The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the tensorrt-llm library from official Python registries and utilizes official NVIDIA Docker images (nvidia/tensorrt_llm:latest). It also provides instructions for fetching model weights from HuggingFace repositories, which is a well-known and recognized service for model hosting.
[INDIRECT_PROMPT_INJECTION]: The skill implements model inference interfaces that ingest untrusted user text. Ingestion points: Prompts are passed through the llm.generate method in SKILL.md and the REST API endpoints in references/serving.md. Boundary markers: No delimiters or special instructions are included in the code examples to distinguish user data from model instructions. Capability inventory: The skill's functionality is limited to model serving and inference; it does not demonstrate any capabilities for file system modification or network exfiltration based on model output. Sanitization: No input validation or sanitization routines are included in the basic usage examples.
[SAFE]: Analysis of the instructions and scripts identified no malicious patterns, hidden code, or unauthorized data access attempts. The skill appears to be a legitimate developer tool for performance optimization on NVIDIA hardware.

tensorrt-llm