The Agent Skills Directory

[SAFE]: The skill uses a bash command for local environment introspection to identify available MLflow scorers. This is a legitimate functional requirement for identifying version-specific capabilities and uses the Bash tool as permitted in the skill configuration.
[SAFE]: No external network connections, sensitive file access, or obfuscated code were detected. The skill relies on local library imports (mlflow).
[SAFE]: The instructions follow best practices for building evaluation suites, emphasizing atomic criteria and cheapest reliable implementations, which serves as a defensive pattern against complex, non-deterministic prompt behaviors.
[SAFE]: The skill correctly handles tool-specific requirements, such as checking for the 'RETRIEVER' span type in MLflow retrieval scorers, which prevents potential runtime errors during evaluation.

build-a-scorer