The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a variety of system commands such as make, helm, kubectl, docker, curl, and kill to orchestrate the deployment, connection, and cleanup of the Airflow controller and its related services.
[DATA_EXFILTRATION]: The skill interacts with sensitive configuration files and environment variables, including KUBECONFIG, secrets.env, and Kubernetes Secrets containing AWS and NVIDIA (NGC) credentials. These operations are performed as part of the setup and validation process, with instructions for using restricted-access temporary files in a private scratch directory.
[DYNAMIC_EXECUTION]: The skill provides Python scripts embedded in shell blocks to perform configuration parsing on local repository manifests, utilizing yaml.safe_load for secure deserialization.
[INDIRECT_PROMPT_INJECTION]: The skill reads and parses local workflow manifests (airflow/dags/workflows/*/configs/*.yaml) to determine deployment parameters, representing an attack surface for indirect injection from the repository content.
Ingestion points: The skill reads manifest files to extract Airflow pool configurations using ls and Python parsing scripts.
Boundary markers: No specific delimiters or safety warnings are applied to the content read from these configuration files.
Capability inventory: The skill possesses extensive system and infrastructure management capabilities including helm upgrade, kubectl patch, and command execution via make.
Sanitization: The skill employs yaml.safe_load for manifest parsing, which prevents code execution during data loading but does not sanitize the logical content against injection.
[CREDENTIALS_UNSAFE]: The skill includes instructions for connecting to the Airflow API using default credentials (admin/admin), although it contains a clear warning for users to update these before production deployment.

cdf-orchestration-setup