The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and analyzing external source documents (plans, policies, contracts, architecture diagrams) which are potentially attacker-controlled or untrusted. This creates a surface for indirect prompt injection where malicious instructions embedded in these documents could attempt to influence the audit results or remediation roadmap.
Ingestion points: External files provided as source documents, including PDF, Markdown, and document paths specified in the Audit Metadata.
Boundary markers: The skill enforces a strict output structure with mandatory sections, but does not explicitly document the use of XML delimiters or 'ignore' instructions for the parsed document content.
Capability inventory: The skill can read local files and produce a prioritized remediation roadmap. It does not have explicit network exfiltration or shell execution capabilities in its defined logic.
Sanitization: No specific content sanitization or instruction-stripping logic for source documents is described.
[DYNAMIC_EXECUTION]: The skill utilizes a 'Framework Adapter Pattern' that dynamically loads structured control catalogs from a local directory (compliance-frameworks/) based on the --framework argument. While designed for modularity and ease of updates, dynamic loading from computed paths requires careful path validation to prevent unauthorized file access or configuration manipulation.

compliance-audit