The Agent Skills Directory

[SAFE]: The skill provides standard implementations for LLM metrics using reputable libraries like nltk, scipy, and pydantic. No malicious command execution or unauthorized data access patterns were detected.
[SAFE]: The included CI/CD integration example correctly demonstrates the use of GitHub Secrets for managing sensitive credentials (OPENAI_API_KEY), adhering to secure configuration practices.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data (LLM outputs and retrieved context) which creates a natural attack surface for indirect prompt injection.
Ingestion points: Untrusted model responses and external contexts are processed in the RAGAS evaluation section and the AgentTaskEvaluator class in SKILL.md.
Boundary markers: The prompt templates for the LLM-as-judge functionality do not use explicit delimiters to isolate untrusted content.
Capability inventory: The skill uses the OpenAI API for evaluative judgments and standard NLP libraries for text processing.
Sanitization: No explicit sanitization or validation of the content being evaluated is demonstrated in the provided code snippets.

llm-evaluation