| name | omg-ask |
| description | Human-only broker for external advisor CLIs (codex/claude/gemini). Use when user wants a second opinion via omg ask. Never shell external CLIs as workers. |
omg-ask — External advisors (user-invoked only)
omg ask is a trusted human broker for Codex / Claude (fable) / optional Gemini. It is not a product executor and not a default worker path.
HARD RULES (non-negotiable)
- Fan-out ONLY via Grok
spawn_subagent (depth=1; children must NOT spawn).
- NEVER invoke
claude / codex / omc team / agy / cursor-agent via run_terminal_command as workers.
- For external second opinions, tell the human to run:
omg ask codex "your question"
omg ask claude "your question"
omg ask gemini "your question"
- Output is advisory under
.omg/artifacts/ask-*.md.
- Do not mark
verified / passes. Do not apply advisor patches automatically.
- Product changes require
omg ulw / omg ralph / omg pipeline implement stages.
Use when
- User asks for Codex review, Fable/Claude second opinion, multi-vendor dual-review.
- High-risk change needs an external advisor in addition to Grok-native critic/verifier.
Do not use when
- Default implement / plan / verify — use Grok-native modes.
- You are an agent tempted to shell
claude -p — stop; use Grok tools only.
Playbook
- Draft the question (scope, files, risks).
- Ask the human to run
omg ask <provider> "…".
- Read the artifact path they paste; treat as advisory.
- Continue implementation with Grok-native tools only.
