Interactive loop that builds and tunes an AI prompt for evaluating whether a company fits a client's ICP. Run after any list-building skill (disco-like, blitz-list-builder, google-maps-list-builder, prospeo-full-export) to qualify companies before scaling. Iterates batches of 10 companies with user feedback, stops when 2 consecutive rounds have zero corrections, saves the final prompt for reuse. Always uses Claude Code Task sub-agents — never an external API key.
