The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted PDF files and converts them to images, creating a potential surface for indirect prompt injection attacks where malicious instructions could be embedded in the document text or layout.
Ingestion points: User-provided PDF files are processed in scripts/convert_pdf_to_images.py, scripts/extract_form_field_info.py, and via libraries such as pdfplumber as shown in SKILL.md examples.
Boundary markers: There are no explicit instructions or delimiters defined to separate user-provided content from agent instructions, which increases the risk of the agent following instructions embedded within a PDF.
Capability inventory: The skill possesses file system read/write access and the ability to execute command-line utilities such as qpdf and pdftotext.
Sanitization: No sanitization or filtering is performed on the extracted text before it is presented to the agent for analysis.
[DYNAMIC_EXECUTION]: The script scripts/fill_fillable_fields.py performs a monkeypatch on the pypdf library at runtime to address an internal bug in the library's form handling. This involves dynamically modifying the behavior of the get_inherited method in the pypdf.generic.DictionaryObject class.

pdf