# Consult

> Ask the advisor a direct question on the user's behalf. Use when the user says "/consult ...", "ask the advisor ...", or "what does the advisor think about ...". Unlike proactive consults, there is no escalation bar - any question the user wants a second opinion on qualifies. Named /consult, not /advisor - Claude Code has a built-in /advisor command that shadows any plugin skill of that name.

- Skill: `alitahir6001/consult` (Agent Skill)
- Install (CLI): `npx skillmds@latest add alitahir6001/consult`
- Raw SKILL.md: https://api.skillmd.com/api/skills/alitahir6001/consult/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: alitahir6001 (https://skillmd.com/u/alitahir6001)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/alitahir6001/consult

---


# Ask the advisor directly

1. Take the user's question as the consult question, verbatim where
   possible. The advisor is stateless: if the conversation holds context it
   needs (the idea under discussion, constraints, repo-relative file
   paths), package that briefly alongside the question.
2. Route like any consult: the consult_advisor MCP tool when available,
   the advisor subagent otherwise.
3. Relay the advisor's reply to the user in full — do not compress it.
   Add your own take afterward only where it differs, clearly labeled.

## Report your own model

Pass `workhorse_model` on every call - the model YOU are running as. The
reply's header prints advisor and workhorse side by side so the user can see
they are not the same model. If you do not know your own model, omit the
field; the header then says "not reported". Never guess it.

Show the header verbatim. Do not summarise it away.