The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands for validating the environment, configuring SSH, and managing jobs on a SLURM cluster. This includes the use of ssh for remote access, sbatch and srun for job orchestration, and squeue, sacct, and scancel for monitoring and control.\n- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the nvidia-tao-sdk package. The documented installation sources include public PyPI and an internal vendor GitLab instance (gitlab-master.nvidia.com), both of which are official resources for the provided functionality.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external inputs such as Lustre filesystem paths and job configurations that influence the generated remote execution scripts.\n
Ingestion points: User-provided storage paths (e.g., lustre:///...) and job specification metadata.\n
Boundary markers: The documentation specifies that the underlying platform (tao-core) validates storage URIs and rejects local or file-based paths for remote backends, which helps prevent unauthorized local access.\n
Capability inventory: The skill is capable of executing remote shell commands over SSH, writing configuration files to shared storage, and launching containerized workloads via SLURM.\n
Sanitization: The skill explicitly guides the agent to use printf piped to ssh for secret installation (e.g., NGC_KEY), ensuring that sensitive values are not recorded in shell history or intermediate logs.

tao-run-on-slurm