Adversarial Spec

This automates what good spec writers already do manually: get multiple perspectives to find weak spots. You feed it a draft PRD or technical spec, it pings GPT, Gemini, Grok, and others through their APIs, collects critiques, and runs rounds until the models stop disagreeing. Claude participates as a debater, not just a moderator. It supports direct API keys, AWS Bedrock for enterprise routing, and even CLI tools for subscription models. The interview mode is thoughtful if you're starting from scratch. Useful when you want external pressure testing without scheduling actual meetings, though you'll still need judgment about when consensus actually means the spec is good versus just bland.

Install to Claude Code

npx -y skills add zscole/adversarial-spec --skill adversarial-spec --agent claude-code