The Agent Skills Directory

[Vendor Resource Integration]: The skill guides users on installing and using the google-agents-cli tool. This is the official command-line interface provided by Google for managing ADK projects and running evaluations.
[Dynamic Metric Extensibility]: The documentation explains how to implement custom evaluation metrics by providing a Python path in the configuration. This allows the framework to dynamically load and execute specialized evaluation logic from the user's project directory, which is a standard feature for developer tools requiring extensibility.
[Cloud Environment Configuration]: Specific evaluation criteria, such as safety metrics, require the configuration of environment variables like GOOGLE_CLOUD_PROJECT. This is standard practice for tools that integrate with Vertex AI services and Google Cloud SDKs.
[Data Ingestion Surface]: The skill processes JSON-based 'evalsets' that contain conversation trajectories and expected responses. While this is core functionality for an evaluation tool, it constitutes an ingestion surface for structured data. The provided examples follow standard schema patterns for agent testing.
[Multimodal Evaluation Patterns]: The skill includes code snippets for handling multimodal inputs (images, audio, files) using the google-genai library. These patterns demonstrate how to pass media content to vision-capable models for quality assessment.

google-agents-cli-eval