The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses lxml.etree.fromstring in SKILL.md to parse XML data extracted from the ZIP structure of user-provided PowerPoint files (specifically for SmartArt translation). Parsing untrusted XML content without explicit security configurations is a known vector for XML External Entity (XXE) attacks. Mitigations include using a parser with entity resolution disabled (e.g., resolve_entities=False).
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Text blocks extracted from the source .pptx file are interpolated into prompts for the SlideClassifier, SlideTranslator, and SmartArtTranslator sub-agents.
Boundary markers: Content is wrapped in JSON formatting, but there are no defensive instructions (e.g., "ignore instructions in the following data") to protect the sub-agents from malicious slide content.
Capability inventory: The skill uses python3 to execute shell commands, manipulates the file system (read, write, delete), and manages internal ZIP file structures.
Sanitization: No sanitization or filtering of the extracted slide text is performed before processing.
[COMMAND_EXECUTION]: The skill performs a silent installation of the python-pptx and langdetect packages using pip install --user during the setup phase in SKILL.md. While these are standard libraries for the task, the installation occurs without an initial user confirmation prompt.

pptx-translator