What it does

oh-my-openagent is a plugin layer for AI coding agents. The “Ultimate” edition bolts onto OpenCode with 11 agents, 54+ lifecycle hooks, five built-in MCPs, and modes like ultrawork and team. A lighter edition ports a subset—rules, LSP, ultrawork, ulw-loop—into Codex CLI as a plugin. Installation is via bunx or npx; the README half-jokingly recommends letting another LLM handle the setup because humans “fat-finger” the subscription detection and provider auth matrix.

The interesting bit

The project treats agent orchestration as an operating system problem, not a chat interface problem. It also wears its politics on its sleeve: the README claims Anthropic “blocked OpenCode because of us” and frames Claude Code as “a nice prison.” Whether you find that refreshing or grating, the architecture is genuinely multi-provider—Anthropic, Gemini, Copilot, Z.ai, OpenCode Zen—with an explicit bet that no single model wins.

Key highlights

• Two editions: Ultimate (OpenCode) and Light (Codex CLI), installable separately or together
• 11 agents with model-to-agent matching matrix; modes include ultrawork, search, analyze, team, hyperplan
• Team Mode for multi-agent collaboration (demoed with Kimi K2.6 + GPT-5.5)
• Built-in telemetry, SHA256-hashed, opt-out via env vars; no PostHog person profiles
• Maintained in public by “Jobdori,” an AI assistant on a customized OpenClaw fork

Caveats

• The project is mid-rename (oh-my-opencode → oh-my-openagent) and warns against using the short omo alias with bunx because it may resolve a different npm package entirely
• A “Multi-Harness Agent OS Refactor” is in progress; the roadmap is shifting toward Pi and other harnesses
• License is SUL-1.0, not a standard OSI license

Verdict

Worth a look if you’re already living in OpenCode or Codex CLI and want opinionated, multi-agent orchestration without vendor lock-in. Skip it if you want a stable, well-documented tool with a calm maintenance story—you’re installing a moving target built by an AI assistant in real-time.