The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external source code and job metadata which could theoretically contain instructions intended to influence the agent's behavior.
Ingestion points: The skill downloads and analyzes Spark source code from Google Cloud Storage (GCS) or local paths specified by the user in SKILL.md.
Boundary markers: The instructions provide explicit constraints to maintain functional logic and production paths, acting as a behavioral boundary.
Capability inventory: The skill possesses the ability to create Dataproc clusters, submit jobs via gcloud dataproc, and perform file operations using gcloud storage and uv run.
Sanitization: The skill performs syntax verification using python3 -m py_compile but does not implement explicit content filtering or sanitization on the ingested Spark scripts.
[COMMAND_EXECUTION]: The skill frequently uses shell commands to interact with the environment and cloud resources.
Evidence: Extensive use of gcloud CLI tools for project description, region validation, cluster management, and job submission across SKILL.md and references/cluster_input_validations.md.
Evidence: Uses uv run to execute the internal helper script scripts/cluster_resolve_version.py in an isolated environment.
[EXTERNAL_DOWNLOADS]: The skill fetches necessary metadata from external sources to perform version resolution.
Evidence: The script scripts/cluster_resolve_version.py uses urllib.request.urlopen to fetch Dataproc release notes and Spark compatibility tables from official Google Cloud documentation at cloud.google.com.

gcp-managed-spark-upgrades