The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted PDF files, which represents an attack surface where malicious instructions could be embedded in document text or metadata.
Ingestion points: PDF content is ingested via pypdf, pdfplumber, pdftotext, and pytesseract as documented in SKILL.md and the scripts.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when the agent processes extracted text.
Capability inventory: The skill possesses capabilities for file system writes and shell command execution, which could be targeted by successful injection.
Sanitization: No specific sanitization logic for extracted PDF text is provided in the skill's code.
[DYNAMIC_EXECUTION]: The skill uses runtime monkeypatching to modify the behavior of an external library.
Evidence: scripts/fill_fillable_fields.py contains a monkeypatch_pydpf_method function that overrides pypdf.generic.DictionaryObject.get_inherited to correctly handle specific PDF field attributes.
[EXTERNAL_DOWNLOADS]: The skill's documentation references multiple well-known and trusted external libraries for its core functionality.
Evidence: Mentions pypdf, pdfplumber, reportlab, pytesseract, pdf2image, pandas, pypdfium2, pdf-lib, and pdfjs-dist in SKILL.md and reference.md.
[COMMAND_EXECUTION]: The skill relies on executing various command-line utilities and local scripts to perform PDF tasks.
Evidence: The agent is instructed to use pdftotext, qpdf, pdftk, and magick (ImageMagick), as well as the provided Python scripts in the scripts/ directory for form extraction and filling.

pdf