The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation establishes an indirect prompt injection surface within the PDF form-filling workflow.
Ingestion points: In forms.md, the skill converts PDF documents into PNG images for visual analysis and extracts metadata using scripts/extract_form_field_info.py.
Boundary markers: Absent. The instructions do not provide delimiters or safety markers to help the agent distinguish between its instructions and potentially malicious content embedded within the PDFs or images.
Capability inventory: The skill has the ability to execute multiple local scripts via subprocess (e.g., scripts/fill_pdf_form_with_annotations.py), write several JSON configuration files (fields.json), and generate modified PDF outputs.
Sanitization: Absent. There is no evidence of validation or sanitization for text and data extracted during the visual OCR or metadata analysis phases.
[COMMAND_EXECUTION]: The skill workflow is heavily dependent on the execution of local scripts and external CLI tools.
Evidence: forms.md requires the use of several utility scripts located in a scripts/ directory to manage form fields, PDF conversion, and annotation. Additionally, SKILL.md and reference.md demonstrate usage of standard CLI utilities including qpdf, pdftotext, and pdftk.
[EXTERNAL_DOWNLOADS]: The instructions direct the user to install external dependencies via standard package managers.
Evidence: SKILL.md contains instructions to pip install pytesseract pdf2image. These are standard libraries for OCR and image conversion tasks.

pdf