full-text search, content deduplication, and cross-namespace experience enhancement. The plugin provides 14 memory tools, such as `memory_write` (mandatory evidence verification), `memory_search`, and `memory_index`, and supports progressive tool exposure. Consistency maintenance relies on deduplication, manual updates, and candidate merging; the system does not automatically detect contradictions, and it is strictly prohibited to physically discard verified facts, allowing only compression.
Read Moredsh-teacher: DSH plugin, turning agents into Socratic teachers
dsh-teacher is a plugin-based teaching tool within the DSH ecosystem, designed to transform traditional "direct instruction" into "Socratic questioning guidance." The plugin loads courses from Markdown question sets and persists data in SQLite. Its core strategy is "ask without answering," guiding user thinking by hiding answer keys while recording "knowledge gaps" in real-time during sessions. It integrates the FSRS-5 spaced repetition algorithm to perform on-demand retesting of overdue knowledge gaps, achieving cross-session long-term memory consolidation. Furthermore, it provides immediate quiz popups (supporting multiple choice and fill-in-the-blank) without requiring LLM involvement, followed by an LLM-based scoring and Socratic review of error points after the quiz. The Web client supports viewing the question bank, the knowledge gap panel, and status. Installation requires DSH rc.6+ and Node >= 22.
Read MoreConnect OpenMAIC Classroom to DeepSeek Harness with dsh-openmaic
### 版本1（正式技术文档风格）： DSH-OpenMAIC is maintained by THU-MAIC from Tsinghua University. It integrates OpenMAIC with DeepSeek Harness: a single request can generate a usable classroom link, and the slides, interactive components and teaching cards written by the intelligent agent can be rendered in real time within the conversation. This article verifies the installation commands, four tools, Socratic skills and configuration items against the catalog page and GitHub repository. ### 版本2（更贴合开源项目介绍口吻）： Maintained by THU-MAIC of Tsinghua University, DSH-OpenMAIC connects OpenMAIC to DeepSeek Harness. With it, one single prompt can spawn a fully functional classroom link, and the slides, interactive widgets and teaching flashcards generated by the AI agent can be rendered on-site right in the chat interface. This paper cross-checks the installation commands, four core tools, Socratic questioning skills and configuration items against the project catalog page and GitHub repository.
Read More