The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes content from external PDF files, which creates a potential surface for indirect prompt injection attacks.
Ingestion points: Text and data extraction are performed using pypdf, pdfplumber, and pdftotext as described in SKILL.md and reference.md.
Boundary markers: The skill does not implement delimiters or provide specific instructions to the agent to treat extracted content as untrusted data.
Capability inventory: The agent can execute local Python scripts (e.g., scripts/fill_fillable_fields.py) and write files to the local system, which could be leveraged if malicious instructions are processed.
Sanitization: No sanitization or filtering is performed on the text extracted from PDF documents before it is used by the agent.
[COMMAND_EXECUTION]: The skill utilizes dynamic code execution and system commands for document processing.
Evidence: The script scripts/fill_fillable_fields.py implements a monkeypatch on the pypdf library's DictionaryObject.get_inherited method to correct a bug in selection list processing.
Evidence: Documentation in SKILL.md and reference.md instructs the agent to execute various command-line utilities including pdftotext, qpdf, pdftk, and pdfimages for PDF manipulation.

pdf