Skills are inscribed knowledge -- reusable instructions that teach AI agents how to perform specific workflows. rune keeps them current across projects via git-based registries.
cargo install rune Skills should be managed like dependencies.
AI coding agents accumulate instructions in markdown files. These files encode hard-won workflow knowledge -- how to run CI, how to write in voice, how to handle upstream contributions. Without a registry, every project maintains its own copy. rune treats skills like packages: versioned, synced, with drift detection when local edits diverge from the registry.
One config file defines your registries. One manifest per project declares which skills to sync. rune handles the rest.
Git repositories containing skill directories. Public registries
for shared workflows, private registries for proprietary methodology.
Configured globally in ~/.config/rune/config.toml.
Each project declares its skills in .claude/rune.toml.
Maps skill names to registries. rune sync pulls the right version of
each skill into .claude/skills/.
A Claude Code hook runs on session start. If local skill files have diverged from their registry source, the agent is told before it starts work. No silent staleness.
rune push sends local edits back to the registry.
Skills evolve where they're used, then propagate to every project
that syncs from that registry.
Import skills from upstream registries (community, vendor) into your own. Track upstream versions. Pull updates selectively. Fork when you need to diverge.
Works with Claude Code, Cursor, Windsurf, Copilot, Gemini CLI. Skills are markdown files in a directory convention. The agent reads them. rune manages them.
rune setupone-time config + hook installrune initcreate project manifestrune add <skill> --from <reg>add a skillrune syncpull latest from registriesrune checkshow driftrune push <skill>push local changes backrune browse <registry>discover available skillsrune import <skill>@<upstream>import from upstream