The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes npx --yes skills commands to manage other agent skills. These commands are executed based on explicit user requests and include safety constraints like avoiding --copy and verifying installations via --json output.
[REMOTE_CODE_EXECUTION]: While the skill downloads other skills from GitHub or URLs via npx skills add <source>, this is the primary, intended purpose of the tool. The instructions explicitly mandate listing skills (--list) and asking for user confirmation before installation if the source contains multiple items.
[DATA_EXPOSURE]: The skill accesses project-local and global configuration files (e.g., ~/.agents/.skill-lock.json, skills-lock.json) to inventory current skills. This data is used solely for identifying update candidates and is not exfiltrated.

install-skills