The Agent Skills Directory

[SAFE]: The skill is designed for legitimate OCR and data extraction tasks on official Israeli documents. All provided scripts, including scripts/preprocess_image.py and scripts/extract_form_fields.py, utilize standard open-source libraries such as OpenCV, Tesseract, and Pillow for image manipulation and text extraction. No unauthorized network activity, credential harvesting, or suspicious file operations were identified.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it ingests untrusted data from scanned images and converts it into text that is processed by the agent. However, the risk is considered low because the skill's environment and tools are restricted to image processing, and there are no capabilities provided that would allow malicious content in a document to execute system-level commands or exfiltrate data.
Ingestion points: Image files are processed via scripts/extract_form_fields.py and scripts/preprocess_image.py.
Boundary markers: No explicit markers are used to encapsulate OCR-generated text in the agent's context.
Capability inventory: The skill only includes libraries for image processing and OCR. No network access or shell execution tools are enabled within the provided scripts.
Sanitization: The normalize_bidi_text function in scripts/extract_form_fields.py provides basic sanitization by stripping Unicode format characters (Cf) and normalizing whitespace.

hebrew-ocr-forms