dsh-bio-human-microbiome-function-mining：人体微生物组目标酶功能挖掘工作流
Liuyixin-lily 维护的 DeepSeek Harness 插件 human-microbiome-function-mining，将目标酶/EC/底物-产物反应标准化为 Target Function Card，经 CD-HIT、DIAMOND、US-align 与条件性 DEER、FedKEA、文献审计等步骤，输出带 overall_rank（默认 TM-score 优先）的候选证据表与实验设计。内置 9 个 skill 与脚本/schema，不打包外部生信工具。安装使用官方 npx @deepseek-ai/dsh plugin add 命令；适用已有候选蛋白组或需 genome 子流程的用户。MIT 0.1.0。
Read Moredsh-memory-fitting：先立靶子再问，收敛后提案待确认
`dsh-memory-fitting` 是一款 DeepSeek Harness (DSH) 插件，用于解决模糊指令痛点。它采用“先立靶子再问”的策略，通过预测 3-5 个意图方向并批量提问，引导用户确认，直至收敛后生成提案。该插件支持本地留档、数据导出训练及安全记忆写入机制，需手动启用。
Read MoreDeepSeek Harness Off-Peak Job Plugin
`dsh-offpeak-job` is an off-peak scheduling plugin for DeepSeek Harness (DSH), designed to reduce costs and avoid peak-time usage when developing intelligent agents. The plugin allows users to select “Off-peak run” in Composer to send a message; the system will then queue it and automatically execute it within the configured off-peak window. Key features include provider-specific peak and off-peak scheduling, custom run windows (e.g., daily or weekdays), pause policies, and a nightly upgrade policy. Configuration and state are stored locally and do not involve telemetry. Installation requires running the official command and restarting the DSH process, and the host must remain powered on during off-peak periods for execution.
Read Moresimple-plan: DeepSeek Harness Plan-Execute-Verify Workflow Plugin
The `simple-plan` plugin for DeepSeek Harness is designed to address issues in multi-agent collaboration, such as coarse task decomposition, lack of independent verification of deliverables, and an uncontrollable repair process. It enforces a Plan-Execute-Verify-Fix loop to transform complex tasks into controllable streams of subtasks. The plugin consists of five core stages: breaking the request down into 2–7 steps, executing the steps with worker subagents, verifying the results with stateless subagents, fixing defects within a maximum of three loops, and finally aggregating the results. It requires Node.js >= 22 and is suitable for complex, multi-step tasks that demand strict quality control.
Read MoreDeepSeek Harness Plugin: dsh-connection-card-host
`dsh-connection-card-host` is a plugin built for DSH to address the lack of a programmatic management layer in multi-session collaborative workflows. It elevates “connections” to first-class citizens and, by defining sessions as nodes, connections as containers, and cards as scopes, provides flexible permission boundaries and resource management for interactions between sessions. The plugin supports multiple interaction modes: automatically pulling the counterpart’s state to enable mutual “seeing” and “conversation,” using mounted “cards” to share tools and dependencies, and providing a “convention box” to manage consensus. Its core mechanism is pull-based awareness, which incurs overhead only when the other side actively queries, enabling efficient isolated collaboration without interrupting sessions.
Read Moredsh-midtalk: In-Task Interjection + Optional Barge-in + Structured Resume Card
`dsh-midtalk` is a Host-side workflow plugin for DeepSeek Harness (DSH), designed to address the operational challenges of "intervening" and "interrupting" during task execution. It implements two slash commands: `/say` for lossless intervention, inserting input at step boundaries; and `/cut` for lossy interruption, which aborts the current step and generates a structured recovery card containing the full state. Installation supports command-line or manual mounting, with no third-party dependencies. This tool helps users fine-grained control over the task workflow and is suitable for intervening at any time during long-running tasks.
Read MoreSpecsRelay for DeepSeek Harness: Turning Conversations into Development Tasks
SpecsRelay is a workflow plugin designed specifically for DeepSeek Harness to address the pain point of manually organizing long conversations during development. It converts DeepSeek conversations into structured development requirements with one click, supporting one-click organization, real-time clarification, continuation of long-running tasks, and reuse of configured models. By default, the plugin is only compatible with the anywhere-labs community edition of DSH Desktop. After installation, users can log in to DeepSeek via the sidebar to discuss, then organize and verify the requirements before sending them to the DSH Agent for execution, significantly improving the efficiency of turning ideas into code.
Read Moredsh-cds-mode: A Working Mode That Truly Brings Multi-Agent Collaboration to Life
CDS Mode is a workflow plugin for DeepSeek Harness, designed to resolve development issues such as unclear requirements and conclusions that are not persisted as deliverables through a multi-agent collaboration model of "one person reports, the team executes." It includes 56 built-in personas, divided into General and Cybersecurity modes, and constrains behavior with nine iron rules and six circuit-breaker mechanisms to ensure rigorous processes. Users only need to interact with the master persona, Chen, who refines requirements through structured questioning, assigns tasks, validates results, and ultimately produces final files persisted to disk. The tool is suitable for scenarios requiring strict process control and multi-role collaboration.
Read MoreIntroduction to the DeepSeek Harness AgentTeams Plugin
The AgentTeams plugin for DeepSeek Harness (DSH) aims to extend agent development capabilities through modularity. It is currently in a basic form, and no official feature list, installation command, or typical usage examples have been provided. Due to the lack of detailed documentation, users must refer to the source code repository themselves for deployment and be sure to verify the license and reliability of the source code.
Read Moredsh-plugin-background-tasks: Asynchronous Execution Plugin for Long-Running Commands
The `dsh-plugin-background-tasks` plugin automatically moves long-running commands to the background after a preset 10-second synchronous window to avoid blocking the session. It supports proactive completion reports, status queries, and controlled termination, ensuring that commands execute safely within the sandbox and approval pipeline. It must be used in combination with `ctx.shell`, the `jobs` runtime, and the approval pipeline. After installation, enabling Background Tasks mode provides a unified operational experience.
Read Moredsh-order-plan: Returns a Hot Pot Meal Plan Based on Parameters
The DSH plugin `dsh-order-plan` is a workflow plugin focused on planning menus for hot pot gatherings. It is implemented purely in Node.js, has no network dependencies, and is maintained by uckkk. The installation command is `dsh plugin add github:uckkk/dsh-order-plan`. It provides the `order_plan` tool, which returns the recommended order for adding ingredients to a hot pot meal after users provide the input parameters. The plugin is suitable for agent workflows, and it is recommended to review the source code and license before use.
Read Moredsh-workout-split: Workout Split Plan
The DSH plugin `dsh-workout-split` is a workflow plugin for programmatically generating training split plans. Maintained by uckkk, it is implemented entirely in Node.js, has no network dependencies, and supports precise parameter-based calculations to return detailed results. Installation requires specific dependency versions; once enabled, its tools can be invoked within a session. Review the source code and MIT license before use.
Read MoreDSH Plugin: Track Driver Development
The Track-driven development feature of the DeepSeek Harness (DSH) plugin aims to improve agent development efficiency. By integrating specifications, execution plans, and quality gates into the development process, it enhances the consistency and controllability of the workflow. The plugin supports a Track-driven development mode, inspired by Agents Conductor. Since installation and detailed usage rely on the GitHub repository documentation, users are advised to consult it directly. When using the plugin, review the source code for security and verify license compliance.
Read MoreIntroducing the DSH Plugin: dsh-mlops (ML Pipeline Workflows)
`dsh-mlops` is a plugin in the DSH ecosystem that aims to simplify the building and maintenance of machine learning pipelines through standardized workflows. It covers the full lifecycle, from data preparation, training, and evaluation through deployment and monitoring. The plugin runs with the permissions of the current DSH process, but official resources currently do not provide a specific feature list, installation command, or usage examples. It is recommended to inspect the source code before installation to verify that it meets your usage requirements.
Read MoreIntroducing the DeepSeek Harness Plugin: Identity Sentinel
`dsh-io/identity-sentinel` is a performance art plugin for DeepSeek Harness that verifies whether the operator is a Chinese-speaking user. The plugin collects local signals such as region and timezone, then evaluates them through a weighted rule engine. If the result is “non-Chinese,” it triggers termination of the workflow, prints an artistic banner, and exits; if it passes, it registers the tools. The plugin is a pure-function library and supports configurations such as custom thresholds. Installation requires attention to pnpm dependency issues, and it should be noted that this is not a security tool; installing it implies acceptance of its performance-art nature.
Read Moredsh-verify-gate: Enforcing Workspace Verification Before Case Closure
The DSH workflow plugin `dsh-verify-gate` is designed to address the issue of case-closing actions failing due to environment state changes during Agent execution. As a pre-execution gate, it requires that the specified verification commands be run in the current workspace before case closure. The workflow can continue only if the verification command executes successfully (exit code 0). This plugin is categorized under workflows, maintained by blueWhalei, and currently has no official installation command or concrete configuration examples. It is suitable for scenarios requiring environment consistency and automated test acceptance.
Read MoreDeepSeek Harness Plugin: dsh-workflows
dsh-workflows is a DeepSeek Harness plugin for persisting and managing JavaScript workflows. It provides core features such as background execution, a visual dashboard, slash commands, and compatibility assessment. Installation requires Node.js 22.19.0+ and pnpm 11.7.0. Its execution model supports pause/resume (within the same process only) and checkpoint replay, but an unexpected process termination will interrupt the run and make it unrecoverable. The plugin is maintained by zaalipro and is released under the MIT License.
Read Moredsh-funnel: Curating Tool Outputs Before the Model Context
dsh-funnel is a plugin for DeepSeek Harness that addresses the issue of long tool outputs consuming the context window and burying critical information. It curates data before it reaches the model, automatically truncating redundant content while preserving the header, footer, and key error/warning lines, and dumping the original full data to disk. Once installed, the model’s view is significantly reduced, with real-world testing showing substantial token savings. It supports zero configuration and on-demand access, effectively improving the efficiency and accuracy of LLM conversations.
Read Moredsh-water-care: Water Quality Maintenance Plugin
The DeepSeek Harness (DSH) plugin `dsh-water-care` is designed to provide agents with aquarium care solutions. Maintained by `uckkk`, it is a pure Node.js implementation that depends on `@deepseek-ai/cordis`. Users can install it with the command `dsh plugin add github:uckkk/dsh-water-care` and call the `water_care` tool to obtain water-quality maintenance recommendations. The plugin runs with the current process permissions, making it suitable for developers automating aquarium-related logic.
Read Moredsh-schedule-day: Daily Scheduling
`dsh-schedule-day` is a DeepSeek Harness (DSH) workflow plugin developed by uckkk for automating schedule management. Implemented purely in Node with no network dependencies, it supports precise date and schedule calculations based on input parameters and returns the results. It can be installed using the official command and is suitable for scenarios requiring automated scheduling. Because it runs with the permissions of the current DSH process, users are advised to review its source code and license before use.
Read Moredsh-cleaning-freq: Calculate Cleaning Frequency Based on Parameters
dsh-cleaning-freq is a workflow extension plugin in the DSH ecosystem, developed by uckkk and licensed under MIT. It is built with Node.js and has no network dependencies. Its core feature, `cleaning_freq`, accurately calculates execution frequency based on input parameters and provides suggestions. Install it with `dsh plugin add github:uckkk/dsh-cleaning-freq`. Before using it, verify compatibility with your dependency environment.
Read Moredsh-virtual-eng-team: Virtual Engineering Team Workflow Plugin
dsh-virtual-eng-team is a DeepSeek Harness (DSH) plugin designed to simulate virtual engineering team collaboration. Maintained by satan9394, it features 23 built-in expert roles and 6 interactive Office Hours questions, covering the complete development workflow from planning and design to review, QA, shipping, and retrospectives, with integrated security auditing to ensure a closed quality loop. It reduces the complexity and barrier to entry of simulated team collaboration through structured skills, making it suitable for developers who need to quickly set up team workflows. Review the source code and license before installing.
Read Moredsh-parallel-dev: DSH Parallel Development Workflow Plugin
`dsh-parallel-dev` is a plugin for DeepSeek Harness (DSH) designed to assist users in handling parallel development workflows. The tool involves architectural concepts such as file ownership, interface contracts, and vertical slicing versus horizontal layering, helping to manage the parallel development of complex tasks. Since detailed official documentation is not provided, users must visit the GitHub repository to install it themselves and inspect the source code to ensure it meets their usage requirements.
Read MoreDSH Plugin: satan9394/dsh-investment-research
The DeepSeek Harness plugin aims to standardize personal investment research methodologies, helping users build a systematic investment research workflow through four core stages (five-dimensional analysis, bull-bear debate, reflective auditing, and daily review). Maintained by the user satan9394, it is suitable for users who want to improve the depth and reusability of their investment research. Since no official installation command has been provided, users must inspect the source code themselves to determine the integration method.
Read Moredsh-trajectory-teacher: Convert AI Agent Execution Trajectories into Retrospective Reports
The DeepSeek Harness plugin `dsh-trajectory-teacher` converts raw AI agent logs into structured review reports, addressing the challenges of noisy execution trajectories and difficulty in learning from them. Maintained by XiaoMoDern, the plugin is triggered manually with the `/trajectory-teacher` command and uses an LLM to analyze key session events, such as decision points, errors, and reusable practices, then automatically generates a Markdown report. It supports noise filtering, API key masking, cost control (with reasoning disabled by default and length limits enforced), and custom model configuration, and is released under the MIT license.
Read MoreDeepSeek Harness Flat Team Plugin Introduction
`dsh-flat-teams` is a workflow plugin for DeepSeek Harness (DSH) designed to address the state-tracking confusion that can occur between independent windows in multi-agent collaboration. By using a flat team architecture, the plugin provides team creation, member management, and structured task dispatching, while supporting task state transitions (pending → completed) and offline delivery. It includes a built-in recorder service and a Web dashboard for visual monitoring of team and task progress. Installation requires restarting the Profile and an environment with DSH 0.1.0-rc.6 or Node 22.19+. It is intended for multi-agent collaboration within a single process.
Read Moreopenspec-agents: Enable agent teams to implement OpenSpec changes following standardized processes.
When working with AI coding tools like DSH, the most common approach is to feed requirements directly into an AI and have it generate the entire codebase from scratch. While this is quick to start, the drawbacks are apparent: the process is opaque, and correctness relies entirely on post-hoc manual inspection; if an error occurs at any step, rolling back is difficult without digging through chat history; and starting a new session breaks the progress. OpenSpec (Fission-AI/OpenSpec, openspec.sh) addresses the first half of this workflow: writing specifications before code, organizing requirements into proposals, designs, tasks, and specs. However, what about after the specifications are written? Implementation, verification, review, and merging still require the user to manually oversee the AI step-by-step. openspec-agents fills precisely this missing gap:
Read More