The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands for environment setup and infrastructure validation, including pip install, prefect server start, git status, and kubectl diff. These are standard tools for its stated DevOps purpose.
[EXTERNAL_DOWNLOADS]: The pipeline templates reference external resources such as Python packages from PyPI and container images from the NVIDIA Container Registry (nvcr.io). These are well-known and trusted services in the ML community.
[INDIRECT_PROMPT_INJECTION]: The skill defines pipelines that ingest external data, which is a common surface for indirect prompt injection.
Ingestion points: Data is fetched via fetch_documents(source_url) and read from local files in batch_inference_flow(input_file).
Boundary markers: No specific boundary markers or "ignore embedded instruction" warnings are included in the code templates.
Capability inventory: The skill utilizes KubernetesPodOperator for pod creation, performs network requests to OpenAI and Qdrant APIs, and writes results to the local filesystem.
Sanitization: The provided templates do not include explicit sanitization for document content or input data before embedding or LLM processing, relying on standard library behavior.

ai-pipeline-orchestration