Lets you bounce ideas off GPT, Gemini, Qwen, or other models without leaving Claude. Useful before committing big architectural changes, when debugging gets circular, or when you want to challenge your own assumptions. Supports single opinions, multi-model consensus, and devil's advocate mode. Fetches current model lists dynamically instead of hardcoding IDs, which is smarter than most wrappers. The AI-to-AI prompt framing is clever: other models advise Claude, not the user directly. Works for code review, architecture decisions, prompt improvement, or general strategy. Costs actual API money per query, so don't use it for trivial syntax checks.
npx skills add https://github.com/jezweb/claude-skills --skill brains-trust