OmO
Oh My OpenAgentv3.17.13

Oh My OpenAgent

A multi-model agent harness for OpenCode. Type ulw and let planning, delegation, execution, and verification run as one loop.

Oh My OpenAgent

Oh My OpenAgent (OmO) is a multi-model agent harness for OpenCode. It is not just a prompt pack: it connects intent detection, planning, delegation, execution, verification, and session continuity around OpenCode.

OmO is built around one premise: the human should describe the outcome, not babysit the harness. For quick work you can prompt normally. For complex work you can type ulw or ultrawork. For high-accuracy work you can plan with Prometheus, then execute with /start-work.

The published package and binary are still oh-my-opencode; the compatibility layer prefers the OpenCode plugin entry oh-my-openagent. Use the compatibility command when following upstream installation docs:

bunx oh-my-opencode install

Compatibility note: oh-my-openagent is also exposed as a binary, and both renamed and legacy plugin config basenames are recognized during the transition.

Core Loop

What OmO Adds

LayerWhat it gives the agent
Intent GateReads whether the user wants an answer, investigation, implementation, planning, or delegation.
Discipline agentsSisyphus, Hephaestus, Prometheus, Atlas, Oracle, Librarian, Explore, Metis, Momus, and Sisyphus-Junior each do different work.
Category routingDelegation uses work categories such as visual-engineering, ultrabrain, deep, quick, and writing, instead of forcing the user to pick model names.
SkillsDomain workflows like playwright, git-master, frontend-ui-ux, review-work, and project skills inject instructions and MCP servers only when needed.
Agent-grade toolsLSP, AST-grep, tmux, browser QA, MCP search, session tools, and hash-anchored edits.
Continuation hooksTodo continuation, Ralph/ulw loops, session recovery, runtime fallback, and reminders keep work moving after interruptions.

Documentation Map

On this page