The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs network requests to well-known AI services (Hugging Face and Mistral AI) via official SDK calls to perform inference tasks. It also demonstrates reading from and writing to the local filesystem using the node:fs module for processing media files (audio, images), which is necessary for the multi-modal functionality described.
[EXTERNAL_DOWNLOADS]: The skill uses the standard @huggingface/inference package from the public NPM registry, which is the official and expected dependency for this infrastructure.
[PROMPT_INJECTION]: The skill ingests untrusted user data for processing by external models, identifying an indirect prompt injection surface. Ingestion points: User messages are passed to the chatCompletion and textGeneration methods via the messages and inputs parameters. Boundary markers: The skill recommends using system role messages to define and constrain model behavior. Capability inventory: The skill possesses network access for API calls and filesystem access for media processing. Sanitization: No explicit input validation or escaping is provided in these patterns, as is common for low-level SDK examples.

ai-infrastructure-huggingface-inference