The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions and reference documents recommend installing various third-party PDF processing libraries and command-line utilities, including pypdf, pdfplumber, reportlab, pytesseract, pdf2image, poppler-utils, and qpdf.
[COMMAND_EXECUTION]: The toolkit includes multiple Python scripts in the scripts/ directory that are intended to be executed as part of the PDF processing workflow for tasks such as converting PDFs to images, extracting form field information, and filling PDF forms.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted PDF documents. A malicious PDF could contain visual elements or text designed to mislead the agent during the automated or visual analysis steps.
Ingestion points: PDF files are converted to images by scripts/convert_pdf_to_images.py for visual inspection by the agent.
Boundary markers: The skill does not employ explicit boundary markers or instructions to ignore embedded directives when analyzing PDF content.
Capability inventory: The agent has the capability to write files, modify PDF structures, and execute local scripts based on the analysis of the untrusted data.
Sanitization: No sanitization or validation of the input PDF content is performed.
[REMOTE_CODE_EXECUTION]: The script scripts/fill_fillable_fields.py performs dynamic modification of code at runtime by monkeypatching the pypdf library's DictionaryObject.get_inherited method. While this is used to address a specific bug in the library's form handling, it represents a dynamic execution pattern.

pdf