The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the deepeval package from PyPI using pip install -U deepeval. This is a standard installation procedure for a well-known open-source tool.
[COMMAND_EXECUTION]: The skill utilizes the deepeval CLI (e.g., deepeval test run) to execute evaluation tests. This is the intended functionality of the framework and is performed locally on the user's machine.
[DATA_EXFILTRATION]: The skill mentions deepeval login to integrate with the Confident-AI dashboard. This is an optional, documented feature of the tool for tracking evaluation results and does not involve unauthorized data access.
[INDIRECT_PROMPT_INJECTION]: The skill defines evaluation test cases that ingest external data (inputs, actual outputs, and retrieval contexts). While this presents a surface for indirect prompt injection, it is inherent to the purpose of an LLM evaluation framework and the skill does not grant excessive privileges or perform dangerous operations with this data.

deepeval-evaluation