The Agent Skills Directory

[COMMAND_EXECUTION]: The skill scripts execute external command-line tools to process documents. scripts/make_contact_sheet.py uses pdftoppm for rendering, while scripts/synthesize_signature.py utilizes fc-list for font discovery and pdftotext for bounding box extraction. These calls are implemented using the subprocess list format, which is a secure practice to prevent shell injection. The agent is also instructed to run soffice and pdftoppm via the command line.
[EXTERNAL_DOWNLOADS]: The skill relies on external tools and libraries. It uses uvx to execute noteshrink, a specialized utility for document image enhancement. It also recommends installing system dependencies like poppler and fontconfig via brew. Python dependencies such as img2pdf, pillow, opencv-python-headless, and numpy are defined within the scripts and retrieved from standard registries.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection because it is designed to ingest and process untrusted user data in the form of photos and PDF documents.
Ingestion points: Document images are processed in scripts/photo_to_scan.py, and PDF files are read in scripts/synthesize_signature.py and scripts/make_contact_sheet.py.
Boundary markers: The instructions do not define specific boundary markers or instruct the agent to ignore potential commands embedded within the text of the processed documents.
Capability inventory: The skill provides capabilities for writing files to the local system and executing document manipulation tools.
Sanitization: No evidence of sanitization or filtering was found for the content extracted from user documents before it is used to inform downstream tool parameters.

photo-to-scanned-pdf