The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted document formats which may contain malicious instructions intended to influence the agent's behavior during the conversion process.
Ingestion points: The skill reads external files including PDF, PPTX, and DOCX, as well as raw text input provided by the user.
Boundary markers: The instructions do not define specific delimiters or command-override warnings to the agent when processing content from these untrusted sources.
Capability inventory: The skill utilizes file system access (Read, Write) and shell commands (pdftotext, pdfimages, pdftoppm, grep, magick) to process and transform document data.
Sanitization: Extracted text is placed directly into the output Markdown file without explicit sanitization or escaping of potential instructions embedded in the source documents.
[COMMAND_EXECUTION]: The skill instructs the agent to perform document parsing and image extraction via shell command templates.
The workflow involves executing pdftotext, pdfimages, pdftoppm, and grep to extract text, images, and URLs from PDF files.
It suggests using python-pptx for parsing PowerPoint structures.
While these tools are used for the skill's primary purpose, the execution of shell commands on data derived from untrusted documents represents an attack surface.
[EXTERNAL_DOWNLOADS]: The skill relies on external system utilities and libraries that must be installed on the host system.
Prerequisites include poppler, imagemagick, and the python-pptx Python library. These are standard, well-known tools for document processing tasks.

presentation-docs