The Agent Skills Directory

[PROMPT_INJECTION]: The skill incorporates robust defensive instructions specifically designed to resist indirect prompt injection. It explicitly commands the agent to ignore any directives or override attempts embedded in the documents it processes, treating all external file content as untrusted data.
[DATA_EXFILTRATION]: Analysis of the YAML frontmatter confirms that the skill is restricted to a limited set of file-system tools (Read, Grep, Glob). The absence of network-capable tools or code for external data transmission ensures that the sensitive organizational information accessed during the assessment cannot be exfiltrated.
[SAFE]: The skill processes organizational documents including security policies and network diagrams as part of its primary function. Because the ingestion points are clearly defined and the agent is provided with boundary markers and sanitization instructions, the risk of indirect injection is appropriately mitigated and inherent to the skill's purpose as a vCISO assistant.

nist-csf-assessment