The Agent Skills Directory

[SAFE]: The skill implements a structured framework for prompt engineering and evaluation. It follows best practices for testing AI behaviors and does not include any high-risk operations such as arbitrary command execution, network access, or external script downloads.
[SAFE]: File operations are restricted to specific local directories (promptops/, derived-index/) for reading configuration/datasets and writing run results. No sensitive system paths are accessed.
[SAFE]: While the skill processes external data (datasets and evaluation files) which is a surface for indirect prompt injection, the risk is mitigated by the skill's purpose (evaluating such content) and the lack of dangerous tools or capabilities available to the agent within this context.

apastra-eval