Adversarial Polish

This runs your draft through multiple adversarial rounds where a critic finds problems, a rewriter fixes them, a synthesizer merges the best of both, and three blind judges vote on which version wins. It keeps going until the original wins twice in a row or hits four rounds. Every intermediate artifact gets saved to disk so you can see the evolution. The whole thing depends on spawning isolated sub-agents so each role only sees what they should, no shared context pollution. If you're in Claude Code this works as designed. On claude.ai it degrades to theater since everything shares one context window. Good for high-stakes writing where you'd normally do multiple editing passes yourself, overkill for casual stuff.

Install to Claude Code

npx -y skills add jimliu/illustrated-agent-skills --skill adversarial-polish --agent claude-code