Oh My MiMo Code

English | 简体中文

OpenCode-style multi-agent workflows for MiMo Code.

Oh My MiMo Code is a MiMo-Code plugin and agent workflow adapter. It is based on the ideas behind Oh My OpenAgent, but it is not a blind copy: MiMo Code is built on OpenCode, yet OpenCode plugins are not directly compatible with MiMo Code. This project adapts the agent workflow, model routing, commands, and prompt design for the MiMo Code harness.

You still use the familiar mimo interactive TUI. OMM changes what sits behind it: role-based agents, workflow keywords, model routing, lightweight system prompts, and configuration templates.

Why This Exists

MiMo Code is getting attention fast, but there is a practical gap:

• MiMo Code is based on OpenCode.
• OpenCode plugin workflows do not run in MiMo Code as-is.
• Heavy system prompts and large tool schemas can stress the MiMo Code harness.
• Complex coding tasks need planning, execution, critique, search, and verification as separate behaviors.

Oh My MiMo Code fills that gap by adapting a stronger agent workflow layer specifically for MiMo Code.

What You Get

Capability	What it means
MiMo TUI enhancement	Keep using mimo; no strange new CLI entry point.
Multi-agent roles	Main worker, deep executor, planner, critic, librarian, code explorer, and more.
Workflow keywords	Use ultrawork, ulw, team, hyperplan, analyze, and similar prompts inside the TUI.
Model routing	Assign different models to different agents, with optional fallbacks.
Lightweight MiMo harness prompts	Reduce oversized system prompt and tool-schema pressure.
Chinese mythology naming	User-facing role names are friendlier, while runtime names stay ASCII-compatible.
Compatibility templates	Plugin and agent/model config templates for MiMo Code.
MiMo alias patch	Optional compatibility patch for MiMo-Code checkouts that do not resolve the new agent aliases yet.

Core Ideas

Area	Description
Autonomous main agent	Yugong - ultraworker keeps pushing toward the goal, edits code, and verifies results.
Deep execution	Luban - Deep Agent is for cross-file implementation, complex fixes, and repository exploration.
Planning and critique	Fuxi builds plans; Xiezhi challenges weak assumptions and missing tests.
Research and code lookup	librarian is for external docs; explore is for local code search and call-path mapping.
Model routing	Use stable models for main work, stronger coding models for deep execution, and faster models for search.
Context slimming	The default MiMo path avoids pushing every heavy delegation tool into the main agent prompt.

Quick Start

The easiest path is to give this repository URL to MiMo Code and ask it to install OMM using prompts/setup-with-mimo-code.md.

It is the agent era, after all. You probably do not want to install an agent plugin by hand if the agent can do it for you.

You can say:

Open this GitHub repository and follow prompts/setup-with-mimo-code.md to install and initialize Oh My MiMo Code:
<paste the OMM GitHub URL here>

MiMo Code should clone, build, install the wrapper, write config files, and verify the installation. The default model is mimo/mimo-auto; MiMo Code only needs to ask more questions if you want different agents to use different providers or models.

Manual installation:

cd oh-my-openagent
bun install
bun run build
cd ..
MIMOCODE_SOURCE_DIR=/absolute/path/to/MiMo-Code/packages/opencode \
  ./scripts/install-mimo-wrapper.sh

The installer:

• Creates a ~/bin/mimo launcher.
• Writes ~/.config/mimocode/mimocode.jsonc so MiMo Code loads this plugin.
• Writes ~/.config/mimocode/oh-my-openagent.jsonc with the default mimo/mimo-auto agent model.
• Backs up existing config files before replacing them.

If your MiMo-Code checkout does not resolve aliases such as sisyphus -> Yugong, apply the compatibility patch:

cd /absolute/path/to/MiMo-Code
patch -p1 < /absolute/path/to/oh-my-mimo-code/patches/mimocode-agent-aliases.patch

Make sure ~/bin is in your shell PATH:

export PATH="$HOME/bin:$PATH"

Then open the normal MiMo Code TUI:

mimo

Inside the TUI, switch to Yugong - ultraworker from /agent if needed.

Model Configuration

The recommended setup flow is not to hand-edit JSON from scratch. Give the repository link to MiMo Code, let it read prompts/setup-with-mimo-code.md, and let it ask what providers/models you have.

If you are unsure, keep the default:

{
  "agents": {
    "sisyphus": {
      "model": "mimo/mimo-auto"
    }
  }
}

Advanced example:

{
  "agents": {
    "sisyphus": {
      "model": "mimo/mimo-auto"
    },
    "hephaestus": {
      "model": "openai/gpt-5.5",
      "variant": "high"
    },
    "librarian": {
      "model": "google/gemini-3-flash"
    },
    "explore": {
      "model": "github-copilot/grok-code-fast-1"
    }
  }
}

Fallback models are supported:

{
  "agents": {
    "sisyphus": {
      "model": "mimo/mimo-auto",
      "fallback_models": [
        "openai/gpt-5.5",
        { "model": "google/gemini-3-pro", "variant": "high" }
      ]
    }
  }
}

Common TUI Entries

Entry	Purpose
mimo	Open the MiMo Code interactive TUI.
mimo /path/to/project	Open the TUI in a specific project directory.
/agent	Switch agents. The default main agent is Yugong - ultraworker.
/model	Switch models. Default config uses mimo/mimo-auto.
/goal	Define a completion target for long-running tasks.
/dream	Distill reusable memory from recent sessions.
/distill	Turn repeated workflows into skills, subagents, or commands.
/voice	Voice input, depending on MiMo login and runtime support.

OMM Workflow Keywords

These are not shell commands. Open mimo, then send them as normal messages to the agent.

Keyword	Use it when	What it does
ultrawork / ulw	You want high-intensity progress without manually splitting the task.	Clarifies acceptance criteria, separates exploration/implementation/verification, and keeps pushing.
team	The task can be split across multiple lanes.	Uses a team-mode style coordination pattern when the current MiMo Code runtime supports it.
hyperplan	You have a plan but want it challenged first.	Runs critical planning views to find hidden constraints, risks, and missing tests.
search	The task needs external docs or community references.	Routes toward research instead of guessing from memory.
analyze	You need the project mapped before edits.	Reads the codebase first, then reports structure and likely change points.

Slash Commands

Command	Use it when	What it does
/init-deep	A project is new or deeply nested.	Builds layered AGENTS.md project context.
/start-work [plan-name]	You already have a plan.	Starts execution from a planning artifact.
/ralph-loop "goal"	A long task should keep moving.	Runs a persistent completion loop until done, capped, or cancelled.
/ulw-loop "goal"	You want looped progress plus ultrawork intensity.	Combines Ralph-style continuation with ultrawork-style parallelism.
/cancel-ralph	A Ralph/ulw loop is going the wrong way.	Cancels the active loop.
/stop-continuation	You do not want continuation systems taking over.	Stops continuation mechanisms.
/handoff	You need to continue in another session.	Generates a compact handoff with status, files, and next steps.
/refactor <target>	You want structured refactoring.	Uses code intelligence, architecture review, and tests where available.
/remove-deadcode	You want unused code removed carefully.	Combines static scans and reference checks before deletion.
/security-research	You want security-focused review.	Checks for injection, auth, config leaks, and other risk classes.
/get-unpublished-changes	You are preparing a release.	Compares published versions against current HEAD and summarizes impact.
/publish <patch|minor|major>	You maintain the package release.	Runs release checks and validates GitHub/npm results.
/omomomo	You want the small easter egg.	Prints the inherited project easter egg.

Agents

The project keeps legacy configuration keys, but the main display names use Chinese mythology / legend-inspired names. Runtime names stay ASCII because agent names may enter HTTP headers.

Config key	Chinese name	Display name	Best for
sisyphus	愚公	Yugong - ultraworker	Default main worker: understand, edit, verify, and keep moving.
hephaestus	鲁班	Luban - Deep Agent	Deep implementation and cross-file changes.
prometheus	伏羲	Fuxi - Plan Builder	Requirement clarification and planning.
atlas	大禹	Dayu - Plan Executor	Executing existing plans, TODOs, PRDs, or test specs.
sisyphus-junior	小愚公	Xiao-Yugong	Small, focused execution tasks.
metis	九天玄女	Jiutian-Xuannu - Plan Consultant	Finding hidden constraints and plan gaps.
momus	獬豸	Xiezhi - Plan Critic	Challenging vague requirements, logic issues, and missing tests.
athena	西王母	Xiwangmu - Council	Multi-perspective review and major decision checks.
athena-junior	小西王母	Xiwangmu-Junior - Council	Lighter review for smaller scopes.
oracle	神谕	oracle	Architecture, diagnostics, and hard problems.
librarian	藏书官	librarian	External docs, API references, and community research.
explore	探路者	explore	Local code search, file lookup, and call-path mapping.
multimodal-looker	观象者	multimodal-looker	Image, screenshot, and visual material analysis.

Choosing an Agent

Situation	Recommended agent
I just want it to finish the task	sisyphus / Yugong - ultraworker
The codebase needs deep exploration	hephaestus / Luban - Deep Agent
Requirements are unclear	prometheus / Fuxi - Plan Builder
I already have a plan	atlas / Dayu - Plan Executor
I want the plan challenged	momus / Xiezhi - Plan Critic
I need external docs	librarian
I only need code lookup	explore

MiMo Compact Mode

MiMo Code can be sensitive to context size, system prompt weight, and tool schema volume. OMM therefore avoids stuffing every heavy multi-agent capability into the default main agent.

By default, Yugong - ultraworker:

• Uses a shorter system prompt.
• Keeps core direct execution, verification, and code-editing behavior.
• Disables heavier nested delegation, team delegation, and background delegation by default.
• Preserves legacy agent keys and aliases so existing configs still work.

This is not a downgrade. It is prompt weight control for the MiMo Code harness.

Troubleshooting

Symptom	What to check
Yugong - ultraworker does not appear in the TUI	Run bun run build and verify mimocode.jsonc points to oh-my-openagent/dist/index.js.
sisyphus does not resolve to Yugong	Apply patches/mimocode-agent-aliases.patch to the MiMo-Code checkout.
ConnectionRefused appears	Usually a MiMo hosted service/free-channel issue, not proof that the plugin failed to load.
You do not want old config overwritten	The installer writes .bak.<timestamp> backups first.

License

• Root-level docs, config templates, and scripts use the MIT License. See LICENSE.
• oh-my-openagent/ keeps its own SUL-1.0 License.

See LICENSES.md for the detailed license boundary.

Contributing

Issues and PRs are welcome. Before changing code, read CONTRIBUTING.md, especially the verification commands and the rule about not committing local runtime state.

In One Sentence

Oh My MiMo Code makes MiMo Code feel more like a practical agent workbench: clear entry point, clear roles, clear model routing, and clearer debugging when something goes wrong.