The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from external DOCX and Markdown files, which creates an attack surface where an adversary could embed hidden instructions for the agent within the document data.
Ingestion points: Data is ingested through the read_docx function in scripts/docx_utils.py and the MarkdownParser class in scripts/md_to_docx.py when reading user-supplied files.
Boundary markers: The skill does not implement specific boundary markers or instructions to the agent to treat document content as data rather than instructions.
Capability inventory: The skill allows for file system writes via the doc.save() method used in both scripts/docx_utils.py and scripts/md_to_docx.py.
Sanitization: The implementation uses regex-based parsing to identify document structure but does not perform sanitization of the actual text content to filter out potentially malicious agent instructions.

docx