The Agent Skills Directory

[SAFE]: The skill consists entirely of instructional documentation and evaluation test cases. No executable scripts, network requests, or unauthorized file access patterns were identified.
[DATA_EXFILTRATION]: The skill incorporates security best practices by explicitly instructing the user to redact secrets and Personally Identifiable Information (PII) before sharing data with external evaluation services.
[PROMPT_INJECTION]: The instructions include defensive measures against prompt injection, specifically advising that candidate outputs be treated as untrusted input and recommending the use of hard delimiters and injection canaries when using LLM-as-judge scoring.

evaluate-prompts