Ask HN: How do you manage skills files?
Points and comments are a snapshot, not live.
An Ask HN user seeks better ways to manage, organize, and maintain skills files for AI agents.
The poster asks how people find skills, keep them organized, ensure they work, and improve them over time. They note that general-purpose skills (like design critique) may eventually be subsumed by model capabilities, but custom, insider-knowledge skills remain valuable. They are looking for a better management system until that transition.
What commenters are saying
Commenters split on the value of skills. Some argue concise, deterministic skills (e.g., a script to connect to a server) always beat models' inference-heavy approaches, saving tokens and avoiding errors. Others keep skills minimal, storing only frequently used prompts in shared repos, and rely on AGENTS.md for project specifics. Several share practical tips: use system-level tools like chezmoi for dotfiles, package managers like 'ski', and bootstrap scripts for team-wide deployment. A recurring theme is that skills should be small, scoped, and updated via agent feedback loops.