The Agent Skills Directory

[SAFE]: The skill's primary focus is educational and implementation-oriented for LLM evaluation. No malicious patterns such as credential theft, unauthorized network access, or persistence mechanisms were detected.- [PROMPT_INJECTION]: The skill implements prompts for evaluation that interpolate untrusted data (prompts and responses) into the judge context. While this creates a surface for indirect prompt injection, the skill includes robust mitigation strategies such as Chain-of-Thought reasoning, structured rubrics, and position-bias mitigation. In the context of an evaluation skill, this implementation is standard and well-guarded.- [EXTERNAL_DOWNLOADS]: References to external sources are limited to well-known academic (arXiv.org) and industry research sites, which are used purely for informational purposes.

advanced-evaluation