The Agent Skills Directory

[SAFE]: The skill provides detailed implementation patterns for vision-language models and retrieval-augmented generation using standard libraries such as Transformers, Torch, and FAISS. No malicious code or suspicious execution patterns were detected.
[PROMPT_INJECTION]: The skill includes a dedicated security section that warns about adversarial images and prompt injection, recommending input sanitization and output filtering as essential mitigations for production systems.
[DATA_EXFILTRATION]: Best practices for data privacy are discussed, including the recommendation to strip EXIF metadata from images to prevent the leakage of sensitive location and camera information.
[EXTERNAL_DOWNLOADS]: Dependencies and model weights referenced in code snippets are sourced from well-known technology organizations and official model registries like Hugging Face and OpenAI.

ai-multimodal