The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The skill allows the agent to download and install packages from GitHub using npx skills add <owner/repo>. While GitHub is a common source, the ability to pull from arbitrary, non-vetted repositories represents an external code dependency risk.
[COMMAND_EXECUTION] (MEDIUM): The skill explicitly instructs the agent to run shell commands like npx skills add <package> -g -y. The -y flag is particularly risky as it bypasses user confirmation prompts during the installation process, allowing for silent code execution.
[REMOTE_CODE_EXECUTION] (MEDIUM): By design, this skill installs and potentially executes code from remote sources. The automated installation of external packages without a manual verification step (due to the -y flag) is a high-risk capability, though it is the primary purpose of this specific skill.
[INDIRECT_PROMPT_INJECTION] (LOW):
Ingestion points: The agent ingests search results from the npx skills find command (metadata/descriptions of external skills).
Boundary markers: Absent; the instructions don't specify how to handle malicious content in skill descriptions.
Capability inventory: The skill can execute command-line installation via npx.
Sanitization: None; the agent is expected to present finding details directly to the user.
Risk: Malicious skill metadata in a public registry could potentially contain instructions to trick the agent into performing unauthorized installations.

find-skills