The Agent Skills Directory

[SAFE]: The skill manages sensitive authentication data, such as BREV_API_TOKEN, NGC_KEY, and AWS credentials, through environment variables and explicit preflight checks rather than hardcoding them.
[SAFE]: Documentation provides specific security guidance for handling secrets, instructing users to pass the NGC_KEY via stdin (e.g., using printf and docker login --password-stdin) to ensure sensitive information does not appear in process tables or command history.
[SAFE]: Remote CLI requirements and SDK dependencies are sourced from official NVIDIA documentation and the public PyPI registry, with specific version pinning (nvidia-tao-sdk[brev]==7.1.0rc42) to ensure supply chain integrity.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for indirect injection as it interpolates user-provided parameters (such as instance IDs, cloud credential IDs, and Docker image tags) into shell commands executed on remote instances.
Ingestion points: User-supplied parameters in backend_details and CLI arguments mentioned in SKILL.md.
Boundary markers: Absent; user inputs are directly concatenated into command templates.
Capability inventory: Shell command execution via brev exec and remote Docker container execution.
Sanitization: No explicit validation or escaping logic is defined within the skill's instructional templates.

tao-run-on-brev