A library of 'loops' for coding agents like Cursor and Claude Code that include triggers, feedback gates, and exit conditions. These workflows allow agents to self-pace through tasks like fixing CI failures or verifying deploys until specific goals are met.