A coding agent harness that treats lock-in as a bug

What it does

oh-my-openagent is a plugin-style harness that sits on top of coding agents like OpenCode and Codex CLI. The Ultimate edition injects 11 agents, 54-plus lifecycle hooks, and five built-in MCPs into OpenCode, while a Light edition ports portable components such as ultrawork and lsp into Codex CLI’s narrower plugin surface. Instead of switching between Claude Code, Codex, and random OSS models, you get a single layer that can run multi-model teams and complex workflows.

The interesting bit

The maintainer is literally using another AI assistant named Jobdori to build this in public, which gives the whole project a slightly recursive vibe. More importantly, the README wears its politics on its sleeve: it claims Anthropic blocked OpenCode because of this project, and it frames Claude Code as a “nice prison” inside a $200 walled garden. That anti-lock-in stance is the real engine—oh-my-openagent is betting that models will keep getting cheaper and smarter, so the value is in orchestrating all of them rather than pledging allegiance to one.

Key highlights

• Two editions: a full Ultimate plugin for OpenCode and a stripped Light plugin for Codex CLI, each respecting their host’s constraints.
• Team Mode can run multiple models simultaneously; the README shows it operating with Kimi K2.6 and GPT-5.5 side-by-side.
• Ships with built-in slash commands and modes like hyperplan, analyze, and search, plus an ulw-loop for persistent task execution.
• Anonymous telemetry is on by default and capped at one event per day, with explicit environment-variable opt-outs.
• The codebase is currently mid-refactor to support additional harnesses beyond OpenCode and Codex, including Pi.

Caveats

• The project is undergoing a “Multi-Harness Agent OS Refactor,” so architectural stability is not guaranteed right now.
• Package and command names are in flux between oh-my-opencode, oh-my-openagent, omo, and lazycodex, which the README itself admits is confusing.
• The Ultimate edition setup is complex enough that the documentation recommends letting an LLM agent perform the installation for you.

Verdict

Worth a look if you are already living inside OpenCode or Codex CLI and want richer orchestration, multi-model teams, or just an escape hatch from single-vendor agent lock-in. Skip it if you want a standalone, zero-config CLI that works out of the box without plugin gymnastics.