Foreword

The core philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” To replicate the mature orchestration patterns of oh-my-claudecode on top of DSH’s native primitives (such as workflow, subagent, ralph, and goal), hawk2048 developed oh-my-dsh. It is a multi-agent orchestration layer plugin that encapsulates complex orchestration flows into plug-and-play skills, aiming to reduce the cost for developers to configure complex workflows.

Installation and Enablement

The installation process relies on the DSH CLI and pnpm. Run the following command for a zero-configuration plugin installation:

dsh plugin --profile web add github:hawk2048/oh-my-dsh

Or install it via the npm package name:

dsh plugin --profile web add @hawk2048/oh-my-dsh

After installation, restart the corresponding profile, and the orchestration layer will be ready.

Core Features

The plugin maps oh-my-claudecode patterns onto DSH primitives by encapsulating 30 auto-loaded skills.

1. Team-First Orchestration Workflows

The plugin defines a standardized stage process centered on “teams,” covering the complete lifecycle from planning to correction:
* team-plan: read-only exploration and parallel subagent analysis.
* team-prd: product requirements document generation.
* team-exec: parallel execution and task writing.
* team-verify: independent review and test checking.
* team-fix: issue correction and regression verification.

2. Zero Configuration and Natural Language Interface

No manual configuration files are required. The plugin includes 13 /omd-* slash commands and 30 natural language skills. Users can directly describe requirements in natural language (for example, “plan this” or “autopilot, build a REST API”), and the system automatically matches and invokes the corresponding skill.

3. Automatic Parallelization and Persistent Execution

For complex tasks, the plugin automatically distributes workflows across multiple subagent instances for parallel execution. For persistent execution, it leverages DSH’s ralph and goal primitives to keep the task loop running until verification is complete.

4. Real-Time Visibility and Learning

  • HUD (Head-Up Display): This is a signature feature of the plugin. It inline-renders a dsh-ui progress dashboard in the reply, showing task progress, timelines, and statistics in real time. Unlike a CLI status line, it is structured data embedded in the conversation flow.
  • Experience Learning: Through the omd-skillify and omd-remember skills, the plugin supports extracting reusable skills from past executions and persisting memory across sessions.

5. Zero Runtime Overhead

The plugin does not introduce new host tools or runtime environments; it is built entirely on DSH primitives. This preserves DSH’s own model routing and sandbox capabilities without adding extra system dependencies.

Typical Usage

After installation and starting a session, the orchestration environment is activated.

  1. View candidate commands: Enter /omd to see the list of all available skills and slash commands.
  2. Trigger with natural language: Directly state your intent in the conversation, for example:
    • “plan this” (triggers the planning skill)
    • “autopilot, build a REST API” (triggers the autopilot skill to build an API)
  3. View the HUD: During task execution, pay attention to the inline dashboard in the reply, which provides visualized execution progress.

Notes

  • Dependency Environment: The plugin requires DeepSeek Harness (dsh CLI) and pnpm to be installed locally.
  • Feature Changes: ultrawork mode was removed in version 5.0.0 (retired); the related functionality has been integrated into the execute, team, and ralph modes.
  • HUD Definition: The HUD is a dsh-ui dashboard inline in the reply, not a CLI status line.
  • License: The plugin is open-sourced under the MIT License.

Summary

oh-my-dsh provides DSH users with a shortcut to replicate oh-my-claudecode workflows. By encapsulating complex orchestration logic into 30 skills, it achieves efficient mapping from natural language to DSH primitives while maintaining zero configuration and zero runtime overhead. It is suitable for scenarios requiring rapid setup of multi-agent collaboration systems.