xiongjjlj/claude-debate-skill

A Claude Code skill that runs an adversarial "Builder ↔ Critic" debate loop to stress-test a research topic, pitch, product proposal, or idea. The Builder crafts the strongest possible version of the argument; the Critic generates concrete, content-specific attacks. The loop iterates until the Critic "passes" or a max-rounds cap is reached. Invoked as a slash command (/debate <topic> [--max-rounds N] [--depth quick|deep]). Outputs are saved to ./debates/<YYYY-MM-DD>-<slug>/ with per-round folders containing Builder drafts, the Critic's JSON+markdown, and a final verdict.md. Includes orchestrator and two agent templates (commands/debate.md, agents/debate-builder.md, agents/debate-critic.md). Install by copying files into ~/.claude/commands and ~/.claude/agents. Useful for researchers, product teams, founders, and engineers who want adversarial, iterative critique with configurable depth and round limits.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.