Adds restorative-repair, an external public Claude-compatible skill for response-layer repair when AI agents make output errors.\n\nRepo: https://github.com/MinistaJazz/restorative-repair\n\nThe skill includes trigger taxonomy, severity tiers, crisis suppression rules, bounded loop-exit behavior, and auditable benchmark prompts.