The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the tensorrt_llm Python package from the official PyPI registry and recommends the use of official NVIDIA Docker images (nvidia/tensorrt_llm:latest). All external model resources reference the official Hugging Face repository, ensuring that dependencies are sourced from recognized and trusted providers.
[INDIRECT_PROMPT_INJECTION]: The skill implements model inference capabilities that ingest user-provided prompts. Evaluation shows that the skill does not expose sensitive capabilities (such as arbitrary file writes or persistent network connections) that could be co-opted by malicious input. Standard boundary markers are utilized in the serving implementations to isolate model instructions from data.
[CREDENTIALS_UNSAFE]: Security for production deployments is handled through shell environment variables rather than hardcoding. The documentation correctly demonstrates using openssl to generate and export API keys for the trtllm-serve authentication layer.

tensorrt-llm