OpenCode oh-my-openagent Usage Tutorial

Oh My OpenAgent(abbreviated asOmO) is an open-source AI programming assistant enhancement plugin, formerly calledoh-my-opencode。

oh-my-openagent runs on top of OpenCode, the AI programming tool, turning it into11 professional AI Agentsforming a development team.

Explained in one sentence:

You submit a request → OmO automatically orchestrates multiple AIs to work in parallel → Several times faster than a single AI

Currently has on GitHub54k+ Stars, it is the most popular OpenCode plugin.

Installation

Copy and paste the following prompt into your LLM tool (OpenCode, Claude Code, AmpCode, Cursor, etc.):

按照此链接说明安装并配置 oh-my-opencode：https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/refs/heads/dev/docs/guide/installation.md

After executing, it will install itself. During installation, there may be API key configuration options; you can choose to configure them or not:

Meet the 11 AI Team Members (Agents)

OmO has 11 specialized Agents built in, each with its own role:

Core Agents

Agent Role Default Model When to Use
Sisyphus Main coordinator, all-around developer Claude Opus Entry point for all tasks, automatically orchestrates other Agents
Prometheus Strategic Planner Claude Opus / GPT-5.4 Planning for complex tasks
Atlas Todo task management Kimi K2.5 Track multi-step task progress

Professional Agents

Agent Role Default Model Specializes in
Hephaestus Deep autonomous worker GPT-5.3-codex Long-duration, high-intensity independent coding tasks
Oracle Architecture consultant/debugging expert GPT-5.4 System design, locating difficult bugs
Momus High-precision reviewer GPT-5.4 Code review, output verification
Metis Plan review Claude Opus Check whether Prometheus's plan has any omissions

Tool-type Agents (Fast, Low-cost)

Agent Role Default Model Why use cheap models
Explore Quick codebase search MiniMax / Grok Code Fast Search tasks don't require high intelligence; speed matters more
Librarian Document/code retrieval MiniMax (Free) Document retrieval doesn't require deep reasoning
Multimodal Looker Screenshot/visual analysis Kimi K2.5 Kimi excels at multimodal understanding

💡 Beginner tip: You don't need to manually select Agents. Sisyphus will automatically orchestrate the appropriate Agent combination based on your task type.

Three Working Modes

Mode 1: ultrawork (Easiest)

Add to the promptultraworkorulw, everything is automated:

ultrawork 帮我把这个 React 组件改成 TypeScript

Suitable for: daily development tasks, scenarios where you want AI to complete everything automatically.

Mode 2: Prometheus Planning Mode (Most Precise)

pressTab keyEnter Prometheus mode. The AI will first understand your requirements through Q&A, generate a detailed execution plan, and start executing after you confirm.

Workflow:

  1. pressTabSwitch to Prometheus mode
  2. Describe your requirements; the AI will ask you some clarifying questions
  3. AI generates an execution plan, you confirm
  4. Enter/start-workto start execution

Suitable for: complex feature development, architecture refactoring, and other scenarios requiring precise planning.

Mode 3: Normal Conversation (Most Flexible)

Directly enter your question or requirement without adding any keywords:

解释一下 src/auth.ts 里的 JWT 验证逻辑

Suitable for: lightweight tasks such as asking questions, code explanation, and small modifications.

Configuration File Details

OmO has two configuration files with different priorities:

  • Project-level configuration (higher priority):.opencode/oh-my-opencode.jsonc
  • User-level configuration (global default):~/.config/opencode/oh-my-opencode.jsonc

Configuration files support comments (JSONC format) and JSON Schema auto-completion:

{
  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/dev/assets/oh-my-opencode.schema.json",
  
  "agents": {
    // 主协调 Agent，建议用 Claude Opus 系列或 Kimi
    "sisyphus": {
      "model": "anthropic/claude-opus-4-6"
    },
    // 架构顾问，建议用 GPT 系列
    "oracle": {
      "model": "openai/gpt-5.4",
      "variant": "high"
    }
  }
}

Beginner advice: Don't modify the configuration right after installation; the default settings are already good.

Common Commands Quick Reference

Command Description
opencode Start OmO
bunx oh-my-opencode install Install/reconfigure plugin
bunx oh-my-opencode doctor Diagnose configuration issues
opencode auth login Log in to AI provider account
opencode models View all available models
bunx oh-my-opencode refresh-model-capabilities Refresh model capability cache

Within the conversation interface:

Shortcut/Command Description
Tab Switch to Prometheus planning mode
/start-work Start executing the plan in Prometheus mode
ultrawork / ulw Add to the prompt to activate full-featured parallel mode

Common Issues Troubleshooting

Q: After installation, runningopencodedoesn't show OmO features

Solution: Check whether the plugin is correctly registered in the configuration file:

cat ~/.config/opencode/opencode.json

Confirmpluginfield contains"oh-my-opencode". If not, re-runbunx oh-my-opencode install。

Q: Runtime warning "Using legacy package name"

Solution: Changeopencode.jsonthe plugin name in ... from"oh-my-opencode"to"oh-my-openagent"：

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

Q: Sisyphus Agent performs poorly, feels the same as regular Claude

Cause: Possibly not using the Claude Opus model, or not addingultrawork。

Solution：

  1. Confirm you're logged into a Claude Pro/Max account
  2. Addultraworkkeyword to the prompt
  3. Runbunx oh-my-opencode doctorto check model configuration

Q: Task stops halfway through

Solution: Continue promptingultrawork 继续之前的任务, Sisyphus will resume progress through the Todo system.

Q: Want to disable anonymous telemetry data

OmO enables anonymous telemetry by default (uses PostHog, data is hashed). To turn it off:

export OMO_SEND_ANONYMOUS_TELEMETRY=0
# 或
export OMO_DISABLE_POSTHOG=1
Other Extensions