The Agent Skills Directory

[PROMPT_INJECTION]: The skill maintains an indirect prompt injection surface due to its core functionality of extracting and interpreting content from untrusted document files.
Ingestion points: Automated text extraction from .pptx files via the markitdown module and visual inspections during the QA loop as specified in SKILL.md.
Boundary markers: Instructions lack explicit delimiters or guardrails to prevent the agent from mistaking document content for administrative or steering commands.
Capability inventory: The agent is granted extensive system capabilities including file system access (read/write), shell command execution for various document processing tools, and dynamic Javascript generation for the pptxgenjs library.
Sanitization: Although the code uses defusedxml to prevent technical XML exploits (e.g., XXE), there is no evidence of sanitization for natural language instructions extracted from processed slides.
[EXTERNAL_DOWNLOADS]: The skill relies on and facilitates the installation of standard industry-standard dependencies.
Evidence: SKILL.md and pptxgenjs.md guide the setup of tools like markitdown, Pillow, and pptxgenjs using standard package managers (pip, npm).
Context: These resources are fetched from established official registries and are fundamental to the skill's operation.
[COMMAND_EXECUTION]: The skill frequently invokes subprocesses to perform necessary document manipulation tasks.
Evidence: Usage of LibreOffice (soffice), Poppler (pdftoppm), and custom Python scripts for packing/unpacking PPTX structures (scripts/office/pack.py).
Context: These shell commands are used in a manner consistent with the skill's primary purpose of file conversion and visual auditing.

powerpoint