The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download the SkillSpector tool from NVIDIA's official GitHub repository. This targets a well-known technology organization.
[COMMAND_EXECUTION]: The skill describes using Docker commands to run security scans on other skills. These instructions include safety practices such as using read-only volumes (:ro) and throwaway containers (--rm).
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external skill data, which presents an indirect prompt injection surface.
Ingestion points: Files within the mounted target directory or content from a specified Git URL.
Boundary markers: The scanner documentation mandates the use of the --no-llm flag to prevent the tool from processing instructions as active commands during the scan.
Capability inventory: Uses git to clone the tool and docker to execute the scanning environment.
Sanitization: Relies on static parsing and read-only execution environments to ensure that untrusted data cannot influence the host or the scanner's behavior.

skill-scan