The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the sips command-line utility for image conversion and instructs the agent to generate and execute a Python script to create the PDF output.
[COMMAND_EXECUTION]: The instructions explicitly direct the agent to perform all steps, including script execution and file operations, automatically without asking for user confirmation, which reduces human oversight during the processing of sensitive data.
[EXTERNAL_DOWNLOADS]: The documentation and instructions suggest installing multiple external Python packages (easyocr, pytesseract, reportlab, pyobjc-framework-Vision, pyobjc-framework-Quartz) and system packages via Homebrew (tesseract). These are fetched from official package registries.
[DATA_EXPOSURE]: The skill is designed to process highly sensitive personal identification and financial records (passports, ID cards, bank statements). While no exfiltration behavior is identified, the automated processing and script execution on such high-value data warrants caution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted image data for OCR processing without explicitly defined boundary markers or sanitization steps before the extracted text is used for translation and PDF generation, creating a surface for potential indirect instruction injection from the document content.
Ingestion points: User-provided image files processed via OCR (SKILL.md).
Boundary markers: None identified.
Capability inventory: Subprocess calls (sips), Python script generation and execution, file system read/write operations (SKILL.md).
Sanitization: No sanitization or validation of OCR-extracted text is specified before it is processed by the LLM.

visa-doc-translate