Oh My OpenAgent: The Complete Guide to the Ultimate Agent Harness
Oh My OpenAgent (omo) is a TypeScript-based agent harness that turns AI coding agents into a disciplined engineering team — with named specialist agents, hash-anchored edits, LSP/AST tools, skill-embedded MCPs, and full Claude Code compatibility. Type ultrawork and everything works. 38,100+ stars, 150 releases.
What Is Oh My OpenAgent?
An agent harness (previously oh-my-opencode) that solves the "harness problem" — most agent failures aren't from the model, they're from the edit tool. omo provides hash-anchored edits, multi-model orchestration with named discipline agents, and world-class IDE-level tools.
- Language: TypeScript
- Stars: 38,100+ ⭐
- Forks: 2,873
- Releases: 150
- Homepage: ohmyopencode.org
- Compatible With: Claude Code, OpenCode, OpenAI, Anthropic, Gemini, Kimi
The ultrawork Command
ultrawork # or ulw
One command. Everything below — every feature, every optimization — just works. No configuration.
Discipline Agents
omo's killer feature: named specialist agents, each tuned to its model's strengths.
| Agent | Model | Role |
|---|---|---|
| Sisyphus | claude-opus-4-6 / kimi-k2.5 / glm-5 | Main orchestrator — plans, delegates, drives tasks to completion with parallel execution |
| Hephaestus | gpt-5.3-codex | Deep worker — autonomous end-to-end execution, explores codebase without hand-holding |
| Prometheus | claude-opus-4-6 / kimi-k2.5 / glm-5 | Strategic planner — interviews you, identifies scope, builds verified plan before coding |
| Oracle | - | Architecture/debugging specialist |
| Librarian | - | Docs and code search |
| Explore | - | Fast codebase grep |
| Multimodal Looker | - | Visual analysis |
Agent says what kind of work → harness picks the right model → you touch nothing.
Hash-Anchored Edits (LINE#ID)
The breakthrough that took Grok Code Fast 1 from 6.7% → 68.3% success rate — just by changing the edit tool.
11#VK| function hello() {
22#XJ| return "world";
33#MB| }
Every line comes tagged with a content hash. If the file changed, the hash won't match and the edit is rejected before corruption. No whitespace reproduction errors. No stale-line bugs.
World-Class Tools
| Tool | Capability |
|---|---|
| LSP | lsp_rename, lsp_goto_definition, lsp_find_references, lsp_diagnostics — IDE precision |
| AST-Grep | Pattern-aware code search and rewriting across 25 languages |
| Tmux | Full interactive terminal — REPLs, debuggers, TUI apps |
| Built-in MCPs | websearch (Exa), context7 (docs), grep_app (GitHub search) |
Skill-Embedded MCPs
MCP servers eat your context budget. omo fixed that:
- Skills bring their own MCP servers
- Spin up on-demand, scoped to task
- Gone when done — context window stays clean
Built-in skills: playwright (browser automation), git-master (atomic commits, rebase), frontend-ui-ux (design-first UI).
Deep Initialization (/init-deep)
Generates hierarchical AGENTS.md files throughout your project:
project/
├── AGENTS.md ← project-wide context
├── src/
│ ├── AGENTS.md ← src-specific context
│ └── components/
│ └── AGENTS.md ← component-specific context
Agents auto-read relevant context. Zero manual management.
Claude Code Compatibility
Full compatibility — every hook, command, skill, MCP, and plugin works unchanged. If you dialed in your Claude Code setup, it transfers directly.
Oh My OpenAgent vs Alternatives
Category: This is an agent harness / AI coding agent enhancement layer.
| Feature | Oh My OpenAgent | Claude Code (vanilla) | CC Switch |
|---|---|---|---|
| Focus | Agent harness with discipline agents | Official CLI coding agent | Multi-tool provider manager |
| Stars | 38.1K ⭐ | Anthropic product | 25.3K ⭐ |
| Discipline Agents | ✅ Sisyphus/Hephaestus/Prometheus | ❌ Single agent | ❌ Not an agent |
| Hash-Anchored Edits | ✅ 6.7→68.3% success | ❌ Standard edits | ❌ |
| Multi-Model Orchestration | ✅ Auto model-matching | ❌ Single model | ✅ Provider switching |
| LSP Integration | ✅ rename/goto/refs/diagnostics | ❌ | ❌ |
| AST-Grep | ✅ 25 languages | ❌ | ❌ |
| Skill-Embedded MCPs | ✅ On-demand, scoped | ❌ | ✅ MCP management |
| Background Agents | ✅ Parallel execution | ❌ | ❌ |
| Deep Init | ✅ /init-deep | ❌ | ❌ |
| Claude Code Compatible | ✅ Full | ✅ Native | ✅ Config management |
| 25+ Built-in Hooks | ✅ | ✅ | ❌ |
| JSONC Config | ✅ | JSON | ❌ |
When to choose Oh My OpenAgent: You want the most advanced agent harness — discipline agents for multi-model orchestration, hash-anchored edits for reliable code changes, LSP/AST-Grep tools, skill-embedded MCPs, and Claude Code compatibility.
When to choose vanilla Claude Code: You want the official Anthropic experience without enhancement layers.
When to choose CC Switch: You need a visual desktop manager for switching between multiple CLI tools and providers.
Conclusion
Oh My OpenAgent is the most ambitious agent harness in the ecosystem. With named discipline agents (Sisyphus, Hephaestus, Prometheus), hash-anchored edits that 10x edit success rates, IDE-level LSP/AST tools, skill-embedded on-demand MCPs, deep initialization, background parallel agents, and full Claude Code compatibility — it transforms AI coding from a single-agent tool into a disciplined engineering team. At 38.1K stars with 150 releases, it's production-hardened and relentlessly iterated.