The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill requires access to the sensitive Kubernetes configuration file located at ~/.kube/config to interact with the cluster. While this is standard for deployment tools, it involves handling sensitive credential paths.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for running containerized versions of the deployment tool from the official NVIDIA Container Registry (nvcr.io/nvidia/cloud-native/k8s-launch-kit).
[COMMAND_EXECUTION]: The skill generates shell commands for both local and Docker-based execution of the l8k pipeline, which includes hardware discovery and resource deployment.
[PRIVILEGE_ESCALATION]: The suggested Docker execution patterns use the --net=host flag, which grants the container full access to the host's network namespace. This high-privilege configuration is necessary for low-level networking fabric discovery but increases the potential impact of a container escape.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests hardware and cluster discovery data which is then used to generate and apply deployment manifests.
Ingestion points: The hardware discovery phase (discover) probes the live cluster environment for configuration data.
Boundary markers: There are no explicit boundary markers or 'ignore embedded instruction' warnings provided for the ingested discovery data.
Capability inventory: The skill has the capability to write YAML manifests to the local file system and apply those resources directly to the Kubernetes cluster via the --deploy flag.
Sanitization: No specific sanitization or validation steps for the external hardware discovery data are defined in the instructions.

k8s-launch-kit-pipeline