The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes instructions and scripts for extracting text and structure from PDF files, such as scripts/extract_form_structure.py and scripts/extract_form_field_info.py. This data is ingested into the agent's context, creating a potential surface for indirect prompt injection if a processed PDF contains malicious instructions.
Ingestion points: Text and form structure extraction occurs in SKILL.md code examples and via scripts using pdfplumber and pypdf.
Boundary markers: The provided scripts do not implement specific delimiters or 'ignore' warnings for the extracted content before presenting it to the agent.
Capability inventory: The skill possesses file-writing capabilities (e.g., writer.write() in scripts/fill_fillable_fields.py) and allows the execution of multiple internal Python scripts.
Sanitization: The scripts perform coordinate extraction and text retrieval without explicit sanitization or filtering of the content within the PDF files.
[DYNAMIC_EXECUTION]: The script scripts/fill_fillable_fields.py implements a monkeypatching technique within the monkeypatch_pydpf_method function. This function dynamically replaces the DictionaryObject.get_inherited method of the pypdf library at runtime to adjust the handling of inherited field attributes.
[EXTERNAL_DOWNLOADS]: The documentation and skill instructions reference standard installations for well-known PDF processing libraries and utilities, including pypdf, pdfplumber, reportlab, pytesseract, pdf2image, poppler-utils, and qpdf. It also references official legal terms on Anthropic's website.

pdf