Overview

Relevant source files

Purpose and Scope

This page introduces oh-my-openagent (formerly oh-my-opencode) as an OpenCode plugin and explains its architecture, initialization flow, and core components. It provides a high-level understanding of how the plugin extends OpenCode with multi-model AI agent orchestration, 54–61 lifecycle hooks, 20–39 tools, and batteries-included tooling for code editing, task delegation, and background execution.

For installation and setup instructions, see Quick Start. For detailed agent behavior and model matching, see Key Concepts and Agents. For tool usage patterns, see Tools and Features.

Sources: README.md103-110

What is Oh My OpenAgent?

Oh-my-openagent (npm package: oh-my-opencode, dual-published as oh-my-openagent during its rename transition AGENTS.md48) is an OpenCode plugin that transforms the base platform into a multi-agent orchestration system. It provides:

11 specialized agents with role-specific prompts and tool permissions (Sisyphus, Hephaestus, Oracle, Librarian, Explore, Atlas, Prometheus, Metis, Momus, Multimodal-Looker, Sisyphus-Junior) docs/reference/features.md4-6
54–61 lifecycle hooks organized into tiers (Session, Tool-Guard, Transform, Continuation, Skill) AGENTS.md48
20–39 built-in tools including native search, task delegation, background execution, and session management AGENTS.md48
Hash-anchored editing via the hashline_edit tool, which validates LINE#ID content hashing to ensure modification integrity AGENTS.md48
3-tier MCP system merging built-in Tier-1 (LSP, AST-grep), Claude Code, and skill-embedded MCP servers AGENTS.md48
Team Mode for parallel multi-agent coordination, enabling complex delegation across member sessions AGENTS.md48

The plugin targets technical users who require reliable multi-model orchestration with automatic fallback chains and zero-configuration agent-to-model matching. It ships in two editions: Ultimate (for OpenCode) and Light (for Codex CLI) docs/guide/installation.md3-8

Sources: README.md103-110 AGENTS.md48 docs/guide/installation.md3-8 docs/reference/features.md4-6

System Architecture at a Glance

Plugin initialization and component assembly

This diagram shows how the pluginModule at packages/omo-opencode/src/index.ts4-6 serves as the entry point, delegating to createPluginModule packages/omo-opencode/src/index.ts4 The initialization flow moves from config loading to manager creation, tool registration, and hook composition, finally assembling the PluginInterface AGENTS.md58-72

Sources: packages/omo-opencode/src/index.ts1-19 AGENTS.md58-72

Plugin Lifecycle

Plugin initialization sequence from load to runtime

The plugin follows a strict 7-step initialization order starting from pluginModule.server AGENTS.md69-80 Configuration loading merges user and project settings and validates via Zod. Managers are created to handle background tasks, terminal multiplexing, and MCPs AGENTS.md48

Sources: packages/omo-opencode/src/index.ts1-19 AGENTS.md58-72

Core Components

Agents (11 total)

The plugin provides 11 specialized agents, each with role-specific prompts and model configurations:

Agent	Purpose	Mode	Recommended Model
Sisyphus	Main lead orchestrator	primary	Claude Opus / Kimi K2.6
Hephaestus	Autonomous deep worker	primary	GPT-5.5
Atlas	Todo-list orchestrator	primary	Claude Sonnet 4.6
Prometheus	Strategic planner (interview mode)	primary	Claude Opus
Oracle	Read-only architecture consultant	subagent	GPT-5.5
Librarian	External documentation search	subagent	Gemini 3 Flash
Explore	Fast codebase exploration	subagent	GPT-5.4-mini-fast
Multimodal-Looker	Visual content specialist	subagent	GPT-5.5
Metis	Plan consultant	primary	Claude Sonnet 4.6
Momus	Plan reviewer	primary	GPT-5.5
Sisyphus-Junior	Category-spawned executor	subagent	Category-dependent

For detailed agent behavior, see Agents.

Sources: docs/reference/features.md4-33 docs/guide/agent-model-matching.md13-39

Tools (20–39 total)

Tools are organized into functional categories gated by configuration:

Always On: grep, glob, task (delegate), session_read, skill, background_cancel AGENTS.md48
LSP & AST-grep: Served via built-in Tier-1 MCPs (lsp_rename, ast_grep_search) AGENTS.md48
Conditional: look_at (vision), interactive_bash (tmux integration), hashline_edit (hash-anchored code edits), and team_* tools AGENTS.md48

For detailed tool usage, see Tools and Features.

Sources: AGENTS.md48

Hooks (54–61 total)

Hooks intercept OpenCode events at multiple tiers AGENTS.md48:

Session Hooks: Lifecycle management such as session create/delete/idle packages/omo-opencode/src/hooks/AGENTS.md1
Tool Guard Hooks: Enforce safety and policy during tool execution packages/omo-opencode/src/hooks/AGENTS.md1
Transform Hooks: Modify chat messages, e.g. implementing the ultrawork keyword detector packages/omo-opencode/src/hooks/AGENTS.md1
Continuation Hooks: Manage task continuation loops like createTodoContinuationEnforcer packages/omo-opencode/src/hooks/AGENTS.md1

For hook internals, see Hook System.

Sources: AGENTS.md48 packages/omo-opencode/src/hooks/AGENTS.md1

Configuration System

Multi-Level Merging

Configuration is loaded and merged across levels docs/reference/configuration.md44-58:

Defaults: Built-in system defaults.
User config: Global settings (e.g., ~/.config/opencode/oh-my-openagent.jsonc).
Project config: Local workspace settings (e.g., .opencode/oh-my-openagent.jsonc).

Sources: AGENTS.md48 docs/reference/configuration.md44-58

Configuration Schema

The configuration provides extensive customization for agents and features:

Field Category	Purpose
Disabling Features	Arrays controlling disabled MCPs, agents, skills, hooks, tools assets/oh-my-opencode.schema.json32-84
Agent Overrides	Model, fallback_models, temperature, and per-agent thinking budget overrides assets/oh-my-opencode.schema.json108-183
Behavior Flags	Flags such as `hashline_edit`, `model_fallback`, and `new_task_system_enabled` assets/oh-my-opencode.schema.json11-105

For full configuration details, see Configuration Reference.

Sources: assets/oh-my-opencode.schema.json1-101

Key Workflows

Ultrawork Mode

The ultrawork (or ulw) command engages the Sisyphus orchestrator in "just do it" mode, running multiple agents aggressively in parallel to accomplish the requested task autonomously, until completion README.md109

Planning Workflow

Using /start-work initiates strategic planning with the Prometheus agent in interview mode to break down and clarify requirements before execution docs/reference/features.md22

Hash-Anchored Editing

The hashline_edit tool enforces editing safety by verifying line content hashes (LINE#ID), preventing accidental modification mismatches and race conditions when applying code changes AGENTS.md48

Sources: README.md109 docs/reference/features.md22 AGENTS.md48

Quick Start

To get started quickly with oh-my-openagent, follow these steps:

Installation

The recommended install method is to have an LLM agent perform Ultimate installation for you, automating subscription detection, model configuration, provider authentication, and config migration:

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

This prompt can be pasted into your LLM session (e.g., Claude Code, AmpCode, Cursor) to handle setup.

Sources: docs/guide/installation.md20-29

Alternatively, see Installation and Setup for manual installation and provider authentication guidance.

Your First `ultrawork` Command

After installation, initiate autonomous orchestration by running:

ultrawork

This command starts the Sisyphus agent to plan, delegate, and execute the full work cycle automatically.

Press Tab during execution to enter interview-style planning with Planning Agents if desired, then resume with /start-work for full orchestration.

Sources: README.md109

For further details on these and other core ideas, see the Quick Start and Key Concepts pages.