| name | consult-llm |
| description | Use when the user wants a second opinion from Gemini or OpenAI/ChatGPT, asks to "consult" another model, or wants to compare responses across models. This is the interactive MCP chat path for a second opinion or model comparison; to run a structured review gate over code, diffs, plans, or large doc corpora through a headless CLI, use cli-review instead. |
consult-llm
You have access to two MCP tools for consulting other LLMs:
mcp__chat-gemini__chat-with-gemini — sends a message to Google Geminimcp__chat-openai__chat-with-openai — sends a message to OpenAI (GPT)
When This Skill Applies
Use this skill when the user:
- Asks to "consult", "ask", or "check with" another LLM (Gemini, GPT, OpenAI, ChatGPT)
- Wants a second opinion from a different model
- Says "what does Gemini think", "ask GPT", "get another model's take", etc.
- Wants to compare responses across models
How to Use
Selecting the right model
- If the user specifies "Gemini" or "Google" → use
mcp__chat-gemini__chat-with-gemini
- If the user specifies "GPT", "OpenAI", or "ChatGPT" → use
mcp__chat-openai__chat-with-openai
- If unspecified or user wants both → consult both in parallel
Crafting the prompt
- Send the user's question or context as-is, or distill it into a clear, self-contained query
- If you need to include code or context, include enough for the other model to give a useful answer
- Do not include internal Claude Code instructions or system context in the query
Presenting results
- Quote the other model's response clearly, labeled with which model it came from
- Add your own synthesis or comparison if the user asked for a second opinion
- If both models were consulted, highlight agreements and disagreements
Example Invocations
User: "What does Gemini think about this regex?"
→ call mcp__chat-gemini__chat-with-gemini with the regex and context
User: "Ask GPT and Gemini both — which sorting algorithm should I use here?"
→ call both tools in parallel, then compare their answers
User: "Get a second opinion on my architecture"
→ pick one or both models, send the architecture description, synthesize the feedback
