The Agent Skills Directory

Dynamic Execution of Custom Metrics: The evaluation configuration allows for implementing custom metrics via custom_function or custom_function_file. These Python functions are executed locally within the CLI process. This provides flexibility for user-defined evaluation logic and is documented as a standard feature for extensibility.\n- Indirect Prompt Injection Surface: The skill processes conversation traces and user prompts from evaluation datasets (ingestion points). As these datasets contain external data processed by LLM judges, they represent a potential surface for indirect prompt injection. The documentation does not describe specific boundary markers or sanitization processes (boundary markers and sanitization absent). The CLI tool maintains the capability to execute code and perform network operations (capability inventory).\n- Official Tool Installation: The guide directs users to install the google-agents-cli using the uv tool. This is the intended installation path for the official evaluation utilities provided by the vendor.

google-agents-cli-eval