The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes several well-known and reputable external libraries for its core functionality. These include Microsoft's markitdown for document parsing, Pillow for image processing, and the pptxgenjs library for creating presentations. These are standard dependencies for the tasks described.
[COMMAND_EXECUTION]: The skill provides instructions for executing various shell commands and local scripts to process files. This includes using python to run utility scripts (e.g., thumbnail.py, unpack.py, soffice.py) and standard tools like pdftoppm for converting PDFs to images. These operations are limited to handling PowerPoint documents and their converted outputs.
[PROMPT_INJECTION]: The skill involves parsing untrusted content from .pptx files using the markitdown tool, which presents a surface for indirect prompt injection. Malicious instructions embedded within a presentation could potentially be processed by the agent after text extraction.
Ingestion points: Reading and analyzing content from user-provided .pptx files in the "Reading Content" and "QA" workflows.
Boundary markers: The instructions do not currently define explicit boundary markers or "ignore" directives to separate extracted document text from agent instructions.
Capability inventory: The agent has access to shell commands (python, pdftoppm, soffice) and file system operations to support presentation workflows.
Sanitization: No specific sanitization or validation of the extracted text is described before it is processed by the agent.

pptx