The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a detailed set of commands for the nrl-k8s CLI to launch, monitor, and debug NeMo-RL recipes on Kubernetes. These include operations for RayCluster management, job submission, and log retrieval using standard tools like kubectl and curl to local services.
[EXTERNAL_DOWNLOADS]: The instructions reference container images from NVIDIA's official registry (nvcr.io) and logging services from Weights & Biases (wandb.ai). These are trusted vendor resources and well-known services in the machine learning ecosystem. Additionally, the CLI performs routine package updates for ray[default] from official registries to resolve dashboard issues.
[CREDENTIALS_UNSAFE]: The playbook includes instructions for managing sensitive data such as SSH keys and API tokens. It correctly implements best practices by advising the use of Kubernetes secrets (secretKeyRef) and dedicated setup commands (nrl-k8s dev setup-secrets) rather than hardcoding credentials in the skill or configuration files.
[PROMPT_INJECTION]: No malicious patterns or attempts to override agent behavior, bypass safety filters, or extract system prompts were detected in the skill instructions or associated metadata.
[DATA_EXFILTRATION]: There is no evidence of unauthorized data transfer to untrusted external domains. Network operations are appropriately scoped to local cluster services (localhost) and established observability platforms like Weights & Biases.

launch-nemo-rl