The Agent Skills Directory

[No Code] (INFO): The file is a documentation resource (Markdown) and contains no executable scripts, commands, or automation logic.- [Prompt Injection] (SAFE): Includes a system prompt for an evaluation judge; however, the instructions are benign and do not attempt to subvert AI safety controls.- [Indirect Prompt Injection] (LOW): The described evaluation process handles untrusted data (model outputs). While this represents an attack surface for the judge LLM, the output is limited to scoring and reasoning, posing negligible risk in the context of this reference document.

context-compression