oh-my-openagent
oh-my-openagent (OmO) is an OpenCode plugin that transforms a single-agent coding assistant into a multi-agent orchestration system with 11 specialized agents, 8 task categories, 40+ hooks, built-in skills, MCPs, and multi-provider model routing.
Repository: https://github.com/code-yeongyu/oh-my-openagent
Discord: https://discord.gg/PUwSMR9XNk
Config schema: https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/dev/assets/oh-my-opencode.schema.json
When to use this skill
- - User asks how to configure oh-my-openagent or oh-my-opencode
- User wants to change which model an agent or category uses
- User asks about the agent system (Sisyphus, Oracle, Hephaestus, etc.)
- User wants to add, disable, or configure hooks
- User asks about task delegation, categories, or background tasks
- User needs help with provider authentication or model routing
- User wants to create or manage custom skills
- User asks about slash commands like /ralph-loop, /start-work, /refactor
Installation
CODEBLOCK0
On Windows, interactive mode may crash. Configure manually instead:
- 1. Edit
~/.config/opencode/opencode.json:
CODEBLOCK1
- 2. Edit
~/.config/opencode/oh-my-opencode.json for agent/category overrides.
Core concepts
Agents
OmO provides 11 agents, each with a default model and role:
| Agent | Role | Default Model |
|---|
| Sisyphus | Main orchestrator | claude-opus-4-6 |
| Hephaestus |
Deep worker (code-heavy) | gpt-5.3-codex |
| Oracle | Architecture consultant (read-only) | gpt-5.4 |
| Librarian | Documentation and reference search | gemini-3-flash |
| Explore | Fast codebase grep | grok-code-fast-1 |
| Multimodal-Looker | Vision / image analysis | gpt-5.3-codex |
| Prometheus | Planner | claude-opus-4-6 |
| Metis | Plan consultant | claude-opus-4-6 |
| Momus | Plan reviewer | gpt-5.4 |
| Atlas | Todo orchestrator | claude-sonnet-4-6 |
| Sisyphus-Junior | Delegated task worker | (category-dependent) |
See references/agents.md for full details including fallback chains.
Categories
8 built-in task categories determine which model Sisyphus-Junior uses:
| Category | Best For | Default Model |
|---|
| visual-engineering | Frontend, UI/UX, design | gemini-3.1-pro |
| ultrabrain |
Hard logic-heavy tasks | gpt-5.4 |
| deep | Autonomous problem-solving | gpt-5.3-codex |
| artistry | Creative, unconventional approaches | gemini-3.1-pro |
| quick | Trivial single-file changes | claude-haiku-4-5 |
| unspecified-low | Low-effort misc tasks | claude-sonnet-4-6 |
| unspecified-high | High-effort misc tasks | claude-opus-4-6 |
| writing | Documentation, prose | gemini-3-flash |
See references/categories.md for full details including fallback chains.
Configuration
All config lives in ~/.config/opencode/oh-my-opencode.json.
Override any agent:
CODEBLOCK2
Override any category:
CODEBLOCK3
See references/configuration.md for the full config reference.
Provider priority
OmO resolves models through a provider priority chain:
- 1. Native (built-in to OpenCode)
- Kimi for Coding
- GitHub Copilot
- Venice
- OpenCode Go
- OpenCode Zen
- Z.ai Coding Plan
Each agent's default model has a fallback chain. If no provider can serve a model, OmO tries the next fallback. See references/providers.md.
Tools
OmO injects 40+ tools into agents. See references/tools.md for the full list.
Hooks
40+ built-in hooks that fire on events (session start, file save, pre-commit, etc.). See references/hooks.md.
Skills
Built-in skills: playwright, playwright-cli, agent-browser, dev-browser, git-master, frontend-ui-ux. See references/skills.md.
Slash commands
INLINECODE10 , /ulw-loop, /start-work, /refactor, /init-deep, /handoff, /cancel-ralph, /stop-continuation. See references/commands.md.
MCPs
Built-in MCP servers: websearch (Exa), context7 (docs), grep_app (GitHub code search). See references/mcps.md.
Security and privacy
- - OmO runs locally as an OpenCode plugin
- No data is sent to oh-my-openagent servers
- Model API calls go directly to the configured providers (GitHub Copilot, Google, OpenAI, etc.)
- Config files are stored locally in INLINECODE20
- Auth tokens are stored in INLINECODE21
技能名称: oh-my-openagent
详细描述:
oh-my-openagent
oh-my-openagent (OmO) 是一款 OpenCode 插件，可将单智能体编码助手转变为多智能体编排系统，包含 11 个专用智能体、8 个任务类别、40 多个钩子、内置技能、MCP 以及多提供商模型路由。
仓库地址：https://github.com/code-yeongyu/oh-my-openagent
Discord 社区：https://discord.gg/PUwSMR9XNk
配置模式：https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/dev/assets/oh-my-opencode.schema.json
何时使用此技能
- - 用户询问如何配置 oh-my-openagent 或 oh-my-opencode
- 用户想要更改智能体或类别使用的模型
- 用户询问智能体系统（Sisyphus、Oracle、Hephaestus 等）
- 用户想要添加、禁用或配置钩子
- 用户询问任务委派、类别或后台任务
- 用户需要提供商认证或模型路由方面的帮助
- 用户想要创建或管理自定义技能
- 用户询问斜杠命令，如 /ralph-loop、/start-work、/refactor
安装
bash
bunx oh-my-opencode install
在 Windows 上，交互模式可能崩溃。请改为手动配置：
- 1. 编辑 ~/.config/opencode/opencode.json：
json
{
plugin: [oh-my-opencode@latest]
}
- 2. 编辑 ~/.config/opencode/oh-my-opencode.json 以覆盖智能体/类别设置。
核心概念
智能体
OmO 提供 11 个智能体，每个都有默认模型和角色：
| 智能体 | 角色 | 默认模型 |
|---|
| Sisyphus | 主编排器 | claude-opus-4-6 |
| Hephaestus |
深度工作者（代码密集型） | gpt-5.3-codex |
| Oracle | 架构顾问（只读） | gpt-5.4 |
| Librarian | 文档和参考搜索 | gemini-3-flash |
| Explore | 快速代码库搜索 | grok-code-fast-1 |
| Multimodal-Looker | 视觉/图像分析 | gpt-5.3-codex |
| Prometheus | 规划器 | claude-opus-4-6 |
| Metis | 规划顾问 | claude-opus-4-6 |
| Momus | 规划审查员 | gpt-5.4 |
| Atlas | 待办事项编排器 | claude-sonnet-4-6 |
| Sisyphus-Junior | 委派任务工作者 | （取决于类别） |
完整详情（包括回退链）请参见 references/agents.md。
类别
8 个内置任务类别决定 Sisyphus-Junior 使用的模型：
| 类别 | 最适合 | 默认模型 |
|---|
| visual-engineering | 前端、UI/UX、设计 | gemini-3.1-pro |
| ultrabrain |
硬逻辑密集型任务 | gpt-5.4 |
| deep | 自主问题解决 | gpt-5.3-codex |
| artistry | 创意、非传统方法 | gemini-3.1-pro |
| quick | 琐碎的单文件更改 | claude-haiku-4-5 |
| unspecified-low | 低工作量杂项任务 | claude-sonnet-4-6 |
| unspecified-high | 高工作量杂项任务 | claude-opus-4-6 |
| writing | 文档、散文 | gemini-3-flash |
完整详情（包括回退链）请参见 references/categories.md。
配置
所有配置位于 ~/.config/opencode/oh-my-opencode.json。
覆盖任何智能体：
json
{
agents: {
sisyphus: {
model: my-preferred-model,
fallback_models: [fallback-1, fallback-2]
}
}
}
覆盖任何类别：
json
{
categories: {
quick: {
model: fast-cheap-model
}
}
}
完整配置参考请参见 references/configuration.md。
提供商优先级
OmO 通过提供商优先级链解析模型：
- 1. 原生（OpenCode 内置）
- Kimi for Coding
- GitHub Copilot
- Venice
- OpenCode Go
- OpenCode Zen
- Z.ai Coding Plan
每个智能体的默认模型都有回退链。如果没有提供商可以提供模型，OmO 会尝试下一个回退。请参见 references/providers.md。
工具
OmO 向智能体注入 40 多个工具。完整列表请参见 references/tools.md。
钩子
40 多个内置钩子，在事件（会话开始、文件保存、预提交等）时触发。请参见 references/hooks.md。
技能
内置技能：playwright、playwright-cli、agent-browser、dev-browser、git-master、frontend-ui-ux。请参见 references/skills.md。
斜杠命令
/ralph-loop、/ulw-loop、/start-work、/refactor、/init-deep、/handoff、/cancel-ralph、/stop-continuation。请参见 references/commands.md。
MCP
内置 MCP 服务器：websearch (Exa)、context7 (文档)、grep_app (GitHub 代码搜索)。请参见 references/mcps.md。
安全与隐私
- - OmO 作为 OpenCode 插件在本地运行
- 没有数据发送到 oh-my-openagent 服务器
- 模型 API 调用直接发送到已配置的提供商（GitHub Copilot、Google、OpenAI 等）
- 配置文件本地存储在 ~/.config/opencode/
- 认证令牌存储在 ~/.local/share/opencode/auth.json
