The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands for building and deploying the orchestration controller, specifically using make build sdg-controller and make install sdg-controller. These actions are gated by explicit user confirmation requirements.
[COMMAND_EXECUTION]: The skill uses kubectl to query cluster state (e.g., kubectl get pods) and curl to verify Airflow API connectivity during readiness checks. These operations are performed in a read-only manner for environment verification.
[DYNAMIC_EXECUTION]: The skill generates Python-based Airflow DAGs and workflow-local callables based on user requirements. It performs verification by attempting to import the generated modules using python -c to ensure they are valid before deployment.
[INDIRECT_PROMPT_INJECTION]: As the skill translates natural language descriptions of data pipelines into executable Python code, it possesses an inherent attack surface for indirect prompt injection. This risk is mitigated by the requirement for the agent to adapt existing, vetted task group patterns rather than inventing logic, and the mandatory human-in-the-loop confirmation before deployment.
[EXTERNAL_DOWNLOADS]: The skill references and fetches documentation and configuration excerpts from official NVIDIA GitHub repositories (e.g., github.com/NVIDIA/cdf-augmentation). As these are vendor-owned resources, they are considered safe within the context of this skill.

cdf-orchestration-write-dag