oh-my-openagent (omo) — Summary

oh-my-openagent (omo) by code-yeongyu is the largest framework in this batch (59,562 stars, 30 contributors), originally named oh-my-opencode and undergoing a multi-harness refactor to support OpenCode, Codex, Pi, and other agent runtimes. Built as a native OpenCode plugin (TypeScript + Bun), it ships 11 named agent personas (Sisyphus, Hephaestus, Oracle, Librarian, Explore, Atlas, Prometheus, Metis, Momus, Multimodal-Looker, Sisyphus-Junior), ~52 lifecycle hooks, 20-39 tools (config-gated), a 3-tier MCP system, the Hashline LINE#ID edit tool, IntentGate keyword detection, Team Mode (lead + up to 8 parallel members in tmux), and a Boulder feature for work tracking. The flagship model assignment is Sisyphus (planner/orchestrator) on claude-opus-4-7/kimi-k2.6/glm-5.1 and Hephaestus (deep worker) on gpt-5.5 — making this the most aggressive multi-model framework in the batch. The framework received notable notoriety for reportedly causing Anthropic to block OpenCode due to OAuth token handling. Closest seed comparison: resembles claude-flow in its code-class subagents and MCP architecture, but targets OpenCode rather than Claude Code and adds multi-provider model routing as a first-class feature.

oh-my-openagent (omo) — Overview

Origin

Created by code-yeongyu (@yeon_gyu_kim), developed and maintained by an AI assistant called Jobdori (running on a customized fork of OpenClaw). Named oh-my-opencode originally; dual-published during transition as oh-my-opencode and oh-my-openagent.

Notable Background

"Anthropic blocked OpenCode because of us. Yes, this is true. They want you locked in. Claude Code is a nice prison, but it's still a prison."

"You don't need to pay $200 for 2 hours of work. The future isn't picking one winner; it's orchestrating them all."

The framework is philosophically positioned as anti-vendor-lock-in, routing work across Claude (Anthropic), GPT-5.5 (OpenAI), Kimi K2.6 (Moonshot), and GLM-5.1 (Zhipu).

Agent Philosophy: Named Mythological Agents

• Sisyphus: main orchestrator — plans, delegates, drives to completion with parallel execution, does not stop halfway. Runs on claude-opus-4-7/kimi-k2.6/glm-5.1
• Hephaestus: autonomous deep worker — explores codebase, researches, executes end-to-end. Runs on gpt-5.5. Nicknamed "The Legitimate Craftsman" (irony at Anthropic's "legitimate" framing)
• Prometheus: strategic planner — interview mode, asks questions before any code. Runs on claude-opus-4-7/kimi-k2.6/glm-5.1
• Oracle: research specialist
• Librarian: documentation/knowledge specialist
• Explore: codebase exploration specialist
• Atlas: unknown
• Metis: craftsmanship/quality
• Momus: critic/adversarial reviewer
• Multimodal-Looker: visual/multimodal
• Sisyphus-Junior: lightweight orchestration

Key Differentiators

1. Multi-provider model routing: hard-assigns specific models to specific agent roles
2. Hashline LINE#ID edit tool: content-hash validated edits eliminating stale-line errors
3. IntentGate: analyzes true user intent before classifying/acting
4. Team Mode: tmux-based lead + 8 parallel members with real-time visualization
5. Boulder: work tracking state machine with CLI (cli/boulder)
6. Comment Checker: prevents AI-generated filler in code comments
7. Skill-Embedded MCPs: each skill carries its own MCP servers (no context bloat)
8. LSP + AST-Grep via built-in MCP servers

oh-my-openagent (omo) — Architecture

Distribution

• npm package: oh-my-opencode (dual-published as oh-my-openagent during transition)
• Version: 4.5.0
• Primary runtime: OpenCode plugin (TypeScript + Bun)
• Platform binaries: pre-built for darwin-arm64, darwin-x64, linux-arm64, linux-x64, linux-musl variants, windows-x64

Install

Copy and paste to LLM 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

Or fetch install guide via curl:

curl -s https://raw.githubusercontent.com/code-yeongyu/oh-my-openagent/refs/heads/dev/docs/guide/installation.md

Required Runtime

• OpenCode (primary)
• Bun (build)
• Claude Code (compatibility layer)
• tmux (Team Mode)

Directory Tree

oh-my-openagent/
├── src/
│   ├── index.ts              — Plugin entry (18-line wrapper)
│   ├── plugin-config.ts      — JSONC multi-level config (Zod v4)
│   ├── plugin-interface.ts   — 11 OpenCode hook handlers
│   ├── create-managers.ts    — 4 managers (Tmux, Background, SkillMcp, ConfigHandler)
│   ├── create-tools.ts       — ToolRegistry
│   ├── create-hooks.ts       — 5-tier hook composition
│   ├── agents/               — 11 agent definitions
│   ├── hooks/                — ~52 lifecycle hooks (57 dirs)
│   ├── tools/                — 13 native tool dirs
│   ├── features/             — 20 feature modules
│   ├── shared/               — 297 utility files
│   ├── config/               — Zod v4 schema (30 files)
│   ├── cli/                  — install, run, doctor, boulder, mcp-oauth
│   └── mcp/                  — 5 built-in MCPs (3 remote + lsp + ast_grep)
├── packages/
│   ├── oh-my-opencode-{platform}/ — platform binaries (11)
│   ├── lsp-tools-mcp/        — LSP MCP server
│   ├── ast-grep-mcp/         — AST grep MCP server
│   ├── boulder-state/        — work tracking state machine
│   ├── hashline-core/        — Hashline edit primitives
│   ├── model-core/           — model resolution pipeline
│   ├── prompts-core/         — prompt loading + bundled mode prompts
│   ├── rules-engine/         — rule discovery + matching
│   └── web/                  — Marketing site (Next.js 15 + Cloudflare Workers)
├── bin/                      — platform-detection shim
├── .opencode/                — project-scope skills + commands
├── .omo/                     — AI agent workspace (plans, tasks)
└── docs/                     — user guide, reference, legal

Target AI Tools

• Primary: OpenCode
• Secondary: Claude Code (compatibility layer)
• Supports: any MCP client (3-tier MCP system)

State Root

.omo/ — agent workspace (run-continuation, plans, tasks, notepads)

oh-my-openagent (omo) — Components

11 OpenCode Hook Handlers (Plugin Interface)

Agent Personas (11)

Tools (Always-on, 20)

lsp_goto_definition, lsp_find_references, lsp_symbols, lsp_diagnostics, lsp_prepare_rename, lsp_rename, grep, glob, ast_grep_search, ast_grep_replace, session_list, session_read, session_search, session_info, background_output, background_cancel, call_omo_agent, task (delegate), skill, skill_mcp

Built-in MCP Servers (5)

• LSP tools MCP (local stdio)
• AST-grep MCP (local stdio)
• 3 remote MCPs: Exa (web search), Context7 (official docs), Grep.app (GitHub search)

Skills (ultrawork, search, analyze, team, hyperplan, security-research + more)

Loaded from .opencode/ project-scope skills.

Boulder (Work Tracking)

CLI subcommand: cli/boulder
State machine: packages/boulder-state/
Commands: boulder create, list, complete, status, split, etc.

Hook Presets (~52 hooks in 57 dirs)

Categories: session, tool-guard, transform, continuation, skill
Notable: zauc-mocks (5), team-session-events (1)

oh-my-openagent (omo) — Uniqueness

Differs From Seeds

Closest seeds are claude-flow (code-class subagents, MCP architecture, multi-agent coordination) and superpowers (skills-based behavioral framework). Unlike claude-flow, omo targets OpenCode rather than Claude Code and uses real model diversity (Claude, GPT-5.5, Kimi, GLM) rather than a single provider. Unlike superpowers, omo ships a full code-class agent architecture with 13 OpenCode hook handlers (not skill files), explicit model assignments, and the Hashline edit tool for correctness. omo is the only framework in the corpus with a documented Anthropic policy incident (alleged blocking of OpenCode due to this tool's OAuth usage patterns), though the framework claims compliance through subprocess spawning rather than direct API access.

Positioning

• Highest star count in the batch (59,562)
• Primary competitive claim: escape vendor lock-in, orchestrate across Claude/GPT-5.5/Kimi/GLM
• Boulder work tracking is unique in the batch — a state machine for tasks, not just markdown files
• Hashline LINE#ID edit tool is a technical contribution not present in any seed or other batch member
• /init-deep for hierarchical AGENTS.md generation is a unique token-efficiency feature

Observable Failure Modes

• Multi-harness refactor in progress (AGENTS.md explicitly warns: "DO NOT TRUST THE STRUCTURE BELOW AS STABLE")
• Anthropic policy incident: claimed blocking event creates adoption uncertainty
• Package naming confusion: dual-published as both oh-my-opencode and oh-my-openagent during transition
• Anonymous telemetry: default-on PostHog tracking may concern privacy-conscious users
• ~313k LOC / 2167 TypeScript files: significantly more complex than other batch members; high audit surface
• Bun runtime dependency: limits deployment environments vs pure Node.js

Canonical Status

Not a derivative of the oh-my-codex naming line — independent project that targets OpenCode rather than Codex CLI. The naming similarity ("oh-my-...") is intentional but architecturally distinct.