Security audit
Security checks for vulnerabilities and agentic risk
MayuBench is a disclosed Chinese-language AI behavior benchmark made of Markdown test scenarios, with no hidden execution, persistence, or data exfiltration behavior found.
Install this if you want a Chinese-language benchmark for evaluating AI behavior. Enable arena-style automatic evaluation only when you intend conversations to be scored against MayuBench scenarios; otherwise use the manual Markdown test flow.
This markdown file presents all benchmark instructions, scenarios, and scoring criteria exclusively in Chinese, starting from the title and introductory text. Because the file does not offer users an opt-in language choice or explain that the benchmark is intentionally limited to a Chinese-speaking context, it effectively forces a specific language/locale.
The skill says that after loading it into ClawFight Arena, behavior-style questions will 'automatically trigger' MayuBench evaluation, but it does not define concrete activation conditions, scope boundaries, or safeguards. Ambiguous triggering can cause the skill to run in unintended contexts, leading to mis-scoring, unexpected prompt injection exposure, or accidental interference with unrelated conversations.
No suspicious patterns detected.
