Oh My OpenAgent

Oh My OpenAgent is a battle-tested, open-source agent harness built on top of OpenCode that orchestrates specialized AI agents for complex software engineering tasks. With 39.4k GitHub stars and 1.2M+ downloads, it provides a production-proven framework for autonomous coding workflows. The system separates planning from execution, routes tasks to the right model based on intent, and verifies results independently — all from a single CLI command.

• Ultra Work Mode (ulw) — Type three letters to trigger auto-planning, deep research, parallel agents, and self-correction loops with zero babysitting.
• Sisyphus Orchestrator — The primary orchestrator that parses implicit requirements, maps your codebase, delegates to specialists, and resumes interrupted sessions via boulder.json.
• Prometheus & Atlas Pipeline — Prometheus handles strategic planning (never writes code), while Atlas executes verified plans and accumulates wisdom across tasks.
• Hephaestus Deep Worker — Fires 2–5 parallel explore agents before writing any code, built for deep architectural reasoning and complex debugging.
• 11 Specialized Agents — Includes Oracle (architecture), Librarian (docs search), Explore (codebase grep), Metis (plan gap analysis), Momus (ruthless plan review), and dynamic on-demand agents.
• Category & Skill Routing — Intent-based routing maps requests to categories like ultrabrain, visual-engineering, quick, or writing, each backed by the optimal model.
• 40+ Lifecycle Hooks — Extensible hook system covering session recovery, context compaction, comment checking, background notifications, and more.
• MCP Integrations — Built-in MCP servers for web search (Exa), documentation retrieval (Context7), and GitHub code search (grep_app).
• Browser Automation — Supports Playwright for full browser automation and agent-browser for lightweight scraping and lookups.
• Model Optimization — Expensive models (Claude Opus, GPT-5 Codex) for planning; cheap models (Claude Haiku, Gemini Flash) for routine work to maximize output per dollar.
• Session Continuity — Boulder system tracks active work so interrupted sessions resume exactly where they stopped with zero context loss.
• Highly Configurable — Project-level and user-level JSON config files with JSONC support; override models, prompts, permissions, temperature, and more per agent.