The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions facilitate the ingestion and processing of external PDF documents, which serves as a potential attack surface for indirect prompt injection if the content of the PDF contains malicious instructions targeted at the agent.
Ingestion points: File reading via pypdf.PdfReader and pdfplumber.open() in SKILL.md.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore commands embedded within the PDF content.
Capability inventory: Text extraction, table extraction, and file manipulation capabilities are present.
Sanitization: No sanitization or validation of extracted text is described to prevent the agent from executing instructions found within the document data.

pdf