When Claude Code gets stuck on a bug after one try or you're making an architectural decision, this lets it phone a friend: Gemini 2.5 Pro (with web search and thinking mode), OpenAI's GPT-4 via Codex (repo-aware), or a fresh Claude instance. It uses existing CLI tools instead of building MCP servers, which keeps things simple. The real discipline here is the mandatory 5-part synthesis format that forces Claude to compare its analysis with the external AI's take, highlight key differences, and justify a combined recommendation. You approve before it implements anything. Costs run $0.05 to $0.50 per consultation for paid AIs, free for the fresh Claude option. Templates let you customize behavior and .geminiignore keeps secrets out.
npx skills add https://github.com/secondsky/claude-skills --skill multi-ai-consultant