This dataset is proposed in our work, "When Agents Learn to Be You: Benchmarking Privacy Leakage, Impersonation Risk, and Defenses in Persona Skills".
It includes simulated persona grounded user–assistant dialogues as personal traces for skill distillation, consists of:
sync_characters.jsonl contains 50 character profiles.
sync_questions.jsonl contains 2,500 character-grounded questions. Each character has 50 questions: 30 general, 10 math… See the full description on the dataset page:
https://huggingface.co/datasets/yonglixiang/AntiSkillBench.
