the basic verbs: install, update, remove, list, agents. flags map one-to-one with the js api.

# install latest semver tag from a repo (or default branch if no tags)
$ rosie install vercel-labs/agent-skills

# specific skill from a repo
$ rosie install anthropics/skills pdf

# pin to a branch or tag — recorded as `pin` in the lockfile
$ rosie install owner/repo@v1.0.0
$ rosie install owner/repo@develop

# scope to specific agents (repeatable)
$ rosie install owner/repo -a claude -a cursor

# install a local directory as a skill (symlinked, travels with your repo)
$ rosie install ./my-custom-skill

# reinstall everything in .agents/rosie.lock — useful on a fresh clone
$ rosie install

# update lockfile entries
$ rosie update                      # all entries
$ rosie update slack-gif-creator    # one entry

# list installed skills/refs in this project
$ rosie list

# list what's available in a remote repo (without installing)
$ rosie list owner/repo

# remove an installed skill or reference
$ rosie remove skill-name
$ rosie remove skill-name -a claude   # from a specific agent

# show detected + supported agents
$ rosie agents

# skip the confirmation prompt
$ rosie install owner/repo -y

flags

• ▸-a, --agent <name>install to a specific agent (repeatable)
• ▸-g, --globalinstall globally to ~/.<agent>/skills/ · GitHub sources are copied · local paths are symlinked in place
• ▸-l, --localinstall locally with symlinks (default)
• ▸-r, --refinstall as a reference (README, or a SKILL.md via --skill)
• ▸-s, --skill <name>with --ref: install a specific SKILL.md
• ▸-n, --name <name>with --ref: override the default install name
• ▸-N, --npmwith --ref: source from node_modules/<pkg>/
• ▸-I, --include <path>with --npm: file or directory to include (repeatable; replaces default scope)
• ▸--cwd <path>run as if started from <path> (mirrors js api's cwd option)
• ▸--no-lockfiledon't read or write .agents/rosie.lock
• ▸-y, --yesskip the confirmation prompt
• ▸-v, --verboseverbose output

local vs global

two install modes; default is local. global is opt-in via --global.

• ▸ local (default) canonical copy at .agents/skills/<name>/; symlinks from each agent's local dir (.claude/skills/, .cursor/skills/, …) into it. project-scoped. check the lockfile into git for reproducible installs.
• ▸ global (--global) installs into ~/.<agent>/skills/<name>/ for every detected agent. shared across projects. GitHub sources are copied (no lockfile). local paths (rosie install ~/skills/my-skill -g) are symlinked directly at the source and tracked in ~/.agents/rosie.lock — re-run with no args to re-link after a machine wipe.