Install
openclaw skills install @its-how/aha-codex-omxMode switch for Codex+OMX: full delegation x full capability surface x per-task OMX sub-mode matching. NOT for trivial edits, quick Q&A, read-only exploration, or when user says 'do it natively'.
openclaw skills install @its-how/aha-codex-omxActivates the full Codex+OMX (oh-my-codex, https://github.com/Yeachan-Heo/oh-my-codex) delegation surface. This is a stance declaration, not a procedure library.
omx --versionMap task type to OMX sub-mode CATEGORY. Do not hardcode individual skill names — OMX has 48 catalog entries with frequent renames and deprecations (16 deprecated as of v0.18.1).
| Task type | OMX category |
|---|---|
| Planning, design, architecture | planning |
| Execution, implementation | execution |
| Team orchestration, parallel lanes | team |
| Review, QA gate | review |
| Research, investigation | research |
Run omx list for the current active skills — this is the dynamic source of truth.
Autopilot still requires the task to genuinely need a full gate loop. Do not default to autopilot for straightforward or bounded work.
Full delegation does NOT extend to the following. Subagents must pause and ask how:
If a delegated subagent returns garbage or errors:
Do NOT silently accept garbage output.
This mode is explicit opt-in per task chain. Native-first is the default resting state. When the task chain completes or the user does not re-invoke the mode, return to native-first execution.
Multi-agent delegation consumes more tokens than native single-pass. If the user signals cost concern, fall back to native-first.
~/.codex/skills/aha-codex-omx/ directory, then restart Codex — Codex caches skill metadata until restart.