Install
openclaw skills install @its-how/aha-opencode-omoFull-power + full-delegation mode for OpenCode+OMO. Only when user explicitly invokes 'aha-opencode-omo'. NOT for trivial edits, Q&A, or read-only.
openclaw skills install @its-how/aha-opencode-omotask() to OMO (oh-my-openagent, https://github.com/code-yeongyu/oh-my-openagent) subagentstask() categories availableopencode agent list and confirm OMO agents present (oracle, momus, sisyphus, explore min.)Do NOT default to sisyphus for every task. Match per nature:
| Task type | Agent(s) | task() category |
|---|---|---|
| Planning / architecture | prometheus, metis | unspecified-high |
| Multi-step execution | sisyphus, atlas | quick/deep/ultrabrain |
| Review / critique | momus | unspecified-high |
| Deep diagnosis | oracle, hephaestus | deep/ultrabrain |
| Discovery / research | explore, librarian | unspecified-low |
| Writing / docs | any | writing |
| Visual / UI | multimodal-looker | visual-engineering |
Full delegation does NOT extend to hard-deny actions. These remain forbidden without how confirmation:
~/.local/share/opencode/auth.jsonopencode plugin <module> / opencode mcp add/auth/logout--dangerously-skip-permissionsexplore/librarian directly)If a delegated subagent returns garbage or errors:
ultrabrain or oracle)Do NOT silently accept garbage.
aha-opencode-omo is explicit opt-in per task chain. Native-first is the default. When not activated, execute natively via opencode run, opencode agent, or project-local rules.
Multi-agent delegation consumes significantly more tokens than native execution.
~/.config/opencode/skills/aha-opencode-omo/ and restart OpenCode if hot-reload does not catch it