The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (SAFE): The skill downloads model weights and tokenizers from HuggingFace (meta-llama/LlamaGuard-7b). HuggingFace is a trusted platform, and the Meta-Llama organization is a verified trusted source.
[CREDENTIALS_UNSAFE] (SAFE): No hardcoded API keys or secrets were found. The skill correctly follows security best practices by instructing the user to use the huggingface-cli login command for authentication.
[COMMAND_EXECUTION] (SAFE): Shell commands are restricted to standard package management (pip install) and official tool authentication. No suspicious piped execution or arbitrary command strings were detected.
[INDIRECT_PROMPT_INJECTION] (SAFE): Although the skill is designed to process untrusted user content for moderation, it uses apply_chat_template for safe formatting and lacks any high-privilege capabilities (such as shell access or file writing) that could be exploited by malicious input data.
[DATA_EXFILTRATION] (SAFE): No unauthorized network operations or data exfiltration patterns were observed. Network activity is limited to downloading model assets and standard API serving via FastAPI.

llamaguard