The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill follows secure practices by instructing users to avoid hardcoding secrets, recommending silent input via read -rsp for API keys, and suggesting restrictive file permissions (chmod 600) for environment files. All examples use non-sensitive placeholders.
[EXTERNAL_DOWNLOADS]: The instructions involve cloning official repositories from NVIDIA's GitHub organization and accessing models from Hugging Face, which are recognized as trusted and well-known sources respectively.
[COMMAND_EXECUTION]: Shell commands provided for docker compose and curl are appropriately scoped to the deployment and health verification of the microservices described in the documentation.
[DATA_EXFILTRATION]: No unauthorized data transfer or exfiltration patterns were identified. Network operations are confined to standard interactions with user-defined inference endpoints and trusted software registries.

rtvi-vlm-customize-model