Nice that OpenClaw is already supported. Curious how the scoring holds up against auto-generated agent prompts — the repetition metric especially.