The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the npx skills CLI for searching, updating, and installing packages.
[REMOTE_CODE_EXECUTION]: The skill facilitates the installation and execution of external code through the npx skills add <package> command. It specifically advises using the -y flag to bypass user confirmation prompts and the -g flag for global installation, which increases the risk of automated installation of malicious extensions if a search returns untrustworthy results.
[EXTERNAL_DOWNLOADS]: The skill interacts with external sources, including a dedicated registry at skills.sh and third-party GitHub repositories, to fetch and install new capabilities. It references well-known repositories such as vercel-labs/agent-skills for extension sources.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect injection:
Ingestion points: User-provided search queries in SKILL.md and the output (descriptions/metadata) retrieved from external skill packages via npx skills find.
Boundary markers: No explicit delimiters or warnings to ignore instructions embedded in skill descriptions are present.
Capability inventory: The agent has the ability to execute shell commands and install new executable code (npx skills add).
Sanitization: There is no mention of sanitizing or validating the content returned from the external search registry before presenting it to the user or acting upon it.

find-skills