oh-my-unified

An OpenCode plugin providing Norse Pantheon AI agents with MCP integration, skill mapping, and LLM-driven orchestration.

Quick Start

1. Install: npm install -g oh-my-unified
2. Configure: Add to ~/.config/opencode/opencode.json:

   {
     "plugin": ["oh-my-unified"]
   }

3. Restart OpenCode Desktop
4. Try it: Type /plan build a simple todo app

Available Commands

| Command | Description | |---|---| | /plan | Full Norse pipeline: assess → assemble → improvise → act | | /status | Show pipeline information | | /health | System health report | | /diagnose | 12 parallel diagnostic checks | | /agents | List all agents with modes |

Agent Reference

Primary Agents (Selector Dropdown)

| Agent | Role | Description | |---|---|---| | @odin | Strategist | Chief strategist, interviews, researches, plans | | @njord | Orchestrator | Manages parallel execution, delegates to specialists |

Pipeline Agents (@ autocomplete + task tool)

| Agent | Role | Description | |---|---|---| | @mimir | Advisor | Architecture review, complex debugging, design tradeoffs | | @frigg | Analyst | Gap analysis, risk identification, blind spots | | @forseti | Council | Multi-LLM deliberation, adversarial critique | | @thor | Builder | Code implementation, refactoring, building features | | @vidar | Mapper | Codebase exploration, structure mapping | | @tyr | Critic | Quality review, standards validation |

Specialist Agents (@ autocomplete only)

| Agent | Role | Description | |---|---|---| | @sif | Scout | Fast codebase search, finding files | | @eir | Researcher | Documentation lookup, API references | | @freyr | Designer | UI/UX design, visual styling | | @hermod | Fixer | Quick bug fixes, single-file changes | | @heimdall | Watcher | Visual analysis, screenshots, diagrams | | @magni | Follower | Focused task execution | | @hod | Voter | Independent council perspective |

How /plan Works

The /plan command uses LLM-driven orchestration:

1. Phase 1 (Frigg): Gap analysis — identifies missing requirements, risks
2. Phase 2 (Mimir): Architecture — designs component structure, data flow
3. Phase 3 (Forseti): Critique — adversarial review, identifies issues
4. Phase 4 (Thor): Implementation — writes production-ready code

Each phase chains results from the previous phase via the task tool.

How @ Mentions Work

Type @agentname in the chat to invoke a specific agent:

@forseti analyze this code for security vulnerabilities
@mimir review this architecture
@thor implement the login feature

Model Assignment

| Model | Used By | |---|---| | mimo-v2.5-free | mimir, forseti, frigg, tyr (deep analysis) | | deepseek-v4-flash-free | njord, vidar, thor, hermod, magni (fast execution) | | big-pickle | sif, eir, freyr, heimdall (balanced) | | nemotron-3-super-free | hod (reliability anchor) |

Each agent has fallback models for automatic retry on failure.

MCP Integration

The plugin auto-discovers MCPs from opencode.json and maps them to agents based on their roles. Agents only access MCPs relevant to their capabilities.

Skills

Built-in skills are mapped to agents based on their roles:

• gap-analysis → frigg
• reasoning, code-review, simplify → mimir
• consensus, multi-perspective → forseti
• codemap, architecture-analysis → vidar

Configuration

Customize agent behavior via ~/.config/opencode/oh-my-unified.json:

{
  "agents": {
    "frigg": {
      "model": "opencode/mimo-v2.5-free",
      "fallback_models": ["opencode/deepseek-v4-flash-free"]
    }
  }
}

Key Features

• Assess, Assemble, Act Workflow: Structured Norse agent pipeline methodology.
• Improvisation Loop: Multi-agent adversarial feedback and quality control.
• Transparency Log: Queryable transparency log for full auditability of agent actions.

Architecture

See ARCHITECTURE.md for system design and module structure. See CLAUDE.md for AI agent working instructions. See HARD_RULES.md for inviolable constraints.