gcp-agent-eval-metric-configurator

This skill helps you configure sophisticated automated evaluation metrics. Grounded in evaluation_blog.md, it supports computation-based, rubric-based, and managed Vertex AI metrics.

Usage

Ask Antigravity to:

• "Configure Grounding metrics for my researcher agent"
• "Add a Tool Use Quality evaluator to my pipeline"
• "Set up a ResponseMatch check against my reference answers"
• "Configure an adaptive rubric for style alignment"

Metric Taxonomy

1. Computation-Based: JSON validity, Execution trajectory matching.
2. Managed Rubric-Based (Vertex AI):
   • GROUNDING: Ensures responses are fully supported by context (RAG).
   • TOOL_USE_QUALITY: Checks if the right tool was called with correct parameters (no reference needed).
3. Adaptive Rubrics: Use LLM-as-a-judge to grade responses based on unique criteria generated for each prompt.

Metric Templates

Refer to resources/metric_templates.json for standard definitions.