Skills command
Install the Faros skill for Claude Code and Codex.
kubectl faros skills fetches the skills/ directory of the faros repository
and installs it where AI coding assistants look for skills. No git checkout or GitHub token is needed: one request downloads the source tarball for the chosen ref and keeps only skills/<name>/ directories that contain a SKILL.md.
install
kubectl faros skills install [NAME...] [flags]With no name, every skill is installed. With one or more names, only those skills are installed.
| Flag | Description |
|---|---|
--target claude|codex | Install for one client only. The default installs for both. |
--scope user|project | user (default) writes to ~/.claude/skills and ~/.agents/skills. project writes to ./.claude/skills and ./.agents/skills. |
--dir <path> | Write to an arbitrary directory instead of the client defaults, for example ~/.cursor/skills. |
--ref <ref> | Tag, branch, or commit to fetch. Defaults to the repository’s default branch. |
--repo <owner/name> | Fetch from a fork or another repository with the same skills/ layout. |
--force | Replace directories that the command did not create. |
Each installed skill lands in a temporary sibling directory and is renamed into place, so a failed download never leaves a half-written skill. A .faros-skill.json marker inside the directory records the repository, ref, commit, CLI version, and file list.
Re-running install replaces directories that carry the marker. Directories without the marker are treated as your own and are refused unless you pass --force. Symlinks are always refused, so running the command inside a checkout of the faros repository does not clobber the committed .agents/skills/faros link.
list
kubectl faros skills list [-o json|yaml|name]Prints the skills the repository offers at the chosen --ref, with file counts and the resolved commit. --repo and --ref behave as for install.
Errors
An unknown ref returns the GitHub 404 as a clear error. A refused directory names the path and suggests --force. A symlink at the destination is reported and skipped. The full repository tarball is about 13 MB, so the command needs network access to codeload.github.com and a few seconds to complete.
See MCP for AI agents for what to do once the skill is installed.