The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automates the installation of the deepeval and pytest Python packages using pip install in the scripts/init-deepeval.sh file. These are well-known, legitimate packages from the official Python Package Index (PyPI) and are required for the skill's stated purpose of LLM testing.
[INDIRECT_PROMPT_INJECTION]: The skill defines evaluation patterns that process LLM inputs and outputs, which represents an indirect prompt injection surface as these inputs often originate from untrusted sources.
Ingestion points: Data enters the agent context through the LLMTestCase arguments (input, actual_output) in templates/test_basic.py and similar evaluation scripts.
Boundary markers: None identified in the provided templates; evaluation scripts typically pass raw strings to metrics.
Capability inventory: The skill has the capability to execute shell commands (Bash), install packages (pip), and perform file operations (mkdir, cp) during the initialization phase in scripts/init-deepeval.sh.
Sanitization: No explicit sanitization of the evaluated text is present in the templates, as the goal is to evaluate the content's quality and safety using DeepEval's built-in metrics (e.g., ToxicityMetric, HallucinationMetric).
[CREDENTIALS_UNSAFE]: The templates/conftest.py file includes a fixture to retrieve the OPENAI_API_KEY from environment variables. This follows standard security practices for managing API keys in development and testing environments and does not involve hardcoding secrets.

deepeval-testing