The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The references/requirements.txt file specifies several Python package versions that are not currently available on the public PyPI registry, including google-cloud-aiplatform==1.138.0, google-genai==1.73.1, numpy==2.4.2, and pandas==3.0.1. Attempting to install these may result in errors or indicate a configuration that relies on non-standard package sources.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it is designed to ingest and process external datasets from sources like HuggingFace, Google Cloud Storage, or local files.
Ingestion points: Dataset discovery in Phase 1 and data processing scripts (scripts/prepare_dataset.py, scripts/calculate_cost.py).
Boundary markers: The instructions do not define specific delimiters or instructions to ignore potential commands embedded within the training data.
Capability inventory: The skill has the capability to execute shell commands via gcloud and interface with GCP services via Python scripts.
Sanitization: There is no evidence of sanitization or filtering of the dataset content to prevent malicious instructions from influencing the agent's behavior during processing.
[COMMAND_EXECUTION]: The skill frequently guides the agent to execute system commands, including environment setup (venv, pip), CLI updates (gcloud components update), and job management. To mitigate unauthorized execution, the skill includes explicit "Interactive Confirmation Required" instructions and "CRITICAL" checkpoints that mandate user approval before proceeding with job submission or model deployment.

agent-platform-tuning