Argue

This turns Claude into a debate moderator. Instead of one AI giving you an answer, argue spins up multiple agents who independently analyze your question, challenge each other's reasoning across rounds, and vote toward consensus. It's overkill for "what's the syntax for X" but genuinely useful for architecture decisions, risk analysis, or any question where you'd normally want a second opinion to catch blind spots. Debates take 3 to 7 minutes and cost more tokens (grows with agent count squared), so you're trading speed and API spend for higher confidence answers. The sweet spot is 2 to 3 agents on decisions with real stakes, like monorepo versus polyrepo or pre-commit reviews on big changes. Outputs live in structured JSON with a web viewer for the full transcript.

Install to Claude Code

npx -y skills add onevcat/argue --skill argue --agent claude-code