The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to process data from external website audits which presents a surface for indirect prompt injection.\n- Ingestion points: Website audit data is ingested through the squirrel report command, providing content such as meta titles, descriptions, and page items in a token-optimized XML format as described in references/audit-website/OUTPUT-FORMAT.md.\n- Boundary markers: The output format utilizes XML tags (e.g., , ) and textual prefixes (e.g., Desc:, Fix:) to structure data, providing a layer of separation but not full instructional isolation.\n- Capability inventory: The skill utilizes a specialized tool (squirrel) for network-based auditing and report generation, which exposes the agent to potentially untrusted external content from target websites.\n- Sanitization: The documentation in references/audit-website/OUTPUT-FORMAT.md specifies that all text content is XML-escaped, which helps prevent structural breakage but does not filter out instructional content that could influence the AI's behavior.

seo-mastery