The Agent Skills Directory

[SAFE]: The skill provides documentation and scripts centered around the 'skills-il' vendor ecosystem and well-known Hebrew NLP research institutions (Dicta, Bar-Ilan University, ivrit.ai). All referenced HuggingFace model IDs and external URLs point to legitimate, high-reputation resources within the Israeli NLP community.
[COMMAND_EXECUTION]: Provides standard Bash(python:*) and Bash(pip:*) tools for model inference and package management. These are used logically for loading NLP models via the transformers library and running the provided preprocessing script.
[EXTERNAL_DOWNLOADS]: Downloads NLP models from HuggingFace (dicta-il/*, ivrit-ai/*, onlplab/alephbert-base). These are well-known, trusted research organizations in the AI domain.
[DYNAMIC_EXECUTION]: Mentions the requirement for trust_remote_code=True when loading specific character-level models from Dicta (dictabert-large-char-menaked) because they contain custom prediction logic. The skill correctly cautions the user to review the model card before enabling this standard HuggingFace feature.
[SAFE]: The preprocessing script (scripts/preprocess_hebrew.py) is a clean implementation using only Python's standard library to handle UTF-8 normalization and regex-based cleaning of Hebrew diacritics.

hebrew-nlp-toolkit