The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its input validation scripts.\n
Ingestion points: scripts/validate_clip_usage.py processes raw user queries passed as command-line arguments.\n
Boundary markers: The script lacks delimiters or instructions to ignore embedded commands within the processed queries.\n
Capability inventory: The skill's configuration allows for Bash, Read, Write, and Edit operations.\n
Sanitization: No sanitization or escaping is performed on the input string before logic evaluation.\n- [COMMAND_EXECUTION]: The skill uses Bash to manage environment setup and run validation scripts.\n
Executes pip install for dependency management during installation.\n
Runs local validation and diagnostic scripts as part of its documented workflow.\n- [EXTERNAL_DOWNLOADS]: The skill performs downloads of machine learning models and libraries from well-known sources.\n
Fetches model weights and processors from Hugging Face using the transformers library (e.g., models from the openai and facebook organizations).\n
Installs standard Python packages including transformers, torch, and pillow from the Python Package Index (PyPI).

clip-aware-embeddings