This is a Hacker News discussion where developers describe their actual practices for organizing "skills" — prompts, context files, or instruction sets fed to AI coding agents. The concrete patterns that emerge: storing skills in dotfiles (with symlinks), private GitHub repos, or team-shared repositories; using package managers or custom CLI tools to version and distribute them; keeping community skills curated to a short list of trusted authors; and favoring minimal, task-specific prompts over bloated general-purpose ones. One useful tension: whether general-use skills will become obsolete as models improve (the original questioner clarifies they meant only the truly generic ones, not domain-specific institutional knowledge). The practical takeaway is that skills management mirrors documentation management — treat them as living context for your specific system rather than cargo-culted downloads.