The Agent Skills Directory

[SAFE]: The skill contains inconsistent metadata. The LICENSE.txt file claims copyright by Anthropic, PBC, which contradicts the attributed authorship of the skill by atxinsky. This discrepancy in legal documentation could mislead users regarding the skill's provenance and security vetting.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests and processes untrusted PDF files without explicit boundary markers or instructions to the agent to ignore embedded commands. Ingestion points: PDF files are read in scripts/extract_form_structure.py, scripts/fill_fillable_fields.py, and other PDF-processing scripts. Boundary markers: Absent. Capability inventory: The agent has access to file read/write operations and command-line tools. Sanitization: Lacks mechanisms to sanitize text extracted from PDFs before agent processing.
[REMOTE_CODE_EXECUTION]: The script scripts/fill_fillable_fields.py uses runtime monkeypatching to modify the pypdf library's DictionaryObject.get_inherited method. Dynamic modification of class methods at runtime is a form of self-modifying code that can introduce stability or security risks.
[COMMAND_EXECUTION]: The skill instructions guide the agent to use command-line utilities for PDF operations, such as pdftotext, qpdf, pdftk, and magick. These tools are used for legitimate purposes like data extraction, document merging, and image refinement.

pdf