Oh My OpenAgent Makes One Keyword Run Your Whole Agent Setup

Running Codex, OpenCode, and a pile of models means configuring each one and hoping the pieces cooperate. Oh My OpenAgent (OmO) — 69k stars — is the pre-configured alternative: install it, type ultrawork (or ulw), and the harness routes your task across its agent team until the work is done — the README's framing is blunt: "One word. Every agent activates."

Why This Skill Matters

OmO ships in three editions of the same product. Ultimate is the full build for OpenCode: 11 agents, 54+ lifecycle hooks, and four built-in MCPs — websearch, context7 (docs), grep_app (GitHub code search), and lsp — with IDE-grade tools like LSP rename and goto-definition, AST-Grep pattern search across 25 languages, and a full tmux terminal. Light carries the portable components into Codex CLI — rules, comment-checker, git-bash, LSP, ultrawork, ulw-loop, ulw-execute-continuation, and telemetry at the core, plus teammode and supporting components — with plugin-scoped MCPs for grep_app, context7, git_bash, and lsp, and the shared ast-grep skill. Senpi is a standalone beta edition that runs as the native omo command, installed from omo-ai@beta.

One naming trap deserves its own paragraph. The published npm package is still oh-my-opencode, dual-published as oh-my-openagent during the rename — and omo on npm is a different, unrelated package by a different author. Never run bunx omo or npx omo; the README says those commands resolve to the wrong project. The right entry points are below.

Installation

For the Ultimate edition on OpenCode:

bunx oh-my-openagent install

A TUI walks you through platform detection, subscription checks, and provider auth. For the Light edition on Codex CLI, Bun is not required:

npx lazycodex-ai install

The README strongly recommends letting an LLM agent run the Ultimate install, because subscription detection and per-provider authentication are easy to fat-finger. Paste this into Claude Code, Cursor, or any agent:

Install and configure oh-my-openagent by following the instructions here:
https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/refs/heads/dev/docs/guide/installation.md

Real Workflow: Hand Over a Task With One Keyword

1. Install the Ultimate edition and start an OpenCode session in your project.
2. Prefix your task with the keyword:

ulw refactor the auth module and keep every test green

IntentGate detects the ultrawork/ulw keyword (keyword detection, not semantic classification) and the full pipeline activates: the main agent plans, delegates to category workers in parallel, and keeps driving until the task completes. 3. For complex work, plan first: /ulw-plan interviews you, identifies scope and ambiguities, and writes a reviewed plan to .omo/plans/; /ulw-execute then carries that plan out in the same session.

Real Workflow: Turn On Team Mode for a Hostile Review

Team Mode is off by default. Enable it in ~/.omo/omo.jsonc:

{
  "team_mode": {
    "enabled": true,
    "max_parallel_members": 4,
    "tmux_visualization": true
  }
}

Restart OpenCode and the team_* tools appear — a lead agent orchestrating up to 8 parallel members you can watch in a tmux grid. Two skills ride on top: hyperplan sends 5 hostile critics at your plan before any code is written, and security-research runs 3 vulnerability hunters plus 2 PoC engineers over the codebase in parallel.

Tips

• Delegation routes by category, not by model: visual-engineering, deep, quick, and ultrabrain map to specific models automatically — the README names GPT-6 Astra max for ultrabrain and GPT-6 Astra high for deep, both falling back to GPT-5.6 Sol.
• Hash-anchored edits are opt-in: set hashline_edit: true and every line the agent reads gets a stable LINE#ID tag, so stale edits are rejected before they corrupt a file.
• Run /init-deep once to generate hierarchical AGENTS.md files across your project tree.
• Claude Code refugees keep their setup: the Ultimate edition loads your existing hooks, commands, skills, MCPs, and plugins unchanged.
• Anonymous telemetry is on by default — one event per machine per UTC day at most. Opt out with "telemetry": false in the config or OMO_DISABLE_POSTHOG=1.

When Not to Use This

The Senpi edition is beta and beta-only. The discipline-oriented design — a Todo Enforcer that yanks idle agents back to the task — is not what you want when every diff needs line-by-line human review. The Light edition has no OpenCode team_* tools. And if your whole workflow is Claude Code itself, note what compatibility means here: your Claude Code components run inside OpenCode under OmO; this is not a Claude Code plugin.

See the leaderboard for more skills.