The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches machine learning models from the official Hugging Face Hub. This is a trusted source for model weights and configuration.
[EXTERNAL_DOWNLOADS]: References dependencies from the official NPM registry and well-known JSDelivr CDN. These are standard distribution channels for JavaScript libraries.
[DATA_EXPOSURE]: Utilizes a local cache directory (defaults to ./.cache or ~/.cache/huggingface/) to store downloaded models. This is standard behavior for the Transformers.js library to optimize performance and minimize redundant downloads.
[COMMAND_EXECUTION]: Provides instructions for running inference tasks using the pipeline API. No arbitrary command execution patterns or privilege escalation attempts were detected.
[PROMPT_INJECTION]: As a tool designed to process arbitrary data through machine learning models, the skill presents a surface for indirect prompt injection.
Ingestion points: The pipeline function in SKILL.md accepts untrusted text, image, and audio data for analysis or generation.
Boundary markers: Examples do not demonstrate the use of delimiters or "ignore" instructions for processed content.
Capability inventory: Inference is performed via WebAssembly or WebGPU; the skill also interacts with the local filesystem for caching (references/CACHE.md).
Sanitization: No sanitization or input validation logic is present in the provided examples.

transformers-js