The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes multiple local helper scripts including unpack.py, thumbnail.py, rearrange.py, inventory.py, and replace.py to manipulate Office files. It also invokes system utilities like soffice and pdftoppm for document conversion tasks.
[COMMAND_EXECUTION]: A core workflow involves the agent dynamically generating a JavaScript file and executing it via Node.js to create presentations. This pattern of generating and running code at runtime is a risk factor if the generation logic incorporates data from untrusted sources.
[COMMAND_EXECUTION]: Documentation for the skill includes the use of sudo apt-get for installing system-level dependencies like LibreOffice and Poppler utilities.
[PROMPT_INJECTION]: The skill processes content extracted from external .pptx files, which serves as an indirect prompt injection surface where instructions hidden in slides could influence agent behavior.
Ingestion points: Text and metadata extraction from PowerPoint files via markitdown and XML unpacking scripts (e.g., SKILL.md).
Boundary markers: The instructions do not include the use of clear delimiters or instructions to the agent to disregard embedded commands in the extracted text.
Capability inventory: The skill has the ability to execute Python and Node.js scripts, write to the file system, and run shell commands (soffice, pdftoppm).
Sanitization: While defusedxml is used to prevent XML-based attacks, there is no evidence of sanitization or filtering for natural language instructions extracted from the presentation content.

pptx