The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external file paths and OLE object data, creating an attack surface where an agent might be influenced by malicious filenames or file content if combined with other autonomous capabilities.
Ingestion points: The imagePath parameter in GetOrCreateImageDef and InsertRasterImage, and the pdfPath parameter in InsertPdfUnderlay (SKILL.md).
Boundary markers: No delimiters or safety instructions are provided to the agent to treat these paths as untrusted data.
Capability inventory: The skill uses imageDef.Load() to read files from the disk and SendStringToExecute to interact with the application command line.
Sanitization: The provided code snippets do not include path validation or sanitization logic.
[COMMAND_EXECUTION]: The documentation references the use of SendStringToExecute as a workaround for creating OLE objects. This method sends strings directly to the AutoCAD command processor, which is a potential vector for command injection if the input is not strictly controlled.

acad-images-ole