360 Secure SkillHub
360 安全技能中心

Oh-my-openagent

提供 oh-my-openagent(OpenCode 插件)的配置参考,涵盖 11 个智能体、8 个任务类别、模型与提供商优先级、钩子、内置技能、斜杠命令和 MCP 服务器,含安装步骤、本地配置与认证文件位置说明,并附带 references 目录的细分文档。

kolyanberoks-23loud
kolyanberoks-23loud
热度 72.94
最近更新 2026-05-17 14:36:19
分类:开发依赖:代码执行 / 读写文件应用场景:工作流 / 任务管理 / API 集成

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add oh-my-openagent
Metadata
name
oh-my-openagent
description
Configure and operate oh-my-openagent (OmO), the OpenCode plugin that adds multi-agent orchestration, skills, hooks, MCPs, and multi-provider model routing. Use this skill when the user asks about OmO configuration, agent setup, model selection, category tuning, hook management, or provider authentication.
homepage
https://github.com/code-yeongyu/oh-my-openagent
metadata
{"clawdbot":{"emoji":"🤖"}}

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

bunx oh-my-opencode install

On Windows, interactive mode may crash. Configure manually instead:

  1. Edit ~/.config/opencode/opencode.json:
{
  "plugin": ["oh-my-opencode@latest"]
}
  1. 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:

{
  "agents": {
    "sisyphus": {
      "model": "my-preferred-model",
      "fallback_models": ["fallback-1", "fallback-2"]
    }
  }
}

Override any category:

{
  "categories": {
    "quick": {
      "model": "fast-cheap-model"
    }
  }
}

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)
  2. Kimi for Coding
  3. GitHub Copilot
  4. Venice
  5. OpenCode Go
  6. OpenCode Zen
  7. 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

/ralph-loop, /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 ~/.config/opencode/
  • Auth tokens are stored in ~/.local/share/opencode/auth.json
Oh-my-openagent_360安全技能中心