https://github.com/alvinunreal/oh-my-opencode-slim/blob/master/README.zh-CN.md
如果你手上全是 gpt 模型，那么建议使用 omo-slim，而不是使用 omo。
安装
手动安装：
bunx oh-my-opencode-slim@latest install
# 如果上面那个报错就试试下面这种写法
bun x oh-my-opencode-slim@latest install这个命令运行完之后，脚本会自动在 opencode 配置中添加下面的字段：
"plugin": [
"oh-my-opencode-slim"
],
"agent": {
"explore": {
"disable": true
},
"general": {
"disable": true
},
"build": {
"disable": true
},
"plan": {
"disable": true
}
},并会在 tui.json 中添加：
{
"plugin": [
"oh-my-opencode-slim"
]
}
然后就完事了。
注意，omo 和 omo-slim 不要同时打开，在 opencode 的插件配置中干掉一个。
2.0 新版本有关后台子 agent 相关配置
新版本 omo-slim 调用子 agent 的方式不再是同步阻塞式，而是类似 omo 的后台异步式，这需要 opencode 的相关配置支持。在更新 omo-slim 时安装脚本应当已经修改了环境变量以启用该配置，具体来说，是在终端配置中加入了：
# >>> oh-my-opencode-slim background subagents >>> export OPENCODE_EXPERIMENTAL_BACKGROUND_SUBAGENTS=true # <<< oh-my-opencode-slim background subagents <<<如果没有可是手动加入，Windows 也是同理。
之后，我们直接在配置目录中新建或修改一个 oh-my-opencode-slim.jsonc，我的配置与官方建议的不同。尤其是大多数子代理，作者都使用了非顶级模型的中等推理强度，这可能有他的道理，但是我依然对这些进行了升级。
针对推理强度选择
2.2.5 版本中，作者终于舍弃了低强度模型，把大部分模型的推理强度都换成了 xhigh。
我使用了很多的 max。
不过考虑到现在的成本，luna-max 或许是最佳的选择，可以看到 luna-max 非常省钱，并且性能甚至比 xigh 的 5.5 还要好。
具体可以看我的
openai-easy和codelink-gpt-easy预设。
并且，omo-slim 支持配置多种预设，使用的时候直接更改配置文件中的 "preset" 字段即可，可以看到我就创建了很多预设，针对不同供应商或者是不同的模型组合。
除了默认的 agent 之外，你还可以自定义 agent，比如 writer、helper 都是我的自定义 agent。
同时，你还可以给各个子 agent 配置默认的 skill 和 mcp 以方便它们使用。下文的预设就是我在参考了该插件作者的预设后调整的，skill 几乎都是与前端相关的，更多请看这篇文章。
{
"$schema": "https://unpkg.com/oh-my-opencode-slim@latest/oh-my-opencode-slim.schema.json",
"preset": "codelink-gpt&torch-grok",
"presets": {
"codelink-gpt&torch-grok": {
"orchestrator": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
],
"prompt": "准备向用户输出可见回复或对外展示可见输出时，先加载并使用 shuorenhua skill；内部工作、子 agent、工具调用和内部文档不使用该 skill。"
},
"oracle": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [],
"mcps": []
},
"librarian": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [],
"mcps": []
},
"writer": {
"model": "codelink-gpt/gpt-6-astra",
"variant": "medium",
"skills": [
"deliverable-document-writer",
"create-readme"
],
"mcps": [],
"prompt": "你是文档写作专员。负责创建、修改和整理面向用户、贡献者、项目维护者或交付对象的正式文档，包括 README、指南、API 文档、技术报告、示例和变更日志。写作前核对代码和现有文档中的事实。写完后必须按所用 skill 读回成稿，按句清除无源否定、无端假设、预防式前提、无端转折和过程泄漏，并把修改写回文件；没读回成稿、没把修改写回文件，不算完成。不要接管 project-context、.agent-context、agent memory、session notes、任务日志、内部工作记录、临时计划或其他仅供 agent 内部使用的文档。除非文档任务明确要求，不要修改业务源代码。回复调用方时说明修改了哪些文档，并必须附上写后审计（只给调用方，不得写入文档）：\\n## 写后审计\\n- 已读回成稿：是\\n- 功能异常句：删除或改写过的原句；没有则写无\\n- 抽查句：至少 5 句。格式为「原句 → 通过/删除/改写」；通过时写明读者相反预期来自哪里。优先抽否定句、如果/注意/然而/虽然、安装或对比句。缺这一节不算完成。抽查句或功能异常句里标「删除」或「改写」的，必须已经写回文件。文件仍是原稿却汇报已改，不算完成。不要把未落地的修改交给调用方。",
"orchestratorPrompt": "@writer\n- **Delegate when:** 用户明确要求创建、修改、重写、整理或生成面向用户的文档；或者 Orchestrator 判断当前实现任务为了完成交付，需要自动新建或更新 README、Markdown、指南、API 文档、技术报告、示例、变更日志或其他正式项目文档，即使用户没有单独提出文档要求。\n- **Don't delegate when:** 用户只是询问文档内容或要求解释而不需要写入文档文件；或者文档仅供 agent 内部使用，例如 project-context、.agent-context、agent memory、session notes、任务日志、内部工作记录、临时计划和中间分析。\n- **Required:** 对所有符合条件的正式文档任务，无论需求来自用户还是 Orchestrator 自己的实现计划，都必须先委托给 @writer，再创建或编辑文档文件。内部 agent 文档不受此规则限制。"
}
},
"codelink-gpt&micu-grok": {
"orchestrator": {
"model": "micu-grok/grok-4.6",
"variant": "high",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
],
"prompt": "准备向用户输出可见回复或对外展示可见输出时，先加载并使用 shuorenhua skill；内部工作、子 agent、工具调用和内部文档不使用该 skill。"
},
"oracle": {
"model": "micu-grok/grok-4.6",
"variant": "high",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "micu-grok/grok-4.6",
"variant": "high",
"skills": [],
"mcps": []
},
"librarian": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "micu-grok/grok-4.6",
"variant": "high",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "micu-grok/grok-4.6",
"variant": "high",
"skills": [],
"mcps": []
},
"writer": {
"model": "codelink-gpt/gpt-6-astra",
"variant": "medium",
"skills": [
"deliverable-document-writer",
"create-readme"
],
"mcps": [],
"prompt": "你是文档写作专员。负责创建、修改和整理面向用户、贡献者、项目维护者或交付对象的正式文档，包括 README、指南、API 文档、技术报告、示例和变更日志。写作前核对代码和现有文档中的事实。写完后必须按所用 skill 读回成稿，按句清除无源否定、无端假设、预防式前提、无端转折和过程泄漏，并把修改写回文件；没读回成稿、没把修改写回文件，不算完成。不要接管 project-context、.agent-context、agent memory、session notes、任务日志、内部工作记录、临时计划或其他仅供 agent 内部使用的文档。除非文档任务明确要求，不要修改业务源代码。回复调用方时说明修改了哪些文档，并必须附上写后审计（只给调用方，不得写入文档）：\\n## 写后审计\\n- 已读回成稿：是\\n- 功能异常句：删除或改写过的原句；没有则写无\\n- 抽查句：至少 5 句。格式为「原句 → 通过/删除/改写」；通过时写明读者相反预期来自哪里。优先抽否定句、如果/注意/然而/虽然、安装或对比句。缺这一节不算完成。抽查句或功能异常句里标「删除」或「改写」的，必须已经写回文件。文件仍是原稿却汇报已改，不算完成。不要把未落地的修改交给调用方。",
"orchestratorPrompt": "@writer\n- **Delegate when:** 用户明确要求创建、修改、重写、整理或生成面向用户的文档；或者 Orchestrator 判断当前实现任务为了完成交付，需要自动新建或更新 README、Markdown、指南、API 文档、技术报告、示例、变更日志或其他正式项目文档，即使用户没有单独提出文档要求。\n- **Don't delegate when:** 用户只是询问文档内容或要求解释而不需要写入文档文件；或者文档仅供 agent 内部使用，例如 project-context、.agent-context、agent memory、session notes、任务日志、内部工作记录、临时计划和中间分析。\n- **Required:** 对所有符合条件的正式文档任务，无论需求来自用户还是 Orchestrator 自己的实现计划，都必须先委托给 @writer，再创建或编辑文档文件。内部 agent 文档不受此规则限制。"
}
},
"codelink-gpt": {
"orchestrator": {
"model": "codelink-gpt/gpt-5.6-sol",
"variant": "max",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "codelink-gpt/gpt-5.6-sol",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "codelink-gpt/gpt-5.6-sol",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "codelink-gpt/gpt-5.6-sol",
"variant": "xhigh",
"skills": [],
"mcps": []
},
"librarian": {
"model": "codelink-gpt/gpt-5.6-sol",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "codelink-gpt/gpt-5.6-sol",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "codelink-gpt/gpt-5.6-sol",
"variant": "max",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "codelink-gpt/gpt-5.6-sol",
"variant": "max",
"skills": [],
"mcps": []
}
},
"codelink-gpt-easy": {
"orchestrator": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "max",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "codelink-gpt/gpt-5.6-sol",
"variant": "high",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "max",
"skills": [],
"mcps": []
},
"librarian": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "max",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "max",
"skills": [],
"mcps": []
}
},
"codelink-gpt-mix": {
"orchestrator": {
"model": "codelink-gpt/gpt-6-astra",
"variant": "medium",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "codelink-gpt/gpt-6-astra",
"variant": "medium",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "codelink-gpt/gpt-6-astra",
"variant": "medium",
"skills": [],
"mcps": []
},
"librarian": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "codelink-gpt/gpt-5.6-luna",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "codelink-gpt/gpt-6-astra",
"variant": "medium",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "codelink-gpt/gpt-6-astra",
"variant": "medium",
"skills": [],
"mcps": []
}
},
"torchai-gpt-easy": {
"orchestrator": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "max",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "torchai-gpt/gpt-5.6-sol",
"variant": "xhigh",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "max",
"skills": [],
"mcps": []
},
"librarian": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "max",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "max",
"skills": [],
"mcps": []
}
},
"torchai-gpt-mix": {
"orchestrator": {
"model": "torchai-gpt/gpt-6-astra",
"variant": "medium",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "torchai-gpt/gpt-6-astra",
"variant": "medium",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "torchai-gpt/gpt-6-astra",
"variant": "medium",
"skills": [],
"mcps": []
},
"librarian": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "torchai-gpt/gpt-6-astra",
"variant": "medium",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "torchai-gpt/gpt-6-astra",
"variant": "medium",
"skills": [],
"mcps": []
}
},
"torchai-gpt-noluna": {
"orchestrator": {
"model": "torchai-gpt/gpt-5.6-sol",
"variant": "high",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "torchai-gpt/gpt-5.6-sol",
"variant": "high",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "torchai-gpt/gpt-5.6-sol",
"variant": "xhigh",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "torchai-gpt/gpt-5.6-sol",
"variant": "high",
"skills": [],
"mcps": []
},
"librarian": {
"model": "torchai-gpt/gpt-5.6-terra",
"variant": "high",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "torchai-gpt/gpt-5.6-terra",
"variant": "high",
"skills": [],
"mcps": []
},
"designer": {
"model": "torchai-gpt/gpt-5.6-sol",
"variant": "high",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "torchai-gpt/gpt-5.6-sol",
"variant": "high",
"skills": [],
"mcps": []
}
},
"torch-gpt&torch-grok": {
"orchestrator": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
],
"prompt": "准备向用户输出可见回复或对外展示可见输出时，先加载并使用 shuorenhua skill；内部工作、子 agent、工具调用和内部文档不使用该 skill。"
},
"oracle": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [],
"mcps": []
},
"librarian": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [],
"mcps": []
},
"writer": {
"model": "torchai-gpt/gpt-6-astra",
"variant": "medium",
"skills": [
"deliverable-document-writer",
"create-readme"
],
"mcps": [],
"prompt": "你是文档写作专员。负责创建、修改和整理面向用户、贡献者、项目维护者或交付对象的正式文档，包括 README、指南、API 文档、技术报告、示例和变更日志。写作前核对代码和现有文档中的事实。写完后必须按所用 skill 读回成稿，按句清除无源否定、无端假设、预防式前提、无端转折和过程泄漏，并把修改写回文件；没读回成稿、没把修改写回文件，不算完成。不要接管 project-context、.agent-context、agent memory、session notes、任务日志、内部工作记录、临时计划或其他仅供 agent 内部使用的文档。除非文档任务明确要求，不要修改业务源代码。回复调用方时说明修改了哪些文档，并必须附上写后审计（只给调用方，不得写入文档）：\\n## 写后审计\\n- 已读回成稿：是\\n- 功能异常句：删除或改写过的原句；没有则写无\\n- 抽查句：至少 5 句。格式为「原句 → 通过/删除/改写」；通过时写明读者相反预期来自哪里。优先抽否定句、如果/注意/然而/虽然、安装或对比句。缺这一节不算完成。抽查句或功能异常句里标「删除」或「改写」的，必须已经写回文件。文件仍是原稿却汇报已改，不算完成。不要把未落地的修改交给调用方。",
"orchestratorPrompt": "@writer\n- **Delegate when:** 用户明确要求创建、修改、重写、整理或生成面向用户的文档；或者 Orchestrator 判断当前实现任务为了完成交付，需要自动新建或更新 README、Markdown、指南、API 文档、技术报告、示例、变更日志或其他正式项目文档，即使用户没有单独提出文档要求。\n- **Don't delegate when:** 用户只是询问文档内容或要求解释而不需要写入文档文件；或者文档仅供 agent 内部使用，例如 project-context、.agent-context、agent memory、session notes、任务日志、内部工作记录、临时计划和中间分析。\n- **Required:** 对所有符合条件的正式文档任务，无论需求来自用户还是 Orchestrator 自己的实现计划，都必须先委托给 @writer，再创建或编辑文档文件。内部 agent 文档不受此规则限制。"
}
},
"deepseek": {
"orchestrator": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"skills": [],
"mcps": []
},
"librarian": {
"model": "deepseek/deepseek-v4-flash",
"variant": "max",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "deepseek/deepseek-v4-flash",
"variant": "max",
"skills": [],
"mcps": []
},
"designer": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"skills": [],
"mcps": []
}
},
"openai": {
"orchestrator": {
"model": "openai/gpt-5.6-sol",
"variant": "max",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "openai/gpt-5.6-sol",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "openai/gpt-5.6-sol",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "openai/gpt-5.6-sol",
"variant": "xhigh",
"skills": [],
"mcps": []
},
"librarian": {
"model": "openai/gpt-5.6-sol",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "openai/gpt-5.6-sol",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "openai/gpt-5.6-sol",
"variant": "max",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "xhigh",
"mcps": []
},
"fixer": {
"model": "openai/gpt-5.6-sol",
"variant": "max",
"skills": [],
"mcps": []
}
},
"openai-fast": {
"orchestrator": {
"model": "openai/gpt-5.6-sol-fast",
"variant": "max",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "openai/gpt-5.6-sol-fast",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "openai/gpt-5.6-sol-fast",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "openai/gpt-5.6-sol-fast",
"variant": "xhigh",
"skills": [],
"mcps": []
},
"librarian": {
"model": "openai/gpt-5.6-sol",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "openai/gpt-5.6-sol",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "openai/gpt-5.6-sol-fast",
"variant": "max",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "xhigh",
"mcps": []
},
"fixer": {
"model": "openai/gpt-5.6-sol-fast",
"variant": "max",
"skills": [],
"mcps": []
}
},
"openai-easy": {
"orchestrator": {
"model": "openai/gpt-5.6-luna",
"variant": "max",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "openai/gpt-5.6-luna",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "openai/gpt-5.6-sol",
"variant": "high",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "openai/gpt-5.6-luna",
"variant": "max",
"skills": [],
"mcps": []
},
"librarian": {
"model": "openai/gpt-5.6-luna",
"variant": "xhigh",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "openai/gpt-5.6-luna",
"variant": "xhigh",
"skills": [],
"mcps": []
},
"designer": {
"model": "openai/gpt-5.6-luna",
"variant": "max",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "openai/gpt-5.6-luna",
"variant": "max",
"skills": [],
"mcps": []
}
},
"openai-mix": {
"orchestrator": {
"model": "openai/gpt-5.6-sol",
"variant": "xhigh",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "openai/gpt-5.6-luna",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "openai/gpt-5.6-sol",
"variant": "xhigh",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "openai/gpt-5.6-sol",
"variant": "xhigh",
"skills": [],
"mcps": []
},
"librarian": {
"model": "openai/gpt-5.6-luna",
"variant": "xhigh",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "openai/gpt-5.6-luna",
"variant": "xhigh",
"skills": [],
"mcps": []
},
"designer": {
"model": "openai/gpt-5.6-luna",
"variant": "max",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "openai/gpt-5.6-luna",
"variant": "max",
"skills": [],
"mcps": []
}
},
"openai-normal": {
"orchestrator": {
"model": "openai/gpt-5.6-sol",
"variant": "xhigh",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "openai/gpt-5.6-sol",
"variant": "xhigh",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "openai/gpt-5.6-sol",
"variant": "max",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "openai/gpt-5.6-sol",
"variant": "xhigh",
"skills": [],
"mcps": []
},
"librarian": {
"model": "openai/gpt-5.6-sol",
"variant": "medium",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "openai/gpt-5.6-sol",
"variant": "medium",
"skills": [],
"mcps": []
},
"designer": {
"model": "openai/gpt-5.6-sol",
"variant": "xhigh",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "xhigh",
"mcps": []
},
"fixer": {
"model": "openai/gpt-5.6-sol",
"variant": "xhigh",
"skills": [],
"mcps": []
}
},
"grok": {
"orchestrator": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [
"*",
"!make-interfaces-feel-better"
],
"mcps": [
"*",
"!context7"
]
},
"oracle": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"momus": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [
"simplify",
"web-perf"
],
"mcps": [
"*"
]
},
"council": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [],
"mcps": []
},
"librarian": {
"model": "torchai-grok/grok-4.6",
"variant": "low",
"skills": [
"customer-research"
],
"mcps": [
"websearch",
"context7",
"grep_app"
]
},
"explorer": {
"model": "torchai-grok/grok-4.6",
"variant": "low",
"skills": [],
"mcps": []
},
"designer": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [
"ui-ux-pro-max",
"agent-browser",
"make-interfaces-feel-better",
"motion",
"better-icons"
],
"mcps": []
},
"helper": {
"model": "deepseek/deepseek-flash",
"variant": "max",
"mcps": []
},
"fixer": {
"model": "torchai-grok/grok-4.6",
"variant": "high",
"skills": [],
"mcps": []
}
}
},
"multiplexer": {
"type": "auto",
"layout": "main-vertical",
"main_pane_size": 60
},
"council": {
"presets": {
"default": {
"alpha": {
"model": "torchai-gpt-luna/gpt-5.6-luna",
"variant": "max"
},
"beta": {
"model": "deepseek/deepseek-flash",
"variant": "max"
},
"delta": {
"model": "deepseek/deepseek-v4-flash",
"variant": "max"
},
"musk": {
"model": "torchai-grok/grok-4.6",
"variant": "high"
}
}
}
}
}其中 helper 和 momus 是我自己添加的。helper 是为了搞一个 agent 来使用别的模型，5.6 增加了一个推理等级，因此为了最佳，我直接用了 max，当然也有一般工作的预设版本，使用的是 xhigh。momus 也是自己配置的，默认不会使用。
使用
omo平替
以下分界线内的内容在 2.0 版本之后已经失效
omo-slim 并没有 omo 的那些 ulw、ulw-loop 之类的指令，取而代之的是 /goal 和 auto-continue。
# goal 设定一个目标，在达成之前不会停止。
/goal 完成下面的任务并自动用Oracle审计。
xxxxxx
开始任务吧。
# auto-continue 在 todo 列表未完成的时候会自动继续，一般情况下使用 /goal 后该选项也会自动开启。
xxxxx
开始完成任务吧。
/auto-continue on分界线结束
2.0 版本之后已经没有
/goal与auto-continue命令了作者表示，当前 omo-slim 会自动 continue，并且加入了
/deepwork这个重型指令来代替/goal。也就是说，由于现在的子 agent 任务的异步化，主 agent 会被自动唤醒，并且现在任务或多或少都被当成了 “goal”。
/deepwork
Deepwork 适用于重型编码会话：大范围重构、多阶段功能、高风险架构变更，或需要持久计划的工作。它会创建本地 markdown 进度文件，使用 Oracle 评审关卡，并保持实现阶段结构化。
启动方式：
/deepwork <heavy coding task>就算你不显式使用此功能，在面对针对复杂任务的提示词时，该功能也有可能自动启用。
你也可以在每次任务后面加上“自动让 Oracle 审计”这句话，有时候一般任务 omo-slim 不会自动调用审计。当然加上之后也会让整个流程变的特别长。
更多的请查阅官方文档。
自动compact
omo-slim 与 opencode 的自动 compact 配合的不是很好，因此我建议使用 magic context 这个插件搭配 omo-slim 来弥补这个缺陷。
tmux集成
omo-slim 安装之后会自动开启 tmux 集成，你可以看 omo 的那个教程来感受 tmux 集成的效果，并且 配置这个 来让 tmux 集成能正确显示出来。
问题
2.0无法自动更新
该问题有可能已经被解决
更新到 2.0 之后版本可能会出现和 omo 一样的问题，再检查确认了插件缓存目录还是旧版本之后：
可以删除 oh-my-opencode-slim@latest 目录然后启动 opencode 等他重新拉一个新版的插件。
流程突然停止
在 2.2.3 附近的某个版本中，omo-slim 在分配后台子任务的时候，有可能会卡住并迟迟无法进行下一步。这时，有可能会产生前端显示后台任务已经完成但是主 agent 依然认为其在运行中，没有停止。
https://github.com/alvinunreal/oh-my-opencode-slim/blob/master/docs/background-orchestration.md
这时，你需要明确跟主 agent 说“后台任务卡住，需要重试”。只有这样，主 agent 才会调用它的 cancel_task 去取消这个任务并重试，不然有可能会卡一万年。
websearch mcp 消失
omo-slim 在 2.2.10 版本移除了它自己带的 websearch mcp，于是我们改用 opencode 自带的联网搜索功能。