The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary workflow involves generating Python scripts using scripts/generate_eval_code.py and executing them locally to perform evaluations. This is a standard and necessary function for the skill's intended purpose of running the Vertex AI Evaluation SDK.
[REMOTE_CODE_EXECUTION]: The skill supports the CodeExecutionMetric feature of the SDK, which allows users to define custom evaluation logic in Python strings. These strings are executed in a secure, sandboxed remote environment managed by Google Cloud, as described in the references/metric_registry.md and references/sdk_patterns.md files.
[DATA_EXFILTRATION]: Evaluation data (prompts, responses, and conversation traces) is transmitted to Google Cloud Vertex AI endpoints for processing. This network activity is documented, restricted to the official vendor's infrastructure, and is a functional requirement for providing AI-assisted evaluation metrics.
[PROMPT_INJECTION]: The skill contains instructional headers (e.g., 'CRITICAL') to emphasize configuration steps like setting the GCP Project ID. These are legitimate instructional reinforcements and do not attempt to bypass agent safety filters or override system constraints.
[SAFE]: All external libraries (vertexai, google-genai, pandas) and infrastructure references are associated with the trusted vendor 'GoogleCloudPlatform' or well-known cloud services. The chmod +x operation in scripts/generate_eval_code.py is applied only to the locally generated evaluation script to enable execution, representing standard utility behavior.

quality-flywheel