AI Agent Hub
Back to plugins
oh-my-dsh preview

oh-my-dsh

Workflow Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install LiuMengxuan04/oh-my-dsh

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install LiuMengxuan04/oh-my-dsh in your DeepSeek Harness environment to install, with the full source code hosted at https://github.com/LiuMengxuan04/oh-my-dsh.

About this plugin

While DeepSeek Harness' native Goal mode excels at straightforward "keep going" workflows, it often struggles when development tasks span multiple model turns or face unexpected interruptions. Oh My DSH (also known as DSH Autopilot) addresses this by introducing a durable control layer designed specifically for long-running, complex automation that requires planning persistence, state recovery, and strict oversight.

At its core, the plugin orchestrates work through a persistent dependency graph (DAG) paired with bounded execution budgets and automated verification gates. It manages task delegation, tracks completion evidence, and freezes project checklists to prevent policy drift. If a process crashes or an Agent fails, the system safely resumes from the last stable checkpoint. With built-in pause, resume, audit logging, and read-only reviewer stages, every run maintains a clear, auditable trail until it passes final validation.

This tool is ideal for developers and engineers who need reliable, multi-stage AI workflows with verifiable outcomes rather than open-ended generation. It complements rather than replaces native DSH permissions, acting as a structural overlay that transforms unpredictable AI sessions into controlled, recoverable engineering pipelines. Currently in developer preview, it encourages isolated experimentation while delivering enterprise-grade task governance features.

Screenshots

Use Cases

  • Complex code refactoring that spans multiple model turns and must survive system restarts.
  • Development pipelines requiring strict dependency ordering, verifiable artifacts, and multi-stage review gates.
  • Budget-constrained automated experiments that demand crash recovery and zero-progress-loss guarantees.

Best For

  • Senior engineers who require auditable, traceable, and deterministic AI-assisted development workflows.
  • Developers experimenting with early-stage plugin ecosystems and handling unpredictable Agent callbacks.
  • Team leads building resilient automation pipelines or long-running CI/CD validation environments.