The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external content from software repositories, which creates a potential attack surface.
Ingestion points: The AnalyzeRepositoryRisk logic in SKILL.md processes repository metadata such as descriptions and topics provided by external sources.
Boundary markers: The skill lacks explicit boundary markers or instructions that would prevent an agent from inadvertently executing commands embedded within the external repository data it is analyzing.
Capability inventory: The skill contains instructions for powerful PowerShell commands, including Disable-NetAdapter (which can modify network state), Get-Process (system monitoring), and recursive file listing in C:\Windows\System32.
Sanitization: No input validation or sanitization is applied to the repository metadata before it is processed by the analysis logic.
[COMMAND_EXECUTION]: The skill provides numerous PowerShell code snippets for system-level operations, including process management, network configuration, and scheduled task inspection. If these are utilized by an agent as part of its execution path, they represent a significant capability for system interaction.

malware-distribution-awareness