The Agent Skills Directory

[COMMAND_EXECUTION]: Executes shell commands via the npx skills CLI to perform searches, list installed skills, and manage the local agent skill directory.
[EXTERNAL_DOWNLOADS]: Downloads configuration files, skill packages, and metadata from well-known sources including the official skills.sh directory and public Git repositories on GitHub and GitLab.
[REMOTE_CODE_EXECUTION]: Facilitates the installation of third-party agent skills from external repositories. These skills contain instructions and scripts that are executed by the agent. The process leverages the official skills package from Vercel Labs, which is a trusted organization.
[INDIRECT_PROMPT_INJECTION]: The skill maintains an attack surface by ingesting untrusted data from third-party Git repositories during the installation process.
Ingestion points: Third-party GitHub and GitLab repositories accessed via the npx skills add command.
Boundary markers: The skill implements strong boundary instructions in SKILL.md (e.g., 'Ask First, Never Mutate to "Show" Something') and includes a dedicated 'Security Review' workflow in references/workflows.md that requires the agent to halt for user confirmation and conduct a manual audit of the content.
Capability inventory: The skill can execute shell commands (npx skills), write files to the filesystem (.claude/skills/), and perform network operations to retrieve repository data.
Sanitization: The skill relies on human-in-the-loop validation and agent-led manual content review rather than automated sanitization of the fetched repository content.

find-skills