oh-my-deepseek-harness

Multi-agent orchestration for DeepSeek Harness.

What it does

Turns dsh into a multi-agent development environment with:

• 5 specialized agents: Explorer, Architect, Implementer, Reviewer, Debugger
• 4 workflow skills: Autopilot, Deepwork, Review, Reflect
• Preset system: Switch model configurations at runtime
• Task orchestration: Parallel execution with dependency tracking

Install

dsh plugin --profile your-profile add oh-my-deepseek-harness

Agents

| Agent | Role | Best For | |-------|------|----------| | explorer | Scans codebase | Starting new tasks, finding relevant code | | architect | Designs solutions | Planning, technical decisions | | implementer | Writes code | Implementation, file changes | | reviewer | Reviews code | Quality checks, security audits | | debugger | Fixes bugs | Diagnosing issues, tracing errors |

Skills

autopilot

Fully autonomous execution. Breaks down a goal, assigns to agents, executes in parallel.

/autopilot "add input validation to auth module"

deepwork

Focused work session for complex, multi-file changes.

/deepwork "refactor the database layer"

review

Systematic code review with structured findings.

/review src/auth.ts

reflect

Retrospective analysis after completing work.

/reflect

Presets

Switch model configurations:

| Preset | Description | |--------|-------------| | balanced | Mix of quality and speed (default) | | quality | Strongest models for all agents | | speed | Fastest models for all agents | | cost | Cheapest models for all agents |

Configuration

- id: oh-my-dsh
  name: oh-my-deepseek-harness
  config:
    enableAutopilot: true
    enableDeepwork: true
    enableReview: true
    enableReflect: true
    defaultModel: deepseek-v4
    maxParallel: 3

License

MIT -- YYTbit