The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of PDF files provided by the user to perform whitespace trimming.
Ingestion points: The scripts/proportional_trim.py script reads the input PDF file specified by the user in the input_pdf argument.
Boundary markers: The script operates on visual renders (pixmaps) of the PDF pages rather than parsing or interpreting text for instructions, which naturally isolates the logic from content-based injections.
Capability inventory: The script's capabilities are limited to reading the input PDF, writing the transformed PDF to a specified path, and optionally writing a JSON report. It does not perform network operations, execute arbitrary shell commands, or use dynamic execution functions like eval().
Sanitization: The script treats the PDF content as image data for bounding box detection, preventing the interpretation of document text as code or instructions.
[EXTERNAL_DOWNLOADS]: The skill's Python script requires two external libraries to be installed in the environment.
Evidence: import fitz (PyMuPDF) and import numpy as np are used for PDF manipulation and image processing.
Context: These are well-known, industry-standard libraries used for their intended purposes in this script. They are typically installed via official package registries like PyPI.

proportional-pdf-trimmer